# Open Factory > Research on the equipment that runs factories and datacenters, written for the people who buy it. Open Factory publishes research, lead-time and price data, models and calculators on machinery, electrical gear, power and cooling, with a procurement focus. Open Factory is a research publication on factory and datacenter equipment procurement: lead times, prices, data models and calculators for transformers, switchgear, gensets, chillers, machine tools and the people who buy them. Machine access: JSON API at https://open-factory.com/api/v1/index.json (OpenAPI at https://open-factory.com/api/v1/openapi.json) and an MCP server at https://open-factory.com/mcp (Streamable HTTP, no key needed for the free tools). Docs: https://open-factory.com/developers/ # Dry Cooler or Cooling Tower: What the Water Bill Says in 15 Markets URL: https://open-factory.com/articles/dry-cooler-vs-tower-water-bill/ Date: 2026-10-16 Cooling / TCO Douglas County $24.80 per kgal, 6 million gallons per MW-year, Google's sites, Houston 3,339 wet-bulb hours, dry-cooler fan penalty, four markets where hybrid pays, the 50 MW BoM A thousand gallons delivered to a large meter in Douglas County, Georgia, costs [$24.80 once it is sent back to the sewer](https://ddcwsa.com/wp-content/uploads/2026/01/2025-2026-Rates-and-Impact-Fee-Chart-External-1.pdf). The authority's current chart splits that into $12.10 for the water in the top tier and $12.70 for the treatment. The same thousand gallons in Baton Rouge costs [$1.51 from the water company](https://lpsc.louisiana.gov/docs/utilities/Comparisons/Water/LPSC-Jurisdictional-WATER-Rate-Fee-Comparison-07-21-26.pdf) and [$7.17 from the parish sewer](https://www.brla.gov/DocumentCenter/View/23356/Sewer-User-Fee-Ordinance-19755-Adopted-February-25-2026), $8.68 together. An open cooling tower rejecting 50 MW of heat needs about 300 million gallons a year, so with a sewer deduct meter in place the gap between those two sheets is $3.6 million a year for the same plant. The [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) now holds the water and sewer tariff for every one of the 15 markets in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/), and this piece is the first side-by-side. The bars run from $24.80 in Douglas County and $23.96 in Santa Clara through $21.14 in Houston to $9.43 in Denver and $8.68 in Baton Rouge, and the sewer segment is the larger of the two in eleven of the fifteen rows. The piece covers the gallons a tower needs, the fifteen tariff sheets in one table, what Google, Microsoft and Meta disclose, the wet-bulb and dry-bulb hours that decide which machine works where, the fan penalty of going dry, the ten-year breakeven by market, why the water fights are about permits rather than bills, and, behind the wall, the 50 MW bill of materials three ways with the vendor shortlist. ## The Gallons a Tower Needs An evaporative tower rejects heat by boiling off part of the water it circulates. The EPA's WaterSense guide puts the figure at [approximately 1.8 gallons evaporated for every ton-hour of cooling, regardless of tower efficiency](https://19january2021snapshot.epa.gov/sites/static/files/2017-12/documents/ws-commercialbuildings-waterscore-mechanical-systems-guide.pdf), and the WaterSense at Work chapter on towers states the same physics as [1 percent of recirculating flow for every 10F of range](https://www.epa.gov/system/files/documents/2023-05/ws-commercial-watersense-at-work_Section_6.3_Cooling_Towers.pdf). One MW of heat is 284 tons, so a tower running flat out for a year evaporates 4.48 million gallons per MW. The dissolved solids stay behind, so the tower bleeds concentrated water to the drain to keep scale off the fill. The FEMP best-practice sheet gives the relationship: [blowdown equals evaporation divided by cycles of concentration minus one](https://www1.eere.energy.gov/femp/pdfs/waterfs_coolingtowers.pdf), and the EPA notes that [many systems operate at two to four cycles](https://www.energy.gov/femp/best-management-practice-10-cooling-tower-management). **At four cycles, a tower draws 5.98 million gallons of makeup per MW-year and sends 1.49 million of them to the sewer.** Loudoun Water, the utility under Ashburn, builds the same assumption into its tariff: its estimated-flow table for a water-cooled data center [assumes a 25% blow down to sanitary sewer](https://www.loudounwater.org/sites/default/files/2025-08/Board%20-%202024%20User%20Rates%20Reclaimed%20Water%20Rates%20Avilability%20Charges%20and%20Other%20Fee%20and%20Charges%20-%20Presentation%20-%20Carnes%20-%20Oct%2010%202024.pdf), which is four cycles exactly. SemiAnalysis's cooling anatomy arrives at the same order of magnitude from the other direction: [a 50 MW datacenter at 60% utilization, PUE 1.25 and a WUE of 2.0 uses 174 million gallons a year](https://newsletter.semianalysis.com/p/datacenter-anatomy-part-2-cooling-systems), and at full load with the heat of the chillers added the figure climbs toward our 300 million. The sewer half of the bill depends on one piece of hardware. Most utilities bill sewer on metered water in, unless the customer installs a deduct meter on the makeup line that proves the evaporated gallons never reached the drain. Salt Lake City bills sewer on a flat [70% of total water consumption](https://www.slc.gov/utilities/fy27rates/), Douglas County on [80%](https://ddcwsa.com/wp-content/uploads/2026/01/2025-2026-Rates-and-Impact-Fee-Chart-External-1.pdf), Phoenix as [88% of January-to-March water](https://www.phoenix.gov/content/dam/phoenix/waterservicessite/documents/sewer_fees_schedule_by_user_2026_0701.pdf), and Chicago at [100 percent of the water charge](https://www.chicago.gov/city/en/depts/fin/supp_info/utility-billing/water-and-sewer-rates.html) with no evaporation credit at all. A tower billed sewer on all of its intake pays treatment on water that left as vapor. With a deduct meter and four cycles, the bill per MW-year is $102,295 in Santa Clara, $91,311 in Douglas County, $73,467 in Houston, $56,296 in Phoenix, $36,270 in Northern Virginia and $19,740 in Baton Rouge. Without the deduct meter the fifteen numbers run from $51,882 to $148,250, with every row in [tower-water-cost-per-mw-year.csv](data/tower-water-cost-per-mw-year.csv). For 50 MW the deduct-meter case is $5.1 million a year in Santa Clara and $1.0 million in Baton Rouge. The [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool prices the electric side of the same fifteen connections, and the water line now sits beside it. ## Fifteen Tariff Sheets, One Table We opened the current rate sheet or ordinance for one water utility in each market, took the volumetric rate a user drawing millions of gallons a month pays on its last gallon, and converted hundred-cubic-foot rates at 748 gallons. The utilities match the markets in [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/): the data center cluster inside each electric territory. | Market | Utility | Water $/kgal | Sewer $/kgal | Combined | Effective | |---|---|---|---|---|---| | Atlanta (Douglas County) | Douglas County WSA | 12.10 | 12.70 | 24.80 | 2025-12-01 | | Bay Area (Santa Clara) | City of Santa Clara | 14.83 | 9.13 | 23.96 | 2026-07-01 | | Houston | Houston Public Works | 9.34 | 11.80 | 21.14 | 2026-04-01 | | Salt Lake City | Salt Lake City DPU | 4.70 | 14.67 | 19.37 | 2026-07-01 | | Carolinas (Charlotte) | Charlotte Water | 4.88 | 9.20 | 14.08 | 2026-07-01 | | Columbus | Columbus DPU | 5.11 | 8.84 | 13.95 | 2026-01-01 | | Northern Indiana (Fort Wayne) | Fort Wayne City Utilities | 3.65 | 9.66 | 13.31 | 2026-06-22 | | Phoenix (APS) | Phoenix Water Services | 8.28 | 4.55 | 12.83 | 2025-03-01 | | Phoenix (SRP) | City of Mesa | 5.39 | 6.47 | 11.86 | 2026-04-01 | | Northern Virginia | Loudoun Water | 4.42 | 6.59 | 11.01 | 2026-01-01 | | West and South Texas (Abilene) | City of Abilene | 6.73 | 4.19 | 10.92 | 2026-10-01 | | Dallas-Fort Worth | Dallas Water Utilities | 5.11 | 5.15 | 10.26 | 2025-10-01 | | Chicago | Chicago DWM | 4.98 | 4.98 | 9.96 | 2026-06-01 | | Colorado (Denver) | Denver Water, Denver DOTI | 3.55 | 5.88 | 9.43 | 2026-01-01 | | Louisiana (Baton Rouge) | Baton Rouge Water, EBR sewer | 1.51 | 7.17 | 8.68 | 2026-01-01 | *Source: [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). CSV with the tier, basis and tariff URL for each row: [water-sewer-rates-by-market.csv](data/water-sewer-rates-by-market.csv)* The top of the table is a tier structure, not a scarcity premium. Douglas County has no commercial class at all and scales its three tiers with meter size, so an 8-inch meter pays [$12.10 on every gallon over 640,000 a month](https://ddcwsa.com/wp-content/uploads/2026/01/2025-2026-Rates-and-Impact-Fee-Chart-External-1.pdf), which for a tower is every gallon. Santa Clara charges [one potable rate of $11.09 per hundred cubic feet to every class](https://www.santaclaraca.gov/our-city/departments-g-z/water-sewer-utilities/water-sewer-and-recycled-water-rates/water-rates) and bills sewer at [$6.83 per hundred cubic feet on 90% of metered water](https://www.santaclaraca.gov/our-city/departments-g-z/water-sewer-utilities/water-sewer-and-recycled-water-rates/sewer-rates) for industrial users not otherwise classified. Houston's April 2026 schedule lists [$9.34 for water and $11.80 for wastewater per thousand gallons](https://www.houstonsecured.org/docs/2026%20APRIL%20WATER%20RATES%20.pdf) on all commercial and industrial volume. The bottom of the table is cheap water with expensive treatment. Baton Rouge Water Company's commercial tail block is [$1.13 per hundred cubic feet](https://lpsc.louisiana.gov/docs/utilities/Comparisons/Water/LPSC-Jurisdictional-WATER-Rate-Fee-Comparison-07-21-26.pdf) in the Louisiana PSC's July 2026 comparison, while the parish sewer ordinance adopted in February 2026 charges [$7.169 per thousand gallons and escalates 4% every January from 2027](https://www.brla.gov/DocumentCenter/View/23356/Sewer-User-Fee-Ordinance-19755-Adopted-February-25-2026). Fort Wayne's declining block, approved by the Indiana commission in June 2026, prices [all usage over 50,000 cubic feet at $2.73](https://www.in.gov/iurc/files/ord_46311_061726.pdf). Denver's Tier 1 is [$3.55 per thousand gallons](https://www.denverwater.org/business/billing-and-rates/2026-rates) up to the customer's average winter use, which a flat load never exceeds, so the [$2.20 drought adder](https://www.denverwater.org/drought/board-resolution-drought-pricing) on Tier 3 from June 2026 does not reach it. **Two of the fifteen utilities already sell a second, cheaper pipe to the same meter.** Loudoun Water prices reclaimed water at [$2.07 per thousand gallons against $4.42 potable](https://www.loudounwater.org/node/1087), delivered [more than 750 million gallons of it to roughly 40 customers in 2025](https://www.loudounwater.org/commercial-customers/reclaimed-water-program), and the county says data centers draw [9% of total potable water on an average day](https://www.loudoun.gov/6402/Water-Considerations). Denver sells recycled water inside the city at [$1.20 per thousand gallons](https://www.denverwater.org/business/billing-and-rates/2026-rates), a third of the potable Tier 1, and Santa Clara at [$6.05 per hundred cubic feet](https://www.santaclaraca.gov/our-city/departments-g-z/water-sewer-utilities/water-sewer-and-recycled-water-rates), a 45% discount. Google's Douglas County site shows what the purple pipe does to a bill: [440.6 million gallons withdrawn in 2025, 394.3 of them reclaimed](https://sustainability.google/files/google-2026-environmental-report.pdf). ## What the Disclosures Say the Fleet Draws Three operators publish enough to check the EPA arithmetic against metered sites. Google's 2026 environmental report lists [1,746.4 million gallons withdrawn at Council Bluffs, 1,396.7 at Mayes County, 1,008.1 at New Albany, 263.6 at Midlothian, 258.4 at Sterling and 19.4 at Mesa](https://sustainability.google/files/google-2026-environmental-report.pdf), Mesa being one of eight sites footnoted as air-cooled. Fleet consumption was [10.9 billion gallons in 2025, up 34%](https://sustainability.google/files/google-2026-environmental-report.pdf). Google publishes no liters-per-kWh figure. Its 2022 cooling post said [the average Google data center consumed approximately 450,000 gallons of water per day in 2021](https://blog.google/outreach-initiatives/sustainability/our-commitment-to-climate-conscious-data-center-cooling/), which at our full-load rate is a 27 MW tower plant. New Albany's 1,008.1 million gallons at the Columbus rate of $13.95 per thousand is a $14 million bill before any deduct credit, the three Loudoun towns drew 573 million gallons together, and Mesa's 19.4 million is 1.1% of Council Bluffs. Microsoft and Meta publish the ratio Google withholds. Microsoft's efficiency page lists a [global WUE of 0.27 L/kWh for FY25 and 0.34 for the Americas](https://datacenters.microsoft.com/sustainability/efficiency/), and its 2026 data fact sheet lists electricity and water by location, which divides to [1.03 L/kWh at Phoenix, 0.55 at Chicago, 0.29 at San Antonio and 0.12 at Boydton](https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/2026-Microsoft-Environmental-Data-Fact-Sheet-PDF.pdf). Meta's data index puts its 2024 fleet at [0.19 L/kWh](https://sustainability.atmeta.com/wp-content/uploads/2025/10/Meta_2025-Environmental-Data-Index.pdf), and its Fort Worth site at 311 megaliters against 1,109,004 MWh, or 0.28. The ladder runs 2.58 L/kWh for the EPA tower rule at four cycles, 2.30 for Microsoft's own first-generation sites, 1.03 at Microsoft Phoenix, 0.36 for LBNL's 2023 US fleet average, 0.28 at Meta Fort Worth, 0.19 for Meta's fleet and zero for the design Microsoft is piloting. Microsoft's June 2026 post states that its [average WUE fell from 2.3 L/kWh to 0.27 in 2025](https://blogs.microsoft.com/blog/2026/06/24/inside-microsofts-two-decade-push-to-cut-water-intensity-while-scaling-for-growth/), that about 90% of the owned fleet runs low-to-zero-water cooling, and that those sites rely on water only when outside temperatures exceed 85F. **The 2.3-to-0.27 drop is the whole argument of this piece: it came from changing the heat-rejection machine, and the water bill it removed is the one charted above.** LBNL's December 2024 report for DOE puts the US fleet at [66 billion liters of direct consumption in 2023 and 60 to 124 billion liters for hyperscale alone by 2028](https://escholarship.org/content/qt32d6m0d1/qt32d6m0d1.pdf), with the highest WUE at water-cooled chiller plants without economizers. The [Cooling Roadmap](https://open-factory.com/roadmaps/cooling-roadmap/) tracks the inlet temperature each rack generation permits, which is what lets a site move down the ladder. ## Wet-Bulb Hours Decide the Tower A tower can only cool water to within a few degrees of the ambient wet bulb. We took the NREL TMY3 weather file for the airport nearest each market, computed the hourly wet bulb from dry bulb, dew point and station pressure, and counted the hours above 70F, where a tower selected at the CTI 78F standard starts losing approach and the chiller behind it works harder. We ran the same count on the TMYx files built from 2009 to 2023 observations. Houston counts 3,339 hours above 70F in TMY3 and 3,732 in the 2009 to 2023 file, Baton Rouge 2,959 and 3,584, Dallas 2,589 and 2,566, Atlanta 1,566 and 1,866, Phoenix 833 and 1,137, Northern Virginia 967 and 927, Columbus 538 and 818, and Denver, Salt Lake City and San Jose 0 to 15 hours in either file, every station and threshold in [wet-bulb-hours-by-metro.csv](data/wet-bulb-hours-by-metro.csv). The 0.4% design wet bulb, the figure a tower is selected against, runs from 64.5F in Salt Lake City to 80.8F in New Orleans in TMY3 and is 2 to 5F higher in the 2009 to 2023 files for Houston, Atlanta and Baton Rouge. The Gulf Coast markets with the most wet-bulb hours are where a tower works hardest and a dry cooler works worst, because dry bulb and wet bulb sit close together and neither machine gets much help from the air. **A tower in Salt Lake City evaporates the same 1.8 gallons per ton-hour as one in Houston. What changes with the wet bulb is the approach and the chiller behind it, not the water.** ## Dry-Bulb Hours Decide the Dry Cooler A dry cooler is a finned coil with fans, and it can only cool fluid to within an approach of the dry-bulb temperature. Evapco rates its eco-Air series at [110F entering and 100F leaving at 92F dry bulb](https://www.evapco.com/media/1645013), an 8F approach, and a Guntner selection in a Massachusetts bid package is [115F to 100F at 95F dry bulb](https://deerfieldma.us/DocumentCenter/View/875/008_Spec-and-Cut-Sheets). A loop at the ASHRAE W32 class, [supply at or below 32C](https://www.cibsejournal.com/cpd/modules/2025-09-lcdca/), or 89.6F, cannot be held by an 8F-approach dry cooler once the air passes about 82F, and that is where a hybrid turns on its spray. Microsoft's hybrid fleet uses [water only when outside temperatures exceed 85F](https://blogs.microsoft.com/blog/2026/06/24/inside-microsofts-two-decade-push-to-cut-water-intensity-while-scaling-for-growth/), the same threshold with a slightly warmer loop. Phoenix counts 2,800 hours above 85F dry bulb in TMY3 and 3,127 in the 2009 to 2023 file, Dallas 1,143 and 1,443, Houston 958 and 1,343, Abilene 906 and 1,281, Atlanta 564 and 613, Northern Virginia 296 and 381, Columbus 69 and 221, and San Jose 48 and 76. Phoenix is the one market with few wet-bulb hours and many dry-bulb hours, the hybrid's home climate: wet assist on 36% of hours, dry on the rest. Microsoft's Phoenix WUE of 1.03 L/kWh is that climate metered, 40% of the open-tower figure, after a [23% year-over-year improvement in FY25](https://blogs.microsoft.com/blog/2026/06/24/inside-microsofts-two-decade-push-to-cut-water-intensity-while-scaling-for-growth/). The price of dry rejection is fan power, and four manufacturers publish enough to measure it. The Marley NC Everest [NC8422R-1 lists 1,311 tons on a 25 hp motor](https://spxcooling.com/wp-content/uploads/TECH-NC-20.pdf), 4.0 kW per MW, the Evapco AT [312-2I36 851 tons on three 10 HP fans](https://www.evapco.com/sites/evapco.com/files/2025-08/AT25ENG_WEB.pdf), 7.5 kW, and the BAC Series 3000 [S3E-1424-14W 1,394 tons on 125 HP](https://baltimoreaircoil.com/download_api_endpoint/536/b8_ct_s3000_20150914.pdf), 19.0 kW, all at the CTI 95F, 85F and 78F wet-bulb basis. The Evapco eco-Air [EAW-DD 33S3 rejects 7,130 MBH on 120 HP](https://www.evapco.com/media/1645013), 43.1 kW per MW, and the twelve-fan EAW-DD 1512 rejects 8,330 MBH on 278 HP, 84.8 kW per MW, at the same 92F dry bulb. Those are nameplate figures at the rating point. Over a year the fans slow on every cooler hour, and the pumps and chiller behind a tower, excluded from the chart, draw more than its fans do. Amazon says [evaporative systems cut operational energy use 25% to 35% during peak summer periods](https://sustainability.aboutamazon.com/stories/just-back-from-weftec-why-does-ai-use-water), and a 2022 study of two Phoenix colocation sites found the air-cooled one ran [about 13% higher PUE for 66% lower source WUE](https://experts.arizona.edu/en/publications/water-energy-tradeoffs-in-data-centers-a-case-study-in-hot-arid-c/). Uptime's April 2026 briefing argues that [well-designed dry and adiabatic systems can match evaporative PUE across ASHRAE climate zones 2 to 6](https://intelligence.uptimeinstitute.com/resource/dry-cooling-energy-performance-can-rival-evaporative-cooling), which holds where the loop is warm enough to need no chiller. ## The Fan Penalty and the Breakeven No manufacturer publishes a dry cooler price, and none of the 2025 and 2026 bid tabulations we opened priced one against its capacity. Towers are documented. South Carolina's Department of Veterans' Affairs put a [370-ton tower replacement with condenser pumps and controls at $625,000 to $700,000](https://scbo.sc.gov/files/scbo/SE-310_E26-9862-JM.pdf) in September 2025, $1,690 to $1,890 per ton installed, and Brownsville's utility valued the [replacement tower for two 1,000-ton chillers at $1,770,769](https://projects.constructconnect.com/details/7237552-unit-10-cooling-tower-replacement-project&find_loc=texas-78520) in October 2025, about $885 per ton. Thunder Said Energy's model uses [$1,000 per ton installed, $300 per kW of heat, with installation about 60% of it](https://thundersaidenergy.com/downloads/industrial-cooling-chillers-and-evaporators/), and Delta Cooling's own FAQ quotes [$120 to $200 per ton for the equipment](https://deltacooling.com/resources/faqs/cooling-tower-costs). Evapco raised list prices on [towers 3% and on dry and adiabatic coolers 4% from January 5, 2026](https://refindustry.com/news/market-news/evapco-north-america-announces-price-increases-from-jan-5). The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds these awards as its first heat-rejection rows, and [Quote Check](https://open-factory.com/tools/quote-check/) flags a tower quote outside the $885 to $1,890 per ton band until more awards land. So we inverted the question and asked what premium the tower's water bill would pay for. We priced the fan penalty at the difference between the Evapco eco-Air 33S3 and the Evapco AT 112, 27.7 kW per MW at design, averaged to half of that over the year, at each state's [July 2026 industrial electricity rate](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a), 7.07 cents in Texas, 10.07 in Virginia and 25.11 in California. The half-load average and the exclusion of tower pumps are our assumptions, marked (A) in the CSV. Both favor the tower. Ten years of water and sewer come to $1,023,000 per MW in Santa Clara against a $305,000 fan penalty at California's rate, $913,000 against $122,000 in Douglas County, $735,000 against $86,000 in Houston, $563,000 against $113,000 in Phoenix, $363,000 against $122,000 in Northern Virginia and $197,000 against $74,000 in Baton Rouge. **The net ten-year saving from going dry is $791,000 per MW in Douglas County, $718,000 in Santa Clara, $649,000 in Houston and $450,000 in Phoenix under APS, and below $250,000 in Northern Virginia, Chicago, Denver and Baton Rouge.** At a premium of $200 per kW of heat, two thirds of Thunder Said's installed tower cost, the dry plant pays back in 2.5 years in Douglas County, 2.8 in Santa Clara, 3.1 in Houston, 4.4 in Phoenix, 8.3 in Northern Virginia and 16.2 in Baton Rouge, with every market and three premium scenarios in [dry-cooler-breakeven-by-market.csv](data/dry-cooler-breakeven-by-market.csv). A buyer who signs the 14-year Dominion contract in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) and installs a tower in Loudoun is making a defensible call. The same buyer in Douglas County is not. The hybrid changes the arithmetic in one market. If a hybrid cooler sprays only on the hours above 85F dry bulb, its water bill scales with those hours, which gives $20,096 per MW-year in Phoenix, $11,263 in Houston, $6,390 in Douglas County and under $2,000 in Northern Virginia, Columbus, Chicago and Santa Clara. In Phoenix that is 36% of the tower bill, and in Columbus 2.5%, which is why a hybrid is the rational Phoenix machine and an unnecessary one in Ohio. Lead times do not decide between them this year. Linesight's June 2026 tracker puts [chillers, CRAHs and cooling towers within circa 20 to 43 weeks](https://insights.linesight.com/cmi-2026-june/industry-trends-and-supply-chain-dynamics-2026-americas/supply-chain), while the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists Origin Partners' [50 to 70 weeks for water-cooled chillers over 1,000 tons](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026), the subject of [Chillers Are the 60-Week Problem](https://open-factory.com/articles/chiller-60-week-problem/). ## The Fights Are About Permits, Not the Bill The water fights of 2026 are not tariff disputes. In Fayette County, Georgia, two industrial hookups feeding QTS's Project Excalibur campus had been billed wrongly, and by the time the error was found [QTS had consumed more than 29 million gallons and paid $147,474 in retroactive charges at a $6.46 construction rate](https://fortune.com/2026/05/13/data-center-georgia-arizona-water-wars/), as Fortune reported in May. Twenty-nine million gallons is five weeks of a 50 MW tower. In Tucson the city demanded that the Project Blue developer [replace two acre-feet, about 650,000 gallons](https://fortune.com/2026/05/13/data-center-georgia-arizona-water-wars/), trucked out of city limits by a contractor, less than a day of a 50 MW tower. The headline volumes are small against Google's table. The politics turn on who was told, not on what was paid. The rules that bind are written into the tariffs as thresholds. Phoenix's March 2024 ordinance requires any new user above [250,000 gallons a day to file a conservation plan, and above 500,000 to meet 30% of demand with recycled or conserved water](https://www.aztechcouncil.org/phoenix-large-water-use-regulations/). A 50 MW tower draws 820,000 gallons a day at full load, so it crosses both lines, and under the hybrid arithmetic above it crosses neither. Mesa's rate book adds a [$2.96 per thousand gallon surcharge on use above the customer's December-to-February average](https://www.mesaaz.gov/files/assets/public/v/21/government/omb/utility-rates/utility-rate-book.pdf), which a flat tower load avoids and a hybrid, which sprays in summer, does not. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) lists these alongside the electric rules. In Abilene, where the Stargate campus sits in AEP Texas territory, the city told the Texas Tribune the site would [initially use 8 million gallons of city water to fill its cooling system](https://www.texastribune.org/2025/09/25/texas-data-center-water-use/), and the developer's chief executive told a Senate committee the running draw would be [less than 25,000 gallons a day](https://www.fox34.com/2026/09/02/data-center-ceos-testify-before-texas-senate-committee/). That is a closed loop with dry rejection, and it is what a buyer in West Texas should expect to be asked for. Certification is the other thing a specification should include. The Cooling Technology Institute's STD-201 program certifies that [all models of a line of evaporative heat rejection equipment will perform thermally in accordance with the manufacturer's published ratings](https://web.archive.org/web/2026id_/https://www.cti.org/cti-certification), and its test agency confirms the program now [covers open and closed-circuit towers and dry coolers, with adiabatic coolers under development](https://www.cleanair.com/insight/why-heat-rejection-equipment-thermal-performance-certification-are-important/). A hybrid bought today is bought on the vendor's own curve, which the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) scores for Evapco, BAC and SPX: a certified rating counts, a brochure percentage does not. ## What to Do Monday Pull the water tariff before the electric one is signed, because the water bill is the one you can still design around. Ask for the rate code a large meter lands in: Douglas County's 8-inch Tier 3, Dallas's [Optional General Services at $5.11 above the first million gallons](https://dallas.gov/departments/waterutilities/DCH%20Documents/monthly_rate_sheet.pdf), Fort Wayne's over-50,000-cubic-foot block, Loudoun's Tier 1 within reserved capacity. Then ask in writing whether a sewer deduct meter is permitted on the makeup line, whether a reclaimed line reaches the parcel and at what rate, and what the large-user threshold is in gallons per day. Write cycles of concentration into the treatment contract as a floor, not a target. The EPA's own figure is that [moving from three to six cycles cuts makeup 20% and blowdown 50%](https://www.epa.gov/system/files/documents/2023-05/ws-commercial-watersense-at-work_Section_6.3_Cooling_Towers.pdf), and where sewer costs more than water, as in Salt Lake City, the blowdown half of that is the larger saving. Specify CTI STD-201 certification for any tower or dry cooler, and ask the hybrid vendor for a test report under the adiabatic code once CTI publishes it. Push the facility water class as high as the IT will take, because a W40 loop lets the fans do the work the water did. [How Hyperscalers Buy Chillers and Colo Developers Do Not](https://open-factory.com/articles/ofci-vs-design-build-chillers/) covers the owner-furnished route, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) puts the 20-to-43-week tower slot against the 50-to-70-week chiller slot so the two arrive in order. Behind the paywall: the 50 MW heat-rejection bill of materials three ways, open tower, dry and hybrid, with the per-MW lines for equipment, installation, fans, water and treatment. The four markets where leaving the tower pays back inside **4.4 years** at a $200 per kW premium and the five where a tower should stay. The six-vendor shortlist with what each publishes, and the five clauses for the water service agreement, all tied to the component lines in the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # Fiber Laser Cutters: What a $500,000 Price Gap Buys URL: https://open-factory.com/articles/fiber-laser-price-gap/ Date: 2026-10-11 Machinery / Deep dive Chinese 12 kW machines from $45k FOB, the 41% duty stack, IPG's 10-K and Raycus at a 2.88% margin, federal awards from $645k to $1.75M, Trumpf and Bystronic terms, the PO checklist A Chinese exporter's own price guide puts a fully enclosed 6 to 12 kW fiber laser with an exchange table at [$45,000 to $85,000 FOB](https://arcuscnc.com/fiber-laser-cutter-price/), hardware only, as of December 2025. Federal Prison Industries paid [$645,000](https://www.usaspending.gov/award/CONT_AWD_15UM1L25F00000571_1542_15UC0C25D00001520_1542) to Trumpf's Pennsylvania dealer for one fiber laser cutting machine in July 2025, and two Army depots paid [$1,610,580](https://www.usaspending.gov/award/CONT_AWD_W911KF25CA003_9700_-NONE-_-NONE-) and [$1,750,000](https://www.usaspending.gov/award/CONT_AWD_W51AA125C0011_9700_-NONE-_-NONE-) for theirs in the same quarter. Land the Chinese machine at the top of its band, pay the 41% duty stack, rig it, and it sits on the floor at about $128,000. The gap to the cheapest Western award is $504,000 on one machine, and this piece is about what that money buys. It is the first worked example for the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) in this machine class, and every price in it is a guide, a listing, an award or a filing, dated, with whose number it is. The ladder runs $45,000 to $85,000 for the exporter's enclosed 6 to 12 kW band, $100,000 and up for Gweike's 12 kW tier, $127,850 for our landed and dutied arithmetic, $185,000 for a Chinese-brand 12 kW at a US dealer, $299,900 for a 12 kW Otinus with a tube cutter, installed and trained, then $645,000, $1,233,175, $1,610,580 and $1,750,000 for the four federal awards. The piece covers where those prices come from, the 41% duty, the laser source both sides buy from the same three suppliers, the service and paper inside the Western price, a ten-year comparison with every assumption marked, what to do Monday, and behind the wall the PO checklist and the shortlist by buyer type. ## Two Price Lists for One Machine Class The Chinese price is public because the exporters publish it to win the quote. Arcus, a Jinan exporter, writes that a [$30,000 FOB machine "will likely cost you $42,000 by the time it sits on your floor in Texas"](https://arcuscnc.com/fiber-laser-cutter-price/) and tells the buyer to ask for a DDP quote. The same page prices the exchange table at about $15,000, a 6025 bed at $15,000 over a 3015, and an IPG source at [$5,000 to $15,000 over a Raycus](https://arcuscnc.com/fiber-laser-cutter-price/), rising to about $20,000 on a 12 kW machine. Gweike's guide puts [12 kW and up at $100,000 plus](https://www.gwklaser.com/fiber-laser-cutting-machine-price.html) with no incoterm stated. STYLECNC lists its 1.5 to 6 kW 3015 at [$19,500 to $31,000](https://www.stylecnc.com/fiber-laser-cutting-machine/Fiber-laser-cutting-machine-1200W-laser-cut-metal-signs.html) with a one-year warranty on the entire machine, updated 28 September 2026. These are asking prices from the seller's own page, not transactions, and they carry no n, which is the first thing [Quote Check](https://open-factory.com/tools/quote-check/) flags when a buyer pastes one in. The Western price exists in public only where a public body bought one. We verified each award against the USAspending API on 11 October 2026. Letterkenny Army Depot paid [$1,233,175 in August 2022](https://www.usaspending.gov/award/CONT_AWD_W911N222C0780_9700_-NONE-_-NONE-) for a 12 kW fiber laser with dual shuttle tables of 242 by 80 inches, on one offer. The Marine Corps paid Cincinnati Incorporated [$1,289,608 in May 2024](https://www.usaspending.gov/award/CONT_AWD_M6700424P5031_9700_-NONE-_-NONE-) on eight offers. An award value includes delivery, installation, training and whatever automation the solicitation asked for, so it is a system price, not a list price, and the kilowatts behind the UNICOR and Tobyhanna numbers sit in attachments the public record does not show. **Neither side publishes a transaction price for a configured 12 kW machine with a dual shuttle, so the honest statement of the gap is a range: $504,000 against the lowest 2025 federal award and $1.1 million against the only 12 kW federal award on record.** The lesson of [What a Haas VF-2 Really Costs](https://open-factory.com/articles/haas-vf2-true-cost/) holds here with the sign reversed: in machining centers Haas publishes a list and nobody else does, and in sheet-metal lasers the Chinese builders publish and the Western builders do not, which is why Trumpf, Bystronic and Amada score where they do on the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) and why the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has to go to federal awards for a Western number at all. ## The 41% Duty Stack and What It Changes A laser cutting machine for working metal enters under HTS 8456.11.10 at a [column 1 rate of 3.5%](https://hts.usitc.gov/reststop/search?keyword=8456.11) in the 2026 schedule, Revision 21. The line sits on Section 301 List 1, which added [25% from 6 July 2018](https://www.federalregister.gov/documents/full_text/text/2018/06/20/2018-13248.txt) under heading 9903.88.01 and was left at 25% by the [September 2024 four-year review](https://www.federalregister.gov/documents/full_text/text/2024/09/18/2024-21217.txt). The July 2026 Section 301 action on forced-labor enforcement added a further [12.5% on products of China](https://www.federalregister.gov/documents/full_text/text/2026/07/28/2026-15181.txt) under heading 9903.05.31 from 24 July 2026, per [CBP CSMS 69326983](https://content.govdelivery.com/accounts/USDHSCBP/bulletins/421d887). Those three instruments sum to 41% of customs value. The IEEPA tariffs on top of that died with [Learning Resources v. Trump](https://www.supremecourt.gov/opinions/25pdf/24-1287_4gcj.pdf) on 20 February 2026, and the 10% Section 122 surcharge that replaced them [expired on 24 July 2026](https://www.federalregister.gov/documents/full_text/text/2026/02/25/2026-03824.txt). China pays 3.5 plus 25 plus 12.5 for 41%, Switzerland and Japan pay 12.5% all-in under the July notice's cap, Germany and Italy pay 10% all-in, and a machine assembled in Farmington, Connecticut or Hoffman Estates, Illinois pays nothing. No 8456.11 line appears on any Section 232 metals annex, including the [407 derivative codes added in August 2025](https://www.federalregister.gov/documents/full_text/text/2025/08/19/2025-15819.txt). The open item is the Section 232 investigation into robotics and industrial machinery, whose scope names ["laser and water-cutting tools and machinery"](https://www.federalregister.gov/documents/full_text/text/2025/09/26/2025-18749.txt) and whose 270-day clock ran out in May 2026 with no proclamation in the Federal Register as of 11 October 2026. The [September 2026 tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) walks every machinery line through the same instruments, and the [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) tracks the docket. **On an $85,000 machine the stack is $34,850, which is real money and still only 7% of the gap to the cheapest Western award.** Duty is assessed on customs value, the FOB invoice for a direct import. A Western machine built abroad pays its 10 or 12.5% on the OEM's transfer price, inside the quoted number, and Bystronic's US terms say so: ["All applicable U.S. customs or duties, Installation, and Commissioning are included in the price"](https://www.bystronic.com/sites/default/files/2025-11/BYSTRONIC%20INC_2025%20Terms%20and%20Conditions%20of%20Sale_11.3.2025%20letterhead.pdf). Bystronic also wrote in its 2025 annual report that it advanced ["the ramp-up of our facility in Hoffman Estates, USA"](https://www.bystronic.com/sites/default/files/2026-03/Short%20Report%20-%20Annual%20Report%202025%20%281%29.pdf) to deliver out of the United States and blunt the tariff, and Trumpf's Farmington site has produced laser machinery, resonators and automation on [about 400,000 square feet](https://www.trumpf.com/filestorage/TRUMPF_US/Product_Brochures_and_publications/TRUMPFInc-FactSheet.pdf) since before the current schedule. The question to put on the RFQ is the country of origin of the machine that will ship, not the address of the brand. ## The Imports Did Not Slow Down The trade data show how little the stack has slowed the flow. US imports of laser-operated machine tools under HS 845611 ran [$587 million in 2025](https://comtradeapi.un.org/public/v1/preview/C/A/HS?reporterCode=842&period=2025&cmdCode=845611&flowCode=M&partner2Code=0&customsCode=C00&motCode=0), with China at $66.5 million and 4,419 machines, Germany at $133.9 million and 2,162, Italy at $103.5 million, Japan at $106.3 million and Switzerland at $45.1 million, per Census figures republished by UN Comtrade. China's first half of 2026 came to [$40.7 million against $33.7 million in the first half of 2025](https://comtradeapi.un.org/public/v1/preview/C/M/HS?reporterCode=842&period=202606&cmdCode=845611&flowCode=M&partnerCode=0,156,276,392,380,757&partner2Code=0&customsCode=C00&motCode=0), summed from the monthly series, through a February with IEEPA, a spring with Section 122 and a July with the new 301 rate. China's line reads $59.4 million in 2022, $65.9 million in 2023, $73.1 million in 2024 and $66.5 million in 2025, while Germany swings from $161.4 million in 2023 to $91.0 million in 2024 and back to $133.9 million. The heading covers every machine tool operated by a laser, marking machines included, so the unit counts overstate cutting machines and the averages below understate them. Japan's 431 machines averaged $246,737 of customs value in 2025, Italy's 1,249 averaged $82,906, Switzerland's 713 averaged $63,240, Germany's 2,162 averaged $61,934 and China's 4,419 averaged $15,057. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) does not yet track laser cutters as a class, and this unit series is the first input when it does. ## The Laser Source Is Not Where the Gap Is Both sides of the gap buy the laser from one of three places: IPG in Oxford, Massachusetts, or a Chinese maker, or the OEM's own disk or fiber plant. IPG's 10-K for 2025 reports that high-power continuous-wave lasers, the cutting and welding sources, brought in [$308.8 million, 30.8% of revenue, against $525.0 million and 40.8% in 2023](https://www.sec.gov/Archives/edgar/data/1111928/000111192826000040/ipgp-20251231.htm), a 41% fall in two years. The 2024 10-K gave the reason in a sentence: ["increased competition impacted sales in cutting"](https://www.sec.gov/Archives/edgar/data/1111928/000111192825000023/ipgp-20241231.htm). The 2025 filing names Wuhan Raycus and Maxphotonics as competitors, warns that ["market participants, particularly in China, have reduced and may continue to reduce, prices"](https://www.sec.gov/Archives/edgar/data/1111928/000111192826000040/ipgp-20251231.htm), and says the Asian competitors ["are beginning to export their fiber lasers and fiber laser-based systems to non-Asia markets."](https://www.sec.gov/Archives/edgar/data/1111928/000111192826000040/ipgp-20251231.htm) IPG's gross margin was 38.0% in 2025 and the company booked about 120 basis points of tariff cost in it. The left panel reads $525 million, $333 million and $309 million for 2023 to 2025, and the right panel reads about 6,800 Raycus lasers of 10 kW and above in 2024, 8,561 in 2025, and more than 10,000 for Max Photonics in 2025. Raycus disclosed that it sold [186,000 lasers in 2025 and cut product cost by more than 20%](https://news.qq.com/rain/a/20260423A02X1G00), as reported by Securities Times from the annual report, and that its [gross margin touched 2.88% in the fourth quarter of 2024](https://www.jiemian.com/article/12648144.html) after a year in which it blamed the revenue miss on price declines, as reported by Jiemian. Max Photonics booked [RMB 3.97 billion in 2025 and 28.6% of the Chinese fiber laser market](https://news.sina.cn/sx/2026-09-30/detail-initpatt7639131.d.html), per the 2026 China Laser Industry Development Report as reported by Sina. nLIGHT, the other US source maker, saw its industrial revenue fall from [$71.0 million in 2023 to $38.8 million in 2025](https://www.sec.gov/Archives/edgar/data/1124796/000112479626000012/lasr-20251231.htm) on lower unit sales of cutting lasers. **A 12 kW source from Raycus or Max is a commodity sold near cost, and the same exporters will fit an IPG for about $20,000 more, so the source explains at most $20,000 of a $504,000 gap.** The exporter's number comes from [Arcus](https://arcuscnc.com/fiber-laser-cutter-price/), and a UK reseller of Chinese machines puts the ratio at [about £20,000 for a 6 kW Chinese source against £100,000 to £120,000 from IPG](https://fabertec.co.uk/news/why-fabertec-fiber-lasers-are-so-affordable), a vendor blog with no date. What an IPG or a Trumpf disk source buys is a longer stated life and a wider warranty, not more kilowatts. IPG's product page claims [electrical efficiencies over 40% and operation "for over a decade"](https://www.ipgphotonics.com/en/products/lasers/high-power-cw-fiber-lasers), and Bystronic's terms offer [sixty months of cover on the source and head](https://www.bystronic.com/sites/default/files/2025-11/BYSTRONIC%20INC_2025%20Terms%20and%20Conditions%20of%20Sale_11.3.2025%20letterhead.pdf) under a premium plan. Bodor's Iowa dealer offers [five years on the source and head](https://smith.machinehub.com/listing/1-new-12kw-bodor-fiber-laser-cutting-machine-model-a6p/) on a 12 kW machine with a Bodor-branded source, which is the Chinese side's answer and worth exactly as much as the dealer's balance sheet. ## What the Western Price Contains A federal award of $1.2 million for a 12 kW machine with dual shuttles is a system: machine, tables, delivery, installation, training and a dealer that answers the phone. The Chinese $85,000 is a crate at a Chinese port. Everything between the two is a line a buyer can price. Installation and acceptance come first. Trumpf's US terms say the machine is ["subject to final acceptance testing after installation at the installation site"](https://trumpf.com/filestorage/TRUMPF_CA/CA_Terms_and_Conditions/TRUMPF_INC_TERMS_AND_CONDITIONS_OF_SALE_FOR_LASER_TECHNOLOGY_EQUIPMENT_AND_AUTOMATION_Y2024-07-01.pdf), run by the seller's operator to the seller's procedure plus any tests "specified in the purchase order", with the machine deemed accepted at production use or three months after delivery, whichever comes first. The same document puts rigging and placement on the buyer and payment at 20% with the order, 70% before shipment and 10% net ten days after installation completion. Bystronic's terms include installation and commissioning in the price and give the buyer three options at the end: accept, accept with conditions, or do neither and ["authorize BYSTRONIC to immediately disable the Product from functioning until Buyer accepts"](https://www.bystronic.com/sites/default/files/2025-11/BYSTRONIC%20INC_2025%20Terms%20and%20Conditions%20of%20Sale_11.3.2025%20letterhead.pdf). The same terms state that production estimates ["are not guaranteed or warranted"](https://www.bystronic.com/sites/default/files/2025-11/BYSTRONIC%20INC_2025%20Terms%20and%20Conditions%20of%20Sale_11.3.2025%20letterhead.pdf), the sentence to read twice before believing any vendor's sheets-per-hour number. Trumpf warrants one year from installation completion, Bystronic twelve months conditional on a ByCare service contract and sixty months on the source and head with premium cover, Mazak's machine-tool terms give [two years from start-up](https://www.mazak.com/us-en/mpower/invoice-installation-terms-and-conditions/) with a 45-day defect notice and liability capped at the purchase price, Bodor through its Iowa dealer offers five years on source and head and three on main parts with 50% down and 50% before shipment, HSG through a Milwaukee dealer offers [three years on the machine and two on the source](https://surplusrecord.com/listing/hsg-g3015x-fiber-laser-cutting-machine-60-width-x-120-l-sheet-size-5550-ipm-6-kw-power-1056016/), and a direct STYLECNC import comes with one year. Amada America publishes no US terms we could find. Every Western document caps the seller's liability at the price of the item and excludes consequential damages, so the warranty is the only downtime insurance in the contract, the point [Liquidated Damages at 0.5% a Week](https://open-factory.com/articles/liquidated-damages/) made for switchgear. Service is the second line, and it has a public rate. Hymson's benchmark table puts a laser or 5-axis specialist at [$250 to $400 an hour, an OEM distributor technician at $150 to $225, and travel at $1.50 to $2.00 a mile](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/), and it puts annual maintenance at 2 to 5% of purchase price. The federal record shows what OEM service costs when a depot buys it as a contract: the Navy paid Trumpf [$39,640 in August 2025](https://www.usaspending.gov/award/CONT_AWD_N3904025P0062_9700_-NONE-_-NONE-) for preventive and corrective maintenance on a TruMatic laser system, and the Army paid Mazak Optonics [$145,008 in October 2025](https://www.usaspending.gov/award/CONT_AWD_W519TC25P2385_9700_-NONE-_-NONE-) for laser preventive maintenance. An Air Force order for one [Trumpf optical cutting head came to $18,600 in March 2022](https://www.usaspending.gov/award/CONT_AWD_FA812522P0025_9700_-NONE-_-NONE-), the spare that decides whether a crash costs a day or a month, and the reason the [recommended spares list](https://open-factory.com/articles/recommended-spares-margin/) belongs in the PO. ## Nozzles, Nitrogen and Power Consumables are the one line that does not separate the two machines, because a 12 kW head eats the same nozzles and windows whoever built the frame. Airgas sells a Bystronic-fit 1.5 mm brass double nozzle at [$8.10 each](https://newemarket.airgas.com/product/Welding-Products/Welders-%26-Accessories/Laser-Cutting-Consumables/p/RAD64006667), a Hypertherm ten-pack for the same machine lists at [$45.55](https://shop.arcsolinc.com/products/hypertherm-centricut-by-nozzle-double-push-fit-1-5-mm-10-pk-bystronic-ref-10035278-nk15-15-by310-5278x), and Laguna's D32 nozzles for Chinese heads run [$31.90 to $35.20](https://parts.lagunatools.com/product-category/cnc-accessories/cnc-laser-accessories/), all read on 11 October 2026. Assist gas is the line that dwarfs them. A public bid at Oklahoma State University prices 99.999% nitrogen at [$0.04 a cubic foot in liquid and $0.07 to $0.10 in 300 cubic foot cylinders](https://airs.okstate.edu/okcorral/files/Compressed%20Gas%20Bid%20Price%20Summary.pdf), and a 2007 article in The Fabricator put high-pressure nitrogen cutting at [2,200 to 3,800 cubic feet an hour](https://www.thefabricator.com/thefabricator/article/lasercutting/achieving-efficiency-with-occasional-high-pressure-high-flow-nitrogen-use), which at the liquid price is $88 to $152 an hour of beam-on stainless. That arithmetic, not the machine, is why shops cut mild steel with oxygen, aluminum with air, and buy a nitrogen generator in year two. Power at the [EIA industrial average of 9.77 cents per kWh for July 2026](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a) is a smaller line again: a dealer listing for a 12 kW Otinus states [73 kW of draw](https://revelationmachinery.com/product/12000-watt-5-x-10-otinus-fiber-laser-cutting-machine-w-tube-cutter-2024-in-stock-install-training-included/), which is about $7 an hour at full load. ## Ten Years of Owning Each One We put one machine from each side through the same ten years, pricing every line that differs between them from the sources above and marking every assumption (A). The Chinese machine is the $85,000 top of the Arcus band, bought FOB. The Western machine is the $645,000 UNICOR award, with duty, installation and commissioning inside the price per Bystronic's terms. Gas, power, nozzles and the operator are the same for both and are left out. | Line | Chinese import | Western OEM | Gap | |---|---|---|---| | Machine, as quoted | $85,000 | $645,000 | $560,000 | | Ocean freight | $5,000 | $0 | -$5,000 | | Duty on customs value, 41% | $34,850 | $0 | -$34,850 | | Broker, entry, FDA Form 2877 (A) | $1,000 | $0 | -$1,000 | | Rigging and forklift | $3,000 | $0 | -$3,000 | | NRTL field evaluation (A) | $7,500 | $0 | -$7,500 | | Installation and training (A) | $5,000 | $0 | -$5,000 | | Landed and running, year 0 | $141,350 | $645,000 | $503,650 | | Maintenance, years 2 to 10, 3.5% of price a year (A) | $26,775 | $203,175 | $176,400 | | Resale at year 10 (A) | -$15,000 | -$100,000 | -$85,000 | | Net ten-year cost | $153,125 | $748,175 | $595,050 | *Source: Open Factory [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/). CSV: [data/ten-year-gap.csv](data/ten-year-gap.csv)* The Chinese machine lands and runs for $141,350 and the Western one for $645,000, a gap of $503,650. Add nine years of maintenance at the midpoint of Hymson's 2 to 5% benchmark and a year-ten resale, and the ten-year net is $153,125 against $748,175, a gap of $595,050. The maintenance line is a share of purchase price, so it widens the gap by construction, and no public source measures what a buyer actually wants to know, which is hours of unplanned downtime a year on each machine. The resale line is anchored on one side only: a [2015 Amada FOL 4 kW sold at $99,990](https://surplusrecord.com/listing/amada-fol-3015aj-4000-watt-laser-5-x-10-122-x-62-y-3-93-z-875-mild-steel-capacity-amnc-f-cnc-control-2015-1098799/) through KD Machinery in 2026, while no ten-year-old Chinese fiber laser with a visible sale price exists on Surplus Record, which is a finding in itself. The bars read $699,990 for a 2020 LVD 10 kW gantry, $450,000 for a 2019 Mazak Optiplex 8 kW, [$279,990 for a 2018 Trumpf TruLaser 3030 fiber 6 kW with a pallet changer](https://surplusrecord.com/listing/trumpf-trulaser-3030-cnc-fiber-laser-6000-watt-5-x-10-table-automatic-pallet-changer-trudisk-6001-siemens-sinumerik-840d-sl-control-2018-1097827/), $274,990 for a 2016 Mazak 6 kW with automation, $189,997 for a 2022 HSG 6 kW with a tube attachment marked sold, $139,500 for a 2026 Bystronic BySmart, $99,990 for the 2015 Amada marked sold, and $69,900 for a Chinese HK 4 kW with a chiller, all on [Surplus Record](https://surplusrecord.com/s/fiber-laser/) on 11 October 2026. Asking is not sold, the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds no laser hammer prices yet, and the method we laid out in [Building the Missing Index](https://open-factory.com/articles/used-cnc-auction-index/) is the only way to turn these into a curve. What the listings already say is that a Western laser is a resalable asset at eight years and a Chinese one has no public price past four. **The gap buys three things with a document behind them, a service network at $150 to $400 an hour that exists, a warranty of one to five years from a seller whose terms are published, and an acceptance test with the buyer's part in it, and one thing with no document behind it, which is uptime.** The machine-tool order data say the market has decided. US orders ran [$3.44 billion over 11,243 units in the first half of 2026](https://www.americanmachinist.com/news/news/55397089/machine-tool-demand-is-booming-in-2026-us-manufacturing-technology-orders-report-june-2026), an average of about $306,000 that we unpacked in [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/), and Trumpf reported the United States as its largest single market for the first time in the year to June 2026 at [about 760 million euros, up 16%](https://www.trumpf.com/en_INT/newsroom/global-press-releases/press-release-detail-page/release/trumpf-leaves-crisis-behind-sales-just-above-previous-year-order-intake-rises-by-7-percent-leibinger-kammueller-slight-economic-turnaround-us-overtakes-germany-as-largest-single-market-10217/). Bystronic posted an [EBIT of minus 23.4 million francs in the first half of 2026](https://www.webdisclosure.com/press-release/bystronic-accelerates-transformation-amid-structural-market-changes-wFKcYUUr5y1) and leans on DNE Laser, its Chinese entry brand, for orders. The Western builders are not losing the shops that run two shifts with a tower. They are losing the single-shift job shop that was never going to pay $645,000, the shop [SendCutSend](https://open-factory.com/articles/sendcutsend/) serves with a part in two to four days instead of a machine. ## What to Do Monday Get the Chinese quote in DDP terms, with the HTS line, the country of origin and the three duty headings written on it, and get the Western quote with the acceptance procedure attached. Then run both through [Quote Check](https://open-factory.com/tools/quote-check/) against the ladder above and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) at 41% for China, 10% for Germany and Italy, 12.5% for Japan and Switzerland, and 0% for a machine built in Connecticut or Illinois. A Chinese quote that does not show the duty is hiding $34,850 on an $85,000 machine, and a Western quote that does not state origin is hiding whether the January 2028 cliff in the [tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) applies to it. Before you wire a deposit on a direct import, confirm two documents exist. The first is the manufacturer's FDA product report, which [21 CFR 1002.10](https://www.ecfr.gov/current/title-21/chapter-I/subchapter-J/part-1002/section-1002.10) requires before a Class IV laser product enters commerce and which the importer declares on Form FDA 2877 at entry per the [FDA's import page](https://www.fda.gov/radiation-emitting-products/getting-radiation-emitting-product-market-frequently-asked-questions/importing-and-exporting-electronic-products). A machine whose maker has never filed one is a machine the port can hold. The second is the plan for electrical approval, because [29 CFR 1910.303(a)](https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.303) makes the equipment acceptable only if approved, and an unlisted machine needs a one-unit field evaluation that no NRTL prices on its website. The Western dealer ships both as a matter of course, which is part of the $504,000 and the part most easily forgotten. Write the acceptance test yourself and attach it to the PO, because Trumpf's terms say the seller's operator runs the seller's procedure plus whatever the PO specifies, and Bystronic's say the machine can be disabled if you neither accept nor condition. The test is a runoff part in your thickest production material, measured against [ISO 9013:2017](https://www.iso.org/standard/60321.html), the thermal-cut quality standard that covers laser cuts from 0.5 to 32 mm, with the perpendicularity and roughness class named and the final 10% tied to it. The [capital equipment RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) has the milestone and FAT language. And put the price you pay, with the configuration, into the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) form, because eight federal awards and two exporter guides are the whole public record for this machine class today, and a cell publishes only when five contributors and eight quotes are in. Behind the paywall: the eleven lines of a fiber laser PO, including the runoff clause that moves the final **10%** from "installation completion" to a measured part, the spares list by part number class, and the shortlist of which machine to buy by buyer type, with the one case where the direct import wins and the two where it does not, built on the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) worked example above. --- # Tariffs, September 2026: The Actual Schedule for Machinery and Electrical Gear URL: https://open-factory.com/articles/tariff-schedule-2026/ Date: 2026-09-09 Procurement / Deep dive IEEPA struck down Feb 20, Section 122 ran to Jul 24, Section 301 forced-labor tariffs on 60 economies, Section 232 metals at 50/25/15/10 with the Annex III 15% tier to Dec 2027, the robotics probe with no rate, HTS by HTS A 2,500 kVA pad-mount transformer entered at Long Beach on September 9, 2026 pays 25% duty whether it was built in Monterrey, Osaka or Mumbai, and 50% if it was built in Jiangsu. A low-voltage panel with the PLC already in it pays 0% from Mexico, 10% from Germany, 12.5% from Japan and 40.2% from China. Those two numbers are the whole story of the 2026 tariff schedule: the metal tariff no longer cares where the box came from, and the country tariff no longer touches the metal. Five legal authorities have carried US tariffs since February 2025 and two of them are dead, so every rate below carries its instrument, its Federal Register or CBP citation, and the HTS revision it was read from ([HTS 2026 Revision 18](https://hts.usitc.gov/)). The [Open Factory Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) holds the same 198 HTS-by-origin cells with the citation on every one. Nine origins, one 25% bar each, and a 50% bar for China: HTS 8504.22.00 sits in Annex I-B of the April metals proclamation, which puts 25% on the full customs value regardless of who made it, and China adds the 25% List 1 rate it has carried since 2018. This piece covers what died and what replaced it, the Section 232 tiers line by line, the two Section 301 actions and how they stack on China, the HTS-by-origin table with the CSV, the $80 billion of 2025 imports now under this schedule, a $306k machine landed from five countries, and, behind the wall, the January 1, 2028 cliff and the three purchase-order clauses that handle it. ## What Died and What Replaced It **Since February 2025 a US equipment buyer has paid under five statutes, and the two that carried the headline rates are gone.** The Supreme Court held 6 to 3 on February 20, 2026 in [Learning Resources v. Trump](https://www.supremecourt.gov/opinions/25pdf/24-1287_4gcj.pdf) that IEEPA does not authorize tariffs, which ended the fentanyl tariffs on Canada, Mexico and China (EOs 14193 to 14195, 90 FR 9113 to 9121) and the reciprocal tariffs of EO 14257 ([90 FR 15041](https://www.federalregister.gov/documents/2025/04/07/2025-06063/regulating-imports-with-a-reciprocal-tariff-to-rectify-trade-practices-that-contribute-to-large-and)). The same day the President signed [Proclamation 11012](https://www.federalregister.gov/documents/2026/02/25/2026-03824/imposing-a-temporary-import-surcharge-to-address-fundamental-international-payments-problems) (91 FR 9339), a Section 122 balance-of-payments surcharge of 10% for the statutory maximum of 150 days, effective February 24, with goods already under Section 232 and USMCA-qualifying goods exempt. It expired at 12:01 a.m. on July 24, 2026. The Court of International Trade struck the Section 122 surcharge too, on May 7, 2026, in Oregon v. United States, but limited relief to the three plaintiffs and the government appealed the next day, so the other 330,000 importers kept paying until it lapsed ([Skadden, May 2026](https://www.skadden.com/insights/publications/2026/05/us-trade-court-strikes-down-section-122-tariffs)). The replacement was ready to the hour. USTR opened 60 Section 301 investigations on March 12, 2026 into economies that do not enforce a forced-labor import ban ([91 FR 12884](https://www.federalregister.gov/documents/2026/03/17/2026-05151/initiation-of-section-301-investigations-of-acts-policies-and-practices-of-various-economies-related-to-the)), found all 60 actionable on June 2, took 1,600 comments and three days of hearings, and imposed tariffs of 10% or 12.5% effective 12:01 a.m. July 24, 2026 ([91 FR 47318](https://www.federalregister.gov/documents/2026/07/28/2026-15181/notice-of-actions-in-section-301-investigations-of-acts-policies-and-practices-of-various-economies)), the minute Section 122 died. Fifteen dated instruments, seven of them in 2026 alone, and the three shaded rows that still bill you are Proclamation 11021 (April 6), the Section 301 forced-labor action (July 24) and the sunset already written into the first for January 1, 2028. Meanwhile the refunds for the dead authority are real money: CBP told the Federal Register that IEEPA duties collected from February 3, 2025 to February 24, 2026 total [an estimated $166 billion across 53 million entry summaries](https://www.federalregister.gov/documents/2026/07/08/2026-13771/agency-information-collection-activities-extension-court-ordered-refunds-under-the-international-emergency), all of which the CIT ordered refunded with interest on March 4, 2026 in Atmus Filtration v. United States. Who receives that money, the importer of record or the buyer who paid the surcharge line on the invoice, is the subject of [Who Gets the IEEPA Refund](https://open-factory.com/articles/ieepa-refund-fight/). The schedule below is about what you pay from here. Section 122 is the second refund fight, and a smaller one: 10% for 150 days on non-232, non-USMCA goods, so a Japanese CNC lathe entered in April 2026 paid 10% plus the 4.4% column 1 rate and nothing else. If the Federal Circuit affirms Oregon v. United States for everyone rather than for three plaintiffs, that 10% comes back only to importers who preserved the claim, which for entries liquidating this autumn means a protest under 19 U.S.C. 1514 within 180 days of liquidation ([Skadden, May 2026](https://www.skadden.com/insights/publications/2026/05/us-trade-court-strikes-down-section-122-tariffs)). Nothing on this schedule moves for a buyer who paid a supplier's DDP price, unless the PO says the refund is theirs. ## Section 232 Metals: One Rate for Everyone **Proclamation 11021 of April 2, 2026 rebuilt the metals tariff into four tiers on the full customs value of the product, and it is the only instrument on this schedule that charges Osaka and Jiangsu the same.** The tiers, from [91 FR 18201](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states) and [CBP CSMS 68253075](https://content.govdelivery.com/bulletins/gd/USDHSCBP-4117593): 50% for articles of steel, aluminum and copper and the derivatives in Annex I-A (heading 9903.82.02); 25% for the derivatives in Annex I-B (9903.82.09); a transitional tier for Annex III "fixed industrial machinery and power equipment" under which column 1 duty plus Section 232 sums to 15% through December 31, 2027 (9903.82.10); and 10% for any derivative whose metal was melted and poured in the United States (9903.82.06), a threshold Proclamation 11032 cut from 95% to 85% of metal weight on June 8 ([91 FR 34085](https://www.federalregister.gov/documents/2026/06/04/2026-11314/further-adjusting-the-tariff-regimes-for-imports-of-aluminum-steel-and-copper-into-the-united-states)). Articles outside chapters 72 to 76 with less than 15% metal by weight pay nothing under 9903.82.03, but the importer reports the metal weight in kilograms on the entry line. The surprise for an electrical buyer is which annex each transformer landed in. The [CBP Metals HTS List](https://content.govdelivery.com/attachments/USDHSCBP/2026/05/06/file_attachments/3641718/Metals%20HTS%20LIST%20050426.docx) that implements the annexes puts liquid-filled units up to 10,000 kVA (8504.21.00 and 8504.22.00) and dry-type units from 16 to 500 kVA (8504.33.00) in Annex I-B at a flat 25%, while liquid-filled units over 10,000 kVA (8504.23.00) and dry-type units over 500 kVA (8504.34.00) sit in Annex III at the 15% cap. A 2,500 kVA pad-mount for a plant pays 25%. A 50 MVA GSU for a datacenter pays 15% all-in, or 13.4% on top of nothing if it enters under KORUS from Hyundai's Ulsan works. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) shows the 50 MVA unit at 128 to 160 weeks and the pad-mount at 40 to 65 weeks, so the cheaper tariff attaches to the unit you cannot get. Switchgear is not on the list at all. Neither 8535 nor 8536 nor 8537 appears in any annex. Only 8538.10.00 (empty boards and cabinets) and one 8536.90 parts line carry the 25%. MV switchgear assemblies (8537.20.00), motor control centers (8537.10.60) and LV panels and PLC cabinets (8537.10.91) pay no Section 232 at any origin. Gas turbines split at 5 MW: 8411.81.80 is an Annex I-B steel derivative at 25%, 8411.82.80 above 5 MW is uncovered. Diesel gensets (8502.11 to 8502.13) are uncovered; spark-ignition gensets (8502.20.00) pay 25%. Chillers (8418.69.01) are uncovered; household refrigeration is not. Of the machine-tool headings 8456 to 8466 only machining centers (8457.10.00) appear, in Annex III, and of the robot lines only 8428.70.00 lifting-and-handling robots, also Annex III; industrial robots under 8479.50.00 pay nothing under 232. Cable is the messiest: LV cable with connectors (8544.42.90, 8544.42.10, 8544.42.20) and 8544.49.10 are copper articles at 25% and now need country of smelt and cast reported ([CSMS 68855869](https://content.govdelivery.com/bulletins/gd/USDHSCBP-41aa83d)), the aluminum lines 8544.49.20, 8544.49.90, 8544.60.20 and 8544.60.60 are Annex I-B at 25%, and MV copper cable over 1 kV (8544.60.40) is on no list. The copper side of that is in [Copper at $6.77](https://open-factory.com/articles/copper-tariff-panel/) and priced weekly in the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/). Two clauses matter for anyone signing a PO this quarter. Clause 6 of Proclamation 11021 lets Commerce and USTR revoke the Annex III 15% tier for any single trading partner whose shipments "increased in a manner that undermines the objectives" of the action, by Federal Register notice, effective on the day of the finding. And clause 7 ends the tier for everyone at 11:59 p.m. on December 31, 2027, after which Annex III articles pay the Annex I-B 25%. Proclamation 11032 added a parallel Annex I-C for mobile industrial equipment (forklifts, dozers, mobile cranes) at 25%, or 15% for products of Japan, Korea, Taiwan, the UK, Switzerland and the EU, or 25% on non-US content with a 15% floor for USMCA goods from Mexico and Canada, with the same December 31, 2027 end date. The 232 cost of a full lineup, transformer plus switchgear plus UPS plus genset, is built SKU by SKU in [Tariff Bill of Materials](https://open-factory.com/articles/tariff-bom-electrical/). The paperwork is now part of the price. Every entry under 9903.82.03, the under-15%-metal exemption, reports the aggregate metal weight in kilograms as a second quantity on the line, which means the importer needs the bill of materials and a weight calculation on file before the container lands ([CSMS 68253075](https://content.govdelivery.com/bulletins/gd/USDHSCBP-4117593)). Steel derivatives still report country of melt and pour and aluminum derivatives country of smelt and cast, and CBP switched on copper smelt-and-cast reporting for the four 8544.42 and 8544.49 cable lines on July 30, 2026 ([CSMS 69252300](https://content.govdelivery.com/bulletins/gd/USDHSCBP-420b4cc)). The 10% US-melted tier needs a mill certificate showing 85% of the metal by weight was melted and poured in the United States, and Proclamation 11032 tells CBP to impose penalties "to the extent permitted by law" where an importer misstates US content. Goods under the metals proclamation entering a foreign trade zone must take privileged foreign status, so the zone freezes the classification but not the rate that applies on the withdrawal date, and manufacturing drawback under 19 U.S.C. 1313(a) and (b) remains available for 232 duties on goods that are re-exported after processing. A vendor quoting DDP has to carry all of this. A vendor quoting DAP hands it to your broker, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) now records which OEMs publish the HTS line and metal content on their submittals. ## Section 301, Twice: The China Stack and the July Tier **The July 24 action has four rate treatments, and the one that matters most for machinery buyers is an exemption: any article already subject to Section 232 pays none of it.** From [91 FR 47318](https://www.federalregister.gov/documents/2026/07/28/2026-15181/notice-of-actions-in-section-301-investigations-of-acts-policies-and-practices-of-various-economies) and the presidential memorandum at [91 FR 47717](https://www.federalregister.gov/documents/2026/07/28/2026-15274/actions-by-the-united-states-in-the-investigations-under-section-301-of-the-trade-act-of-1974-of-the-acts): a flat 10% additive on goods of Canada, Mexico, India, the UK and 13 others that have or have promised a forced-labor import ban; a cap for the EU and Taiwan under which column 1 plus Section 301 equals 10%, with zero added where the MFN rate is already 10% or more; the same cap at 12.5% for Japan, Korea and Switzerland; and a flat 12.5% additive on the other 38 economies including China and Vietnam. Goods that enter free under USMCA are exempt, as are the Annex I product exemptions and, in the notice's own words carried from the June proposal, "all articles and parts of articles subject to tariffs under Section 232" ([Brownstein, July 2026](https://www.bhfs.com/insight/trump-administration-imposes-tariffs-under-section-301-investigation-into-forced-labor-imports/) and [Honigman, July 24, 2026](https://www.honigman.com/alert-3462) both read the final Annex I the same way). That exemption inverts the schedule. A Japanese machining center (Annex III) pays 15% all-in and nothing under the new action. A Japanese CNC lathe (8458.11.00, no 232) pays MFN 4.4% plus 8.1% to reach the 12.5% cap. A German MV lineup pays 10% total. A German 2,500 kVA transformer pays 25% and is untouched by the 301 tier. Japan's 15% all-in deal of September 2025 ([EO 14345, 90 FR 43535](https://www.federalregister.gov/documents/2025/09/09/2025-17389/implementing-the-united-states-japan-agreement)) was an IEEPA instrument and died with it; the 15% number survives only inside Section 232 Annex I-C for mobile equipment. The "Taiwan 32%" that machine-tool buyers were quoted in spring 2025 was the April 2025 reciprocal rate and is gone; Taiwan now pays the EU treatment, 10% all-in, and the [Taiwan trade and security agreement](https://www.federalregister.gov/documents/2026/05/28/2026-10571/implementing-certain-tariff-related-elements-of-a-trade-and-security-agreement-between-the-american-institute) of May 28, 2026 caps Taiwan's 232 exposure on auto parts and lumber at 15%. What that does to a Taiwanese 5-axis landed in Ohio is in [Japan Is Eating the US Capex Boom](https://open-factory.com/articles/japan-eats-capex-boom/). China is the line where the secondary sources appeared to disagree, so we went back to the instruments. Honigman reads the new 12.5% as stacking on the existing 25% List 1 to 3 rates for an aggregate Section 301 of 37.5%. Holland & Knight's country table shows China at "13%" ([H&K, July 30, 2026](https://www.hklaw.com/en/insights/publications/2026/07/and-the-tariff-beat-goes-on)). The table rounds every 12.5% economy to 13%, Vietnam and Algeria included, so the 13% is the new action alone, not a claim that the 2018 lists were lifted. The primary record: 91 FR 47318 imposes 12.5% on products of China "except as provided in Annex I and Annex II, Part A," and contains no offset against the 2018 action; the 2018 action is still in the HTS at headings 9903.88.01 to 9903.88.04, "the duty provided in the applicable subheading plus 25%" ([HTS 2026 Rev. 18, chapter 99](https://hts.usitc.gov/)), and every one of our 22 HTS lines sits on List 1 or List 3 per U.S. note 20 (Lists 1 and 3 at [83 FR 28710](https://www.federalregister.gov/documents/2018/06/20/2018-13248/notice-of-action-and-request-for-public-comment-concerning-proposed-determination-of-action-pursuant-to) and [83 FR 47974](https://www.federalregister.gov/documents/2018/09/21/2018-20610/notice-of-modification-of-section-301-action-chinas-acts-policies-and-practices-related-to-technology)). Two instruments, neither cancelling the other, is a stack. We print China's non-232 lines at MFN plus 25% plus 12.5%, and we flag one gap per rule 48: we could not locate a CBP CSMS message confirming the combined 37.5% collection by September 9, 2026, so the calculator carries the stack as the default reading of the notice with that caveat on the cell. Nine origins for HTS 8537.10.91 and nine different bills: 0% for USMCA-qualifying Mexico and Canada, 9.8% for Korea (the 12.5% cap less the 2.7% column 1 rate KORUS already waives), 10% for the EU and Taiwan, 12.5% for Japan, 12.7% for India, 15.2% for Vietnam and 40.2% for China. Rockwell told analysts on August 4, 2026 that 100 basis points of its 250 basis point price target is tariff pass-through ([Q3 FY2026 call](https://www.investing.com/news/transcripts/earnings-call-transcript-rockwell-automation-beats-q3-2026-estimates-shares-fall-93CH-4834705)); on a panel built in Monterrey that pass-through is recovering a duty that does not exist, which is the argument of [Rockwell's 250 Basis Points](https://open-factory.com/articles/rockwell-250bps/) and the reason [Quote Check](https://open-factory.com/tools/quote-check/) asks for country of origin before it asks for price. ## The Schedule, HTS by HTS **Twenty-two HTS lines by nine origins gives 198 cells, and 108 of them are either 25% or 15% because the metal tariff does not read the country-of-origin field.** The table below is the condensed view; [data/tariff-schedule.csv](data/tariff-schedule.csv) carries every cell with the column 1 rate, the 232 tier and chapter 99 heading, the 2018 list, the July 2026 tier, the total, the caveat and the source string. Mexico, Canada and Korea are shown as FTA-qualifying, which removes the column 1 duty and, for USMCA, the July 2026 tier. A non-qualifying Mexican panel pays 12.7%, not 0%. [CSV: data/tariff-schedule-condensed.csv](data/tariff-schedule-condensed.csv) Read down the China column and the range is 37.5% (chillers, MFN free, no 232) to 52.6% (LV cable with connectors, 2.6% plus 25% plus 25%); read down Mexico and it is 0% on everything without metal and 25% on everything with it. Three rows include a KORUS footnote: for Annex III articles the 232 add-on is 15% minus the column 1 rate, so a Korean machining center that enters free under KORUS pays 10.8%, not 15%, and the same arithmetic gives 13.4% on a Korean or Mexican LPT. The notice does not say how CBP nets the 12.5% cap against a preferential entry, so the Korea column is the one place the calculator asks you to confirm with your broker. Two things are deliberately not in the table: antidumping and countervailing duties (Korean large power transformers have carried an AD order since 2012 and it stacks on all of this), and Section 338, whose 50% Canadian tariffs of August 19, 2026 cover dairy, alcohol and motor vehicles only ([91 FR 46663](https://www.federalregister.gov/documents/2026/07/23/2026-14997/imposing-additional-duties-to-offset-canadian-discrimination-against-the-commerce-of-the-united-states-with)), with the alcohol tranche suspended on August 18 ([91 FR 54789](https://www.federalregister.gov/documents/2026/08/24/2026-17294/temporary-suspension-of-additional-duties-to-offset-canadian-discrimination-against-the-commerce-of-the)). The robotics and industrial machinery investigation is the open item. Commerce opened it on September 2, 2025 with a scope that names CNC machining centers, turning and grinding machines, presses, tool changers, EDM and laser cutters ([90 FR 46382](https://www.federalregister.gov/documents/2025/09/26/2025-18749/notice-of-request-for-public-comments-on-section-232-national-security-investigation-of-imports-of)); comments closed October 17, 2025; the 270-day statutory clock ran out around May 30, 2026. A Federal Register search on September 9, 2026 returns no proclamation, and Business Sweden's August 20 tracker still lists it as pending ([Business Sweden](https://www.business-sweden.com/insights/articles/us-trade-policy-updates-courts-new-investigations-and-targeted-relief-reshape-the-tariff-landscape/)). Until it lands, a lathe, a press brake and a six-axis robot pay the July 2026 country tier and nothing else. The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) tracks the docket. ## What $80 Billion of Imports Now Pays **The nine HTS groups on this schedule covered $80.2 billion of US imports in 2025, and Mexico shipped $31.7 billion of it against China's $6.4 billion.** The figures are US-reported imports at customs value from [UN Comtrade](https://comtradeapi.un.org/), calendar 2025, cut at HS6 where the line allows us to drop power supplies and household refrigerators; the CSV is [data/import-value-2025.csv](data/import-value-2025.csv). Cable is $36.9 billion with $17.9 billion from Mexico; LV panels and PLCs are $20.2 billion with $8.6 billion from Mexico and $1.2 billion from China; liquid transformers are $7.7 billion, of which Mexico $2.2 billion, Korea $1.7 billion, the EU $1.3 billion and China $277 million. The line that pays the most is not the one with the highest rate: the 25% metal tier on $7.7 billion of liquid transformers and roughly the copper and aluminum share of $36.9 billion of cable dwarfs the 40.2% on $1.2 billion of Chinese panels. Machine tools are the mirror image, $5.1 billion with $2.1 billion from the EU and $1.2 billion from Japan, almost all of it outside 232 and inside the 10% and 12.5% caps. The transaction prices behind these customs values, by kVA and by section, are the job of the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), and the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) is collecting the machine-tool quotes. ## A $306k Machine From Five Countries **AMT's USMTO series put the average US machine-tool order at about $306,000 in the first half of 2026 ($3.44 billion over 11,243 units), so that is the invoice we landed.** Two machines, same price, two HTS lines: a machining center (8457.10.00, Annex III) and a horizontal CNC lathe (8458.11.00, no 232). The order data is from [American Machinist's June 2026 USMTO report](https://www.americanmachinist.com/news/news/55397089/machine-tool-demand-is-booming-in-2026-us-manufacturing-technology-orders-report-june-2026); the record half-year with fewer units is [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/). The machining center owes $45,900 from Japan, Germany or Taiwan, $33,000 from Korea under KORUS and $122,400 from China. The lathe owes $38,200 from Japan, $30,600 from Germany or Taiwan, $24,800 from Korea and $128,200 from China. Against a Haas VF-2 built in Oxnard that owes nothing, the Japanese machining center's duty is about 15% of the gap the buyer is weighing, which is the arithmetic in [What a Haas VF-2 Really Costs](https://open-factory.com/articles/haas-vf2-true-cost/). Duty is assessed on the customs value at entry, not the quote, so the freight, rigging and installation lines that [Tariff Bill of Materials](https://open-factory.com/articles/tariff-bom-electrical/) adds on top are outside the percentage. The calculator takes the invoice and the Incoterm and returns the entered value. ## What to Do Monday **Ask for the HTS line and the country of origin on every quote, because the schedule no longer maps to the brand's headquarters.** A Siemens transformer from Guanajuato pays the same 25% as one from Nuremberg; a Fanuc robot from Oshino pays 12.5% and a Fanuc robot from the Rochester Hills integrator pays nothing; a Mazak from Florence, Kentucky pays nothing and a Mazak from Inabe pays 15% now and 29.2% in 2028. Write the HTS number, the origin and the rate you were quoted against into the PO the way [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/) lays out, and run the line through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) before you sign, not after the broker's bill arrives. Second, split the metal from the box. A prefabricated e-house with the transformer inside is one entry line at the classification of the whole. The transformer alone is 8504.22.00 at 25% and the switchgear alone is 8537.20.00 at 0% from Mexico. Whether the skid saves money after the tariff is the subject of [Modular Is Not Cheaper](https://open-factory.com/articles/skids-vs-stick-built/), and the answer moved in April. Third, if you paid an IEEPA surcharge on a 2025 delivery, confirm today that your supplier has registered for ACH refunds with CBP: as of May 11, 2026, 1,880 approved consolidated refunds were sitting untransmitted because the importer of record had not supplied bank details, and CBP will not reissue a failed transmission or pay interest on the delay ([Western Overseas CAPE FAQ, May 28, 2026](https://www.westernoverseas.com/cape-system-faq-ieepa-refunds/)). Fourth, price the schedule into the lead time. A 2,500 kVA pad-mount ordered this month at the [Monitor's](https://open-factory.com/models/lead-time-monitor/) 40 to 65 weeks lands in 2027 under the 25% tier. A 50 MVA unit at 128 to 160 weeks lands in 2028 or 2029, after the Annex III tier is gone, at 25% plus MFN instead of 15%. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) now includes the entry-date tariff on every PO line for exactly that reason, and the [slot deposits](https://open-factory.com/articles/slot-economy/) buyers are paying to hold 2027 delivery positions are, among other things, a bet on which side of December 31, 2027 the truck arrives. Behind the paywall: the eight article classes whose duty rises **10 to 14.2 points** on January 1, 2028, the mechanism that lets Commerce pull the 15% tier from one country early, and the three clauses, tariff adjustment, refund allocation and entry timing, that we would not sign a capital-equipment PO without, with the language mapped to the [Tariff Exposure Calculator's](https://open-factory.com/tools/tariff-exposure-calculator/) citation fields. --- # SendCutSend: How a Bootstrapped Reno Shop Became a $1 Billion Parts Factory URL: https://open-factory.com/articles/sendcutsend/ Date: 2026-09-05 Procurement / Deep dive $110M from Sequoia, Paradigm and the Collisons, 30 million parts, 300,000 customers, 2 to 4 day sheet cutting, Reno, Paris and Arlington plants, 12 aluminum gauges, the $1B pledge, how to order SendCutSend has shipped more than [30 million laser-cut parts](https://sendcutsend.com/blog/sendcutsend-is-building-americas-anything-factory-with-1b-commitment-to-u-s-manufacturing/) to more than 300,000 customers from three plants in Reno, Nevada, Paris, Kentucky and Arlington, Texas, and until May 2026 it did so on a $750,000 laser financed by its founder, bank loans and credit cards. That month Sequoia, Paradigm and Patrick and John Collison put in $110 million at a $1.01 billion post-money valuation, the company's first institutional round eight years after Jim Belosic started cutting parts for his own boat projects. For a buyer the interesting number is not the valuation. It is the published schedule: 2 to 4 production days on a laser-cut aluminum bracket, plus or minus 0.005 inch, no minimum quantity, free shipping over $39. This piece covers how SendCutSend built that schedule, what the plants and headcount look like now, the processing-time stack a buyer can plan against, the quantity breaks, the four stock materials by gauge, the $1 billion pledge, and behind the wall, which parts to send and how to price them. SendCutSend's laser and CNC routing page says 1 to 3 days and its processing-times page says 2 to 4 for sheet cutting including waterjet. OSH Cut in Orem, Utah publishes 2 business days with a same-day option, Xometry 3 business days standard, Protolabs 3 days for cutting under 0.125 inch and 5 days standard. The 14 to 21 day bar is the comparison Belosic himself draws, [as The Hustle reported it](https://thehustle.co/originals/is-this-man-the-future-of-us-manufacturing): "48 hours vs. 2-3 weeks from China". The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) reads the ISM production-materials lead time at 84 days, so a fabricator that puts single-digit days in writing is doing something the rest of the supply chain does not. ## SendCutSend Started With One Laser and a Garage Jim Belosic had run software companies for most of his working life. He co-founded Pancake Laboratories in Reno, whose ShortStack product built social-media campaigns and landing pages, and he built parts for boats and vehicles in his garage on the side. When the projects outgrew hobby tools and no local shop wanted a one-piece order, he [sold his interest in ShortStack](https://www.nnbw.com/news/2026/jan/28/rise-grow-expand-sendcutsend-meteoric-rise-continues-with-100-million-revenue-milestone/) and bought an industrial fiber laser. The Hustle puts the first machine at $750,000, financed with his wife's approval. The company's own [about page](https://sendcutsend.com/about-us/) gives the founding question: "Why can't we create an online laser cutting service with the convenience of Amazon?" The answer was a web page that quotes a part from a CAD file in seconds. A customer uploads a .dxf, .dwg, .ai, .eps, .stp or .step file, or draws the part in the site's Parts Builder, picks a material and quantity, and the price appears. The company [describes the pricing model](https://sendcutsend.com/pricing/) as "material, quantity, part geometry, and total processing time, plus additional proprietary factors", with volume discounts up to 80%, savings from the second part, no minimum order and free US shipping at $39. Crunchbase News reported that Belosic [funded the company with savings, bank loans and credit cards](https://news.crunchbase.com/venture/interesting-startup-deals-custom-metal-group-travel-geothermal-energy/) for its first eight years, and TBPN Digest that it was [profitable before the round](https://www.tbpndigest.com/story/2026-05-19/sendcutsend-raises-110m-from-sequoia-and-paradigm-to-become-the-amazon-of-manufacturing) on about $1,500 a month of Google advertising. The order flow is the product. The site checks the uploaded geometry against design rules for the chosen material and thickness before it prices, so a hole too close to a bend line or a slot narrower than the kerf comes back as a warning rather than a scrapped part. Orders placed before noon Pacific start the same day, the plants cut seven days a week, and the company's own record, [as The Hustle reports it](https://thehustle.co/originals/is-this-man-the-future-of-us-manufacturing), is 32 minutes from upload to the shipping dock. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) scores vendors on whether lead time, price and capacity are published, and a web page that prices and schedules a part without a phone call is the top of that scale. **Cumulative revenue passed $100 million in October 2025 and $200 million by the time of the round, on a business that had never taken a venture dollar.** The $100 million milestone is from [Northern Nevada Business Weekly](https://www.nnbw.com/news/2026/jan/28/rise-grow-expand-sendcutsend-meteoric-rise-continues-with-100-million-revenue-milestone/), which counted 410 employees in January 2026, and the $200 million cumulative figure is [as Tech Funding News reported it](https://techfundingnews.com/sendcutsend-110m-sequoia-paradigm-stripe-cofounders/) in May. The one outside cheque before Sequoia was a [$6 million friends-and-family round](https://sendcutsend.com/blog/sendcutsend-is-building-americas-anything-factory-with-1b-commitment-to-u-s-manufacturing/) from Sandy Kory and Mark Sugarman. Ten million parts by March 2024, thirty million by May 2026, so the second twenty million took about two years and the first ten took six. The Hustle's figure for last year's output is 8 million pieces, which at that pace means the company is on track to ship more parts in 2026 than in its first five years combined. ## Three Plants and a Quarter-Million Square Feet The first plant outside Reno was Paris, Kentucky, and the state's own release gives the numbers: a [nearly 37,000 square foot building on Cleveland Drive](https://newkentuckyhome.ky.gov/Newsroom/NewsPage/20211201_SendCutSend), a $3.9 million investment, 75 jobs at an average of $21 an hour with benefits, opened in December 2021 with up to $800,000 in Kentucky Business Investment incentives over ten years. Belosic's reason was freight: "The Paris facility allows us to reach more than 50% of the United States in two days with our shipping carriers." Arlington, Texas followed with [50,000 square feet on 107th Street](https://sendcutsend.com/blog/prototype-to-production/), and by March 2024 the company had more than 180 employees, a new $1.3 million high-power fiber laser and two more CNC press brakes. Reno remains the center. Northern Nevada Business Weekly counts [130,000 square feet across three buildings on Longley Lane](https://www.nnbw.com/news/2026/jan/28/rise-grow-expand-sendcutsend-meteoric-rise-continues-with-100-million-revenue-milestone/) and nearly a quarter-million square feet in total, and in June 2026 the company was [hiring for a fourth Reno building](https://nevadabusiness.com/2026/06/sendcutsend-expands-its-reno-team-with-two-on-site-hiring-events/) at $24 to $42 an hour for fiber laser, CO2 laser, CNC, press brake, powder coating, anodizing and welding roles, with 100% employer-paid medical and one paid meal a shift. Headcount was "500+" across more than ten states. 130,000 square feet in Reno, 37,000 in Paris, 50,000 in Arlington, and five regions named for the next sites: Pennsylvania, Ohio, Indiana, Las Vegas and Atlanta, at [about $600,000 of deposit per location](https://www.tbpndigest.com/story/2026-05-19/sendcutsend-raises-110m-from-sequoia-and-paradigm-to-become-the-amazon-of-manufacturing) according to TBPN Digest's account of the raise. Machines are financed separately through JPMorgan, so the $110 million goes to software, people and buildings rather than lasers. 180 employees in March 2024, 410 in January 2026, more than 500 by June, with 200 to 300 more hires planned and the software team to be tripled. Belosic's hiring line, in the June release, is "We believe great people can learn great skills", and the Tech Funding News headline put it more bluntly: the company hires from Dollar Tree, donut shops and bars and trains them on lasers. The [electricians piece](https://open-factory.com/articles/electricians-wages-by-metro/) shows what that labor market looks like from the other side. ## How the Round Came Together The introduction came over X. Patrick Collison, who had invested personally, connected Belosic to Matt Huang at Paradigm and Andrew Reed at Sequoia, and Reed, in Belosic's telling to TBPN, "got on a plane, and within nine hours he was at our facility". Reed's public statement in the company release: "Reindustrializing the U.S. will require a hardware ecosystem to emerge, and SendCutSend is at the center of it." Huang called it [Paradigm's first manufacturing investment](https://news.crunchbase.com/venture/interesting-startup-deals-custom-metal-group-travel-geothermal-energy/) and pointed to "intense demand for rapid, on-demand sheet metal and custom parts" from robotics, defense, rocket and electric-vehicle companies. The customer list explains the interest. The company says [59% of the Fortune 500](https://sendcutsend.com/blog/sendcutsend-is-building-americas-anything-factory-with-1b-commitment-to-u-s-manufacturing/) have ordered, TBPN's account has 85% of the top defense primes and names Zipline and Nuro, and The Hustle reports nine of ten private rocket launch companies as customers, alongside high-school robotics teams paying $19.51 for a bracket. Belosic's own summary to The Hustle: "There has never been a better day to start making something in the US than today." The [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) shows the other end of that market, where public buyers pay for switchgear and gensets in bands, and the contrast is the point: SendCutSend's price is the same for the prime and the student. With the money came a pledge. The company committed to [$1 billion over five years](https://sendcutsend.com/blog/sendcutsend-is-building-americas-anything-factory-with-1b-commitment-to-u-s-manufacturing/) in US manufacturing jobs and domestically produced materials, with more than $250 million for facility expansion and new hubs. TBPN's account has 85% of the aluminum already domestic. For a buyer working through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/), that line is the one to check: a part cut in Reno from US-melted 5052 sits outside the Section 232 tiers the [September 2026 tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) lays out for imported articles. The company's own bio lists aerospace, defense, space flight, data centers, robotics, automation, transportation, marine, agriculture and distribution as the industries it serves, and TBPN's account has supply-chain disruptions touching 3 to 4% of customers, which is the exposure a domestic sheet supply is meant to hold down. ## The Processing-Time Stack a Buyer Can Plan Against What makes SendCutSend unusual among fabricators is not the laser. It is that the [processing-times page](https://sendcutsend.com/sendcutsend-processing-times/) states every operation in days and the days add. Sheet cutting is 2 to 4 days and CNC machining 2 to 5. Bending, tapping, countersinking and tumbling add 1 to 2 days each, hardware insertion and dimple forming 1 to 3, powder coating 3 to 5, anodizing 1 to 7, plating 7 to 10. Orders of 100 or more parts add 2 to 4 days, parts over 46 inches add 1 to 2, orders over 80 pounds add 1 to 2 for packaging, and orders needing more than 40 square feet of material add 2 to 5 depending on stock. The cutoff is noon Pacific, Monday to Friday, and the plants cut seven days a week. A flat laser-cut part ships in 2 to 4 days. Add a bend and a tapped hole and it is 4 to 8. Add powder coat and it is 7 to 13. The Hustle reports a standard 48-hour turnaround inside the company and a fastest-ever order of 32 minutes from upload to the shipping dock. That schedule is what the [Quote Check](https://open-factory.com/tools/quote-check/) tool compares a local fabrication quote against, and a local shop that quotes three weeks for the same bracket is quoting its queue, not its process. **Every operation has a tolerance and every tolerance is on the page: plus or minus 0.005 inch on laser and CNC routing, plus or minus 0.009 on waterjet, an automatic 0.010 inch edge break on machined parts, and a 0.002 to 0.005 inch per side powder-coat build.** The [sheet-cutting page](https://sendcutsend.com/services/sheet-cutting/) lists more than 175 materials from 0.015 to 2.25 inches, the [CNC machining page](https://sendcutsend.com/services/cnc-machining/) offers 3-axis, 3+2 and 5-axis work in 6061 from 1 x 1 x 1 to 7 x 7 x 12 inches, and the [powder-coating page](https://sendcutsend.com/services/powder-coating/) gives 15 colors on seven metals up to 30 x 36 inches flat. A buyer who has spent a week getting a tolerance in writing from a job shop, as the [RFQ guide](https://open-factory.com/articles/rfq-capital-equipment/) describes, will notice that here the tolerance is the default. ## Four Stock Materials, 38 Gauges, One Price Curve The aluminum menu is the one most machine shops will use. [6061-T6 comes in 12 thicknesses](https://sendcutsend.com/materials/6061-aluminum/) from 0.040 to 0.750 inch, [5052-H32 in 11](https://sendcutsend.com/materials/5052-aluminum/) from 0.040 to 0.500, and [7075-T6 in three](https://sendcutsend.com/materials/7075-aluminum/), 0.125, 0.190 and 0.250. [Mild steel runs 12 gauges](https://sendcutsend.com/materials/mild-steel/) from 0.030 cold-rolled 1008 to 0.500 hot-rolled A36. Instant pricing covers sheets to 30 x 44 inches and custom quotes reach 41 x 56 on 6061. Where the plate gets thicker than the laser, the [Nox Metals piece](https://open-factory.com/articles/nox-metals/) picks up: that company saws 6061 and 7075 plate from 0.25 to 24.5 inches in Detroit, and Belosic had its founder on his podcast. Twelve gauges of 6061, eleven of 5052, three of 7075, twelve of mild steel, all at plus or minus 0.005 inch and all at 2 to 4 production days. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) collects delivered price per part on cut sheet so that a buyer can see where an uploaded bracket lands against others of its size. The price curve is steep and published. The sheet-cutting page's example aluminum part is $9.90 each at quantity one and $3.70 at one hundred. The powder-coating example is $20.69 at one and $3.31 at one hundred. The pricing page shows a 1.5 x 3 inch part at $19.51 in hardboard, acrylic and 0.187 inch 5052 alike, and marks all of these as estimates that move with the material tape. $9.90 to $3.70 is a 63% drop, $20.69 to $3.31 is 84%, and the site's stated ceiling is "up to 80% off in bulk", which the powder-coat example exceeds. The arithmetic is in data/assumptions.csv. A buyer who batches a month of brackets into one upload of a hundred pays roughly a third of the single-piece price, and the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) shows how much of the remaining cost is metal tape rather than labor. ## What SendCutSend Means for a Job Shop The obvious reading is that SendCutSend competes with the local fabricator. The better reading is that it is the local fabricator's supplier. Belosic told The Hustle the model is a high-mix, low-volume job shop running 24 hours a day with "backup machines for backup machines", and a shop with one laser and one brake can send its overflow, its odd materials and its one-off prototypes to Reno and keep its own machines on the work that pays. We priced a cobot welding cell at $113,336 in [the robot cell piece](https://open-factory.com/articles/robot-61k-cell-150k/). The fixtures and guarding brackets for that cell are exactly the parts that come back from SendCutSend in a week, and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) worked example prices the rest of the cell. The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) tracks where the cells go. The disclosure is also unusual for its completeness. Turnaround, tolerance, material list, sheet size, quantity curve, cutoff hour and shipping threshold are all on public pages, which is more than most switchgear or chiller OEMs put in writing, as the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) shows. The one thing a buyer still has to ask for is a certificate: material certifications on cut parts are handled by request, and a shop supplying a defense prime should confirm the melt origin on the order, since TBPN's 85% domestic aluminum figure means 15% is not. ## What to Do Monday Upload the bracket you order most. Take the .dxf your local fabricator has quoted, put it through SendCutSend's instant quote in 5052 or 6061 at your usual quantity and at one hundred, and write down three numbers: the price per part at each break, the production days from the processing-times page, and the delivered date after the noon Pacific cutoff. Put the local quote beside it through [Quote Check](https://open-factory.com/tools/quote-check/) and the price per part into the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/). Then look at the calendar, not the price. The [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) turns the 2 to 4 day figure plus adders into a promised dock date you can hold a supplier to, and the [171-day capex piece](https://open-factory.com/articles/171-day-capex-lead-time/) explains why a fabricator that publishes days is worth more than one that beats a price by 5% and ships in a month. For thick plate above 0.75 inch, send the cut list to [Nox Metals](https://open-factory.com/articles/nox-metals/) instead, and for a machine to run the parts on, the [Haas VF-2 piece](https://open-factory.com/articles/haas-vf2-true-cost/) has the number. Behind the paywall: the parts SendCutSend fits and the ones it does not, a worked bracket priced at three quantities with the **7 to 13** day schedule attached, the file-preparation rules that keep an order out of the revision queue, and when a local fabricator still wins. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) fabrication rows are in data/turnaround-ladder.csv. --- # Modular Is Not Cheaper: A 50 MW Prefab Electrical Room, Line by Line URL: https://open-factory.com/articles/skids-vs-stick-built/ Date: 2026-09-04 Electrical / TCO Crown Texas, Powell and PCX purchase orders, 22% faster and 1 to 3% dearer, freight and crane at $450k a block, FAT scope, who owns the slot, three public projects compared Austin Energy paid Crown Texas [$725,183 for one prefabricated 16 by 60 foot control house](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX153644) in June 2023. The house was $489,270 of that. The truck was $12,500, the stairs shipped separately for $5,000, and the crew that unloaded and anchored it billed $58,275. Thirteen months later the utility signed a [change order for $144,350](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX199741) to store a house from October 2023 to October 2024, because the pad was not ready when the factory was. That purchase order is the whole argument in one record. A prefabricated electrical room moves work from the field to a factory, adds a truck, a crane and an integrator's margin, and moves the schedule risk onto whoever fails to have the concrete poured. **On a 50 MW datacenter electrical room, the [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lands the skid route at $3.51M per MW against $3.47M stick-built, 1% more, and 27 weeks faster: 95 weeks from power-block PO to energization against 122.** This piece prices the three routes line by line from public awards and purchase orders, shows why the switchgear lead time does not move, prices the freight and the crane from public rate cards, values the weeks, and ends with who should still buy stick-built and the three contract terms that make a skid deal work. Per MW of critical IT load in the mid case, stick-built is $3,469k, the skid route $3,510k and a hybrid $3,514k; equipment is $1,988k in every column, the stick-built field line is $398k against $219k for the skid, and the skid carries $360k of enclosure and $282k of integrator margin that the stick-built room does not. ## The Room We Are Pricing The unit is a 12.5 MW power block, four of them to 50 MW, each with two 15 kV metal-clad lineups, six 2,500 kVA transformers, 15 MW of UPS with batteries, the LV switchgear and the busway to the hall. Gensets, the utility substation and anything inside the white space are out of scope. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) includes the grid side and [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/) includes the tariff side. Turner & Townsend's 2025 index puts the whole electrical package at [54% of an air-cooled build and 48% of a liquid-cooled one](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends), on facility costs of $9.8 per watt in Phoenix and Columbus and $13.3 in Silicon Valley, so the room we price is roughly two thirds of a $5.3M per MW electrical line in a $9.8 per watt market. Three routes buy that room. Stick-built: the engineer of record specifies, the contractor bids the package, the switchgear and UPS are ordered on the contractor's paper through a distributor, and the gear is set in a room the general contractor finished a month earlier. The skid or e-house: an OEM or integrator ships the same gear pre-mounted in transport-rated enclosures with the HVAC, fire detection, lighting and internal bus already wired and tested, and the site sets modules on pads. Powell's 10-K describes the product class exactly: switchgear, breakers, MCCs and bus duct "typically integrated into custom-built enclosures which we also build," which it calls [power control room substations, custom-engineered modules or electrical houses](https://www.sec.gov/Archives/edgar/data/80420/000008042025000152/powl-20250930.htm). The hybrid keeps the MV switchgear and transformers stick-built and owner-furnished, and takes the UPS and LV distribution as a skid, which is 87% of the equipment value at our mid prices. **The three routes buy the same $1,988k per MW of equipment; the argument is entirely about the other $1,500k.** The equipment line is built from public prices with n and dates: a 15 kV lineup at [$916,409 supply-only from Eaton to the Corps of Engineers](https://www.usaspending.gov/award/CONT_AWD_W9127N26CA022_9700_-NONE-_-NONE-) (ten assemblies, May 2026), [$1,310,228 for a 12.47 kV Powell lineup](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX182451) with a 3 kA main and tie at Austin Energy (February 2024), and [$2,561,552 for Riverside's SUB-871](https://riversideca.legistar1.com/riversideca/attachments/37cb493a-5440-49a2-a51f-c84cbe1795ce.pdf) (June 2024, three bids); a 2,500 kVA pad-mount at [$98,000 to $104,000 list](https://store.maddox.com/products/2500-kva-padmount-transformer) (n=3, 2026); and a UPS line whose only public transaction is a Navy award of [$10,594,213 for five 750 kW modules with battery cabinets and 4,000 A switchgear modifications, installed](https://www.usaspending.gov/award/CONT_AWD_N6945025F1412_9700_N6945024D0018_9700), $2,825 per kW in September 2025, which we treat as the ceiling on an estimate of $900 to $1,800 per kW for equipment alone. Every input, its range and its source is in `data/assumptions.csv`. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds the award rows. ## The Lead Time Does Not Move The vendor pitch is speed, so start with the part of the schedule a factory cannot compress: its own. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes Terrapin's June 2026 range of [60 to 80 weeks for 15/27 kV metal-clad switchgear](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026), 40 to 60 weeks for modular UPS above 2,000 kVA and 40 to 65 weeks for pad-mount transformers, and Schneider's September 2026 guide at [45 to 50 weeks for LV switchgear and 24 to 40 weeks for busway](https://www.se.com/us/en/download/document/Monthly_Lead/). The same breakers, the same bus and the same GOES cores go into a skid as into a room; [why your switchgear costs 50% more than in 2021](https://open-factory.com/articles/switchgear-50-percent/) is the same story whether it ships in a crate or a box. Metal-clad MV switchgear at 60 to 80 weeks sits 15 to 30 weeks above the modular UPS at 40 to 60, the LV switchgear at 45 to 50 and the pad-mount transformer at 40 to 65, and 36 to 56 weeks above the pre-pandemic 24-week norm. The same page warns that [a project schedule "now needs to assume 72"](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026). What the skid changes is who owns the production slot. In a stick-built package the order lands on the contractor's paper at the end of design, and Global Data Center Hub's August 2026 note puts the penalty plainly: without a reserved slot, [nominal lead times extend by another eight to sixteen weeks](https://www.globaldatacenterhub.com/p/procurement-strategy-equipment-epc), while hyperscalers reserve slots with 10 to 30% deposits at land close. In a skid package the OEM already holds the slot for its own gear, and the customer's deposit sits on the OEM's balance sheet. Vertiv's current deferred revenue was [$3,633.7M at June 30, 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050609/vrt-20260630.htm), from $358.7M at the end of 2022; [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/) read that line as the accounting shadow of slot ownership, and [Vertiv Stopped Reporting Its Backlog](https://open-factory.com/articles/vertiv-backlog-estimate/) covers what the company no longer prints beside it. Deferred revenue was $358.7M in December 2022, $1,063.3M in December 2024, $1,814.7M in December 2025 and $3,633.7M in June 2026, a tenfold rise in fourteen quarters and 111% of that quarter's $3,274.3M of sales. Schneider wrote the slot into a contract in November 2025: a [$373M supply capacity agreement with Digital Realty](https://techinformed.com/schneider-electric-signs-2-27-billion-u-s-data-center-supply-deals-with-switch-and-digital-realty/) covering UPS, LV switchgear and prefabricated power skids on "a dedicated LVS production line," alongside a $1.9B Switch agreement for power modules and cooling, with deliveries phased across 2025 and 2026. The buyer is paying for a line, not a box. The slot comes with a clock the brochure does not print. Electronate's March 2026 note puts the OEM manufacturing freeze [4 to 8 weeks into the lead time](https://www.electronate.app/blog/switchgear-lead-times-2026-data-center-boom), after which a change to the spec is a change order and a new date, and puts the deposit that holds the slot at 10 to 20% of the switchgear value. In a skid package the freeze covers the enclosure too: the HVAC, the fire panel and the bus layout are frozen with the breakers, so the integration drawings have to be right in week 8, not in week 60 when the room drawings would have been. ## Ninety-Five Weeks Against 122 We built the schedule from public records rather than brochures. Riverside's own calendar for a 15 kV lineup ran [77 weeks from board award to scheduled delivery in 2024 and about 131 weeks in 2025](https://riversideca.legistar1.com/riversideca/attachments/99a5fd23-cb4f-4c0e-b009-d7390ecf7818.pdf), after 25 weeks of bidding each time; [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/) walks that calendar. Austin Energy allowed a contractor [160 working days, 32 weeks, to erect one prefabricated control house](https://austintexas.legistar1.com/austintexas/attachments/4194aa25-de7d-4390-ac7b-1579cd8bb9db.pdf) on a foundation the utility had already built, for $254,243 in 2017. Powell billed Austin [$249,401 for on-site installation, interconnection cables and site acceptance testing](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX92986) of one lineup in 2020. Those are the field weeks a skid is supposed to remove. Austin has since moved the houses themselves onto a factory contract: a five-year, [$10,000,000 not-to-exceed agreement with Crown Texas for prefabricated substation control houses](https://austintexas.legistar1.com/austintexas/attachments/4b615c93-9fe7-444f-98de-df4f35832988.pdf), issued September 2021, three offers, approved May 2022, with the 2023 purchase orders above drawn against it. Stick-built runs 10 weeks of drawings, 12 weeks in the contractor's queue, 70 weeks of switchgear fabrication, 20 weeks to set, cable and terminate in the finished room and 10 weeks of Level 1 to 5 commissioning, 122 in all. The skid runs the same 10 and 70, then 4 weeks of integration and factory acceptance testing, 3 to ship and set, 4 to tie modules together and 4 of site commissioning, 95 in all. **The 27 weeks the skid saves are 12 weeks of contractor queue and 15 weeks of field work; the factory saves nothing.** PCX, the Hubbell unit that builds skids in Knightdale, North Carolina, claims a [30% reduction in deployment time and "18 month buildouts" cut to "six or seven months"](https://info.pcxcorp.com/blog/reducing-data-center-deployment-time-by-30-with-modular-electrical-skids) in a June 2026 post; Siemens and Delta claimed [up to 50% faster time to market and up to 20% lower capex](https://www.engineering.com/delta-and-siemens-partner-on-modular-data-center-power-systems/) for their skid and e-house partnership in November 2025. Our 22% sits under both because we count the switchgear lead time in every route, and because we give stick-built the 8 to 16 week slot penalty rather than the 2 to 3 year datacenter-spec lead time that [Electronate reported in March 2026](https://www.electronate.app/blog/switchgear-lead-times-2026-data-center-boom) for buyers at the back of the queue. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) runs the same phases with your PO dates and the Monitor's current weeks. Stick-built spans 98 to 150 weeks with a mid of 122, the skid 79 to 120 with a mid of 95, and the hybrid 80 to 122 with a mid of 98. The hybrid captures 24 of the 27 weeks because an owner who holds the MV slot directly removes the contractor queue without buying the enclosure. ## The Cost Stack, Line by Line Here is the mid case per MW, in thousands, with the low and high cases in `data/cost-stack-per-mw.csv`. | Line | Stick-built | Prefab skid | Hybrid | What sets it | |---|---|---|---|---| | Equipment (MV, transformers, UPS, LV, busway) | $1,988k | $1,988k | $1,988k | Public awards and list prices, n and date in assumptions.csv | | Room or enclosure | $256k | $360k | $346k | 800 sq ft/MW at $320 inside the shell; 400 sq ft/MW at $900 fitted and transport-rated | | Procurement markup or integrator margin | $99k | $282k | $245k | 5% contractor markup on gear; 12% integrator margin on gear plus enclosure | | Field installation | $398k | $219k | $242k | 20% of equipment stick-built (Powell 24%, Riverside 47% as ceiling); 55% of that remains with skids | | Commissioning and FAT | $99k | $72k | $75k | 5% of equipment; L1 to L3 move to the factory, FAT travel added | | Freight, crane, permits | $40k | $36k | $37k | 2% of equipment; six modules at $54k crane and $22k haul each | | Owner engineering and PM | $274k | $234k | $261k | 10% (Riverside 11%); 8% with the OEM doing integration engineering | | Contingency at 10% | $315k | $319k | $319k | Riverside boards approved 10% in 2018 and 2024, 20% in 2025 | | Total | $3,469k | $3,510k | $3,514k | Skid +1.1%, hybrid +1.3% | *Source: [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/); CSV: `data/cost-stack-per-mw.csv`, inputs in `data/assumptions.csv`* Read the skid column against the stick column and the trade is visible. The skid removes $179k of field work and $27k of commissioning per MW and hands back $104k of extra enclosure and $183k of extra margin. The enclosure line is anchored to the only fitted prefabricated electrical house we can price from a public record, Crown Texas's [$478,484 and $489,270 for two 16 by 60 foot houses](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX153822) in June 2023, about $500 per square foot with the HVAC, panelboards, chargers and hydrogen detector inside; a UPS module has to reject 2 to 3% of 15 MW as heat and carry clean-agent suppression, so we carry $700 to $1,100 per square foot on a footprint PCX says is [up to 50% smaller than a stick-built layout](https://info.pcxcorp.com/blog/why-modular-electrical-skids-beat-traditional-buildouts-a-contractors-guide-to-minimizing-chaos-and-shortening-timelines). The margin line is anchored to what integration is worth to the people who buy integrators: Eaton paid [$1.45B for Fibrebond on about $378M of sales](https://www.sec.gov/Archives/edgar/data/1551182/000155118225000011/etn03312025exhibit99.htm) in April 2025, and Powell's gross margin on integrated work was [31% in the June 2026 quarter](https://www.sec.gov/Archives/edgar/data/80420/000008042026000107/powl-20260630.htm). Stick-built runs $2,259k to $6,344k per MW around a $3,469k mid, the skid $2,296k to $6,466k around $3,510k, and the hybrid $2,308k to $6,445k around $3,514k, so the skid premium is 1.1% in the mid case, 1.6% in the low case and 1.9% in the high case, and the input ranges swamp all three. That is the honest v0 answer to "is modular cheaper": no, and it is not meaningfully dearer either, at the labor package we used. The claim in the title stands because the claim in the brochures does not. Siemens and Delta's [up to 20% capex reduction](https://www.engineering.com/delta-and-siemens-partner-on-modular-data-center-power-systems/) requires the integrator to charge less than the field labor it replaces, and in our stack the integrator's 12% on $2.35M of gear and enclosure is $282k, against $179k of field labor removed. The public awards say the same thing from the other side. Riverside paid Crown Technical Systems [$2,509,063 for a 15 kV metal-clad switchgear enclosure](https://riversideca.legistar1.com/riversideca/attachments/ae7986e2-8ee3-4675-8f9c-b8dbb6c3040f.docx) in July 2023 against a $2.1M engineer's estimate, with Myers at $2,584,678 and Eaton, AZZ and RESA declining to bid; Leavenworth, Kansas opened bids in December 2025 for a [prefabricated enclosure housing a new 15 kV main switchgear at $1,499,117 and $2,012,750](https://www.leavenworthks.gov/sites/default/files/fileattachments/city_clerk/page/18736/wwtp_switchgear_replacement.pdf) against a $2,527,051 estimate; and Austin's fitted house came in at $725,183 all-in. The Riverside memo has the sentence every estimator should frame: ["The last time RPU procured a switchgear was in 2018. Due to the time gap, our most recent historical records for pricing this type of equipment are no longer relevant."](https://riversideca.legistar1.com/riversideca/attachments/ae7986e2-8ee3-4675-8f9c-b8dbb6c3040f.docx) The [Quote Check](https://open-factory.com/tools/quote-check/) tool exists so that the next buyer does not have to write it. ## Freight, Crane and the Storage Clause Freight and crane are the lines the brochure never prints and the purchase order always does. Crown Texas billed Austin [$12,500 of shipping per house, $5,000 to ship the stairs, and $49,652 to $58,275 for "unloading and anchoring"](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX153644) on two houses in June 2023 (n=2). The public rate cards bracket the crane: FEMA's schedule effective July 2025 puts a [125-ton crawler at $338.94 an hour and a 60-ton all-terrain at $285.76](https://www.fema.gov/assistance/public/schedule-equipment-rates); Caltrans's book for April to August 2026 has a [Link-Belt HTC 11100 at $335.52, a Grove TMS 875 at $212.55, a six-axle low-bed at $87.68 and a tractor over 60,000 lb at $101.56 an hour](https://dot.ca.gov/-/media/dot-media/programs/construction/documents/equipment-rental-rates-and-labor-surcharge/book_26v3.pdf), and notes that "pilot vehicles are extra." Texas prices the paperwork: a [$60 single-trip permit plus a $150 to $375 highway maintenance fee by weight tier and a $35 supervision fee over 200,000 lb](https://www.txdmv.gov/motor-carriers/oversize-overweight-permits/general-single-trip), with [one escort over 14 feet wide and two over 16](https://www.txdmv.gov/motor-carriers/oversize-overweight-permits/escort-and-equipment-requirements). A 16-foot house takes two escorts through every state it crosses. Six modules per block cost $324k to crane and anchor, $120k to haul 1,000 miles and $12k in permits and escorts, $456k in all, plus $150k of FAT travel and retest allowance, against $497k of freight and rigging for the same gear arriving loose. The one-year storage charge of $144,350 on one house is the line that is not in either base stack. **Logistics is a wash; the storage clause is the risk.** The prefab house that Austin stored for a year sat on the vendor's yard at $144,350 because the utility's pad was not ready, and Eaton's standard terms filed with Santa Clara say the same thing in contract language: ["Buyer shall reimburse Seller for all costs of storage and handling incurred by Seller after the date that Seller is prepared to make shipment."](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf) A skid arrives on one date as one object; a stick-built room absorbs late gear one crate at a time. The [RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) puts ninety free days in the PO, and Crown's own 30/40/30 milestones, [$217,555 at drawing submittals, $290,073 at material receipt, $217,555 at delivery and acceptance](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX153644), show where the vendor expects to be paid before your pad exists. ## What a Week Is Worth If the skid costs 1 to 3% more and delivers 27 weeks sooner, the question is what a week is worth, and that depends on whose money is waiting. A 50 MW build at Turner & Townsend's [$9.8 per watt](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends) is $490M; at the [4.68% ten-year Treasury of August 2026](https://fred.stlouisfed.org/graph/fredgraph.csv?id=GS10) that is $441,000 a week of carrying cost, and at the 6.32% Baa yield $596,000. Digital Realty signed Americas leases above 1 MW at [$186 per kW per month in the fourth quarter of 2025](https://www.sec.gov/Archives/edgar/data/1297996/000110465926010887/dlr-20260205xex99d1.htm), which is $2.15M a week of rent on 50 MW that is not energized. The [Liquidated Damages at 0.5% a Week](https://open-factory.com/articles/liquidated-damages/) piece built the same ledger; here we set it against the premium. Twenty-seven weeks of carrying cost alone is $11.9M, 6.9% of the $173.5M stick-built room; with rent it is $69.8M, 40%; the skid's 1.1% premium is $1.9M, so the break-even is under two weeks of rent or eight weeks of interest. That arithmetic is why the title is not "Buy Stick-Built." Modular is not cheaper, and it does not need to be: a route that costs 1 to 3% more and is 22% faster pays for itself in the first two weeks of rent it brings forward, provided the weeks are real. The weeks are real only if the OEM's slot is real, the FAT is witnessed, and the pad is poured, which is what the paid section is about. The [Datacenter Electrical Roadmap](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) tracks the capacity additions that decide whether the slot is real. ## Cheap Electricians Punish the Skid The one input that moves the premium most is the wage the skid displaces. Our field line is priced at the [IBEW Local 48 journey-level package of $95.79 an hour](https://www.ibew48.com/join/2026-wage-information/) in Portland, where the premium is 1.1%. Reprice the labor share of the field line at each metro's May 2025 median electrician wage times a 1.55 burden and the premium rises where labor is cheap: [Dallas at $28.37 and Atlanta at $28.20 an hour](https://open-factory.com/articles/electricians-wages-by-metro/) give a 3.3% premium, Chicago at $49.21 and Portland at $50.53 give 1.8 to 1.9%. The premium runs 3.3% in Dallas and Atlanta, 3.2% in Phoenix and Memphis, 3.1% in Columbus, 2.8% in Reno, 2.7% in Washington, 1.9% in Chicago and 1.8% in Portland, a 1.5-point spread that comes entirely from the wage, because the enclosure, the margin and the truck cost the same in every metro. **The skid is a labor arbitrage, and labor is cheapest in the metros where the datacenters are being built.** Datacenter construction is eating your electricians in every one of those markets, and the Local 48 rate is where the bidding ends up when three campuses compete for one hall's worth of journeymen. The [Electricians by Metro](https://open-factory.com/articles/electricians-wages-by-metro/) piece has the pipeline against the headcount. A buyer in Dallas who expects the skid to save money on labor will be disappointed by about $115k per MW. A buyer who expects it to save 27 weeks of a crew he cannot hire will not. ## Who Builds the Boxes The people who build prefabricated electrical rooms have all been bought, and the prices tell you what integration is worth. Hubbell paid [$112.8M for PCX in July 2022](https://www.sec.gov/Archives/edgar/data/48898/000162828023002875/hubb-20221231.htm), calling it "a leading designer and manufacturer of factory built modular power solutions for applications in the data center market." Quanta bought [Cupertino Electric on July 17, 2024](https://www.sec.gov/Archives/edgar/data/1050915/000105091526000006/pwr-20251231.htm) for its "engineering, procurement, project management, construction and modularization services." Eaton paid [$1.45B for Fibrebond, a builder of "pre-integrated modular power enclosures,"](https://www.sec.gov/Archives/edgar/data/1551182/000155118225000011/etn03312025exhibit99.htm) on about $378M of trailing sales, 3.8 times revenue, in April 2025. Vertiv bought [BMarko Structures on April 13, 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026024824/exhibit991vrt-q120264132026.htm) for structural fabrication behind its "prefabricated, and converged infrastructure." Four acquisitions in 45 months, two with disclosed prices ($112.8M and $1.45B), plus three capacity additions: Vertiv's [215,000 square foot Pelzer, South Carolina plant](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm) for "modular solutions, integrated power systems and other prefabricated infrastructure" in 2024, Powell's [$12.4M for 335,000 square feet of power control room laydown at Jacintoport, a 62% increase](https://www.sec.gov/Archives/edgar/data/80420/000008042025000152/powl-20250930.htm), and PCX's [159,600 square foot Knightdale plant](https://info.pcxcorp.com/news/pcx-announces-grand-opening-of-knightdale-manufacturing-facility) opened in October 2021 with 75 electricians. Powell is the cleanest public read on demand, because it is a pure switchgear and e-house builder that reports quarterly. Its backlog went from [$1.4B at September 30, 2025 to $2.4B at June 30, 2026](https://www.sec.gov/Archives/edgar/data/80420/000008042026000107/powl-20260630.htm), with about $800M of data center awards in the fiscal year including one mega order above $400M; the [Book-to-Bill League Table](https://open-factory.com/articles/book-to-bill-league/) has the ratio. Austin Energy's contracts with Powell trace the same curve on a utility's paper: [$22M over five years for indoor arc-resistant switchgear in 2018](https://webapi.legistar.com/v1/austintexas/matters/5478), $5M more in 2024, and a new [$56.9M five-year contract in September 2025](https://austintexas.legistar1.com/austintexas/attachments/79445311-53c5-484b-8917-28a6d2881c4b.pdf), with $16.5M in the first year. Vertiv's 10-K says the company has ["more than doubled its manufacturing capacity for switchgear, busbar and integrated power solutions" since late 2021](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm) and that most of its $15.0B backlog ships within 12 to 18 months. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades who prints what. ## What to Do Monday Price the room three ways before the design freezes, with the same equipment line in each column, and make the contractor and the integrator both quote against it. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) has the eight lines and the [Quote Check](https://open-factory.com/tools/quote-check/) tool bands the gear against the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). Decide the route on the schedule, not the price, because the price difference is 1 to 3% and the schedule difference is 27 weeks; run your PO dates through the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) and see which route energizes in the month your lease starts. If you go stick-built, buy the MV switchgear yourself under a frame agreement with a deposit, because the 12-week contractor queue is the cheapest schedule you will ever buy back. If you go skid, pour the pads before the factory finishes, because the storage clause is real and the invoice was [$144,350](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX199741) for one house. And put the copper index and the tariff instrument in either PO. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) give you the base dates. Behind the paywall: the four buyer profiles that should still build stick-built, with the labor rate and the integrator margin at which the skid loses, from a sensitivity grid that runs from **4.1% cheaper to 6.4% dearer**; the three contract terms that make a skid deal work, slot transfer, FAT witness and integration responsibility, with the clause language and the PCX FAT exclusions that show where integration responsibility leaks; and the named vendor capacity claims with dates, from the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/). --- # The Recommended Spares List Is Where the Margin Hides URL: https://open-factory.com/articles/recommended-spares-margin/ Date: 2026-08-22 Procurement / How-to Spares at 0.8% to 20% of the award, one utility's breaker price 2023 to 2025, PLC spares bid 40% apart, PM at $2,376 to $7,481 per genset, Modicon end-of-service dates, the four clauses The City of Ocala bought 15 ABB R-Mag breakers in November 2023 at [$40,197.50 each](https://ocala.legistar1.com/ocala/attachments/f2f42bf3-9473-4469-b94e-f21839f2e3b8.pdf), four more in November 2025 at [$44,499.75 each](https://ocala.legistar1.com/ocala/attachments/03e6b221-0ef3-4dd1-a4cf-361dfe0b9626.pdf), and it had no second quote either time because WESCO is ABB's [sole distributor for public power in peninsular Florida](https://ocala.legistar1.com/ocala/attachments/69a4d9a4-f8a3-4470-b6a6-58afcfc8333e.pdf). That is the recommended spares list working as designed: the equipment is bid, the parts that keep it running are not. **On the public awards we can read, the spares line is 0.8% of a transformer award, 2% of a genset, and 20% of a breaker buy, and the vendor's margin on it is whatever the channel allows.** This piece prices spares and service against the equipment they belong to from public records, shows where the bid spread collapses to 0.2% and where it opens to 40%, benchmarks genset PM at dollars per unit per year, walks one PLC family's published end-of-service dates, and ends with the four clauses that move spares pricing back onto the award. The [Open Factory Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds every public spares, breaker and PM award used below with a URL on each row. The master file is [data/spares-prices.csv](data/spares-prices.csv), 46 rows. Ocala's PTI Transformers proposal carried [$63,340 of required spares on an $8,070,340 award](https://ocala.legistar1.com/ocala/attachments/c4a725ca-21c9-4c88-bbeb-7ecd55712a22.pdf) (0.78%, two autotransformers, May 2023) and its Virginia Transformer bid tab carried [$48,900 on $5,726,568](https://ocala.legistar1.com/ocala/attachments/bd7ebabf-b20e-47b4-949f-a8fad5275f2c.pdf) (0.85%, three units, December 2023); the city's Cummins 400 kW genset came with a [five-year warranty line of $4,872.28 on $234,567.90](https://ocala.legistar1.com/ocala/attachments/f3785cdb-88ca-46db-b47e-24b8d7140212.pdf) (2.08%, November 2023); Mesa bought [five 72.5 kV SF6 breakers for $192,655 and kept one as a spare](https://legistar1.granicus.com/Mesa/attachments/5c756d9d-26dc-4ff0-845a-3d5f62de9c1a.docx) (20% by count and by dollars, May 2016); and one integral spindle rebuild at the low end of the [Hymson $8,000 to $25,000 range](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/) is 11% of a $70,995 Haas VF-2 list. Five observations, three buyers, not a distribution. The point is the range, not the median. ## Where the Margin Sits at the OEM The consensus is that service is the high-margin business and hardware is the commodity. Rockwell's own segment table says the opposite for the parts that matter here. In fiscal 2025 the [Software & Control segment](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm), which sells the controllers and I/O that end up on every recommended spares list, earned a 29.7% operating margin on $2,383M of sales; Intelligent Devices (drives, components) earned 18.0% on $3,756M; Lifecycle Services, the segment that does the field work, earned 14.5% on $2,203M, down from 16.1% in 2024. The spread is 29.7% against 14.5%, and Lifecycle Services still carried [53% of Rockwell's $2,878M backlog](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm) at September 30, 2025. The labor is the loss leader. The card in the technician's bag is the product. A spare 1756 processor sold to a plant that cannot run without it is booked in the 29.7% segment, which is why the spares list arrives after the PO and at list. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) tracks which OEMs publish list prices at all; for controllers, none of the majors do, and the public awards below are the closest thing to a price sheet. The channel prices the same way. Cummins' Sourcewell contract 092222-CMM publishes a [400 kW diesel at $78,349.43 and a 1,000 kW at $232,619.61](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10592/00004578/Additional%20Documents/Cummins%20092222-CMM%20Price%20Information.pdf), then discounts "Maintenance Agreements" and technician labor by 5% against 10% on startup. The equipment carries the cooperative discount. The maintenance agreement keeps the dealer's price, which is why the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool includes both. ## Breakers: One Buyer, Two Years, Plus 10.7% Medium-voltage breakers are the cleanest public series because utilities buy them in lots and post the quotes. Ocala's R-Mag line moved from [$40,197.50 in November 2023](https://ocala.legistar1.com/ocala/attachments/f2f42bf3-9473-4469-b94e-f21839f2e3b8.pdf) to [$40,981 in September and November 2024](https://ocala.legistar1.com/ocala/attachments/3f2eed0d-8442-42c3-a8b9-b20e9aa3d5cb.pdf) (+1.9%) to [$44,499.75 in November 2025](https://ocala.legistar1.com/ocala/attachments/03e6b221-0ef3-4dd1-a4cf-361dfe0b9626.pdf) for the H5 variant, 10.5% above the H5 price of $40,282.50 two years earlier. Every quote carried a 22 to 27 week lead time and the 2025 quote added that prices may be adjusted for "duties, freight, tariffs, supplier pricing, surcharges, or exchange rate fluctuations." A spare breaker is a lineup item with a switchgear lead time attached. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists 15 kV metal-clad at 52 to 80 weeks, so 27 weeks for a loose breaker is the fast lane. Riverside Public Utilities paid [$470,877.02 for 15 vacuum breakers in May 2021](https://riversideca.legistar1.com/riversideca/attachments/520d79bf-26c0-4956-81cc-743b3f06feed.docx), $31,392 each, with four bids inside a 7.6% spread and installation by its own crews at $17,122.98; the same utility's [24 69 kV breakers in 2019 came to $2,170,143](https://riversideca.legistar1.com/riversideca/attachments/0304104b-9e31-4e3c-9fc9-33f97ecf0ab9.docx), $90,423 each, and the board memo says the bid evaluation "took into consideration the proposal base amount, shipping and storage costs, optional item prices, and spare parts." Riverside's program exists because its older breakers "contain problematic or unavailable parts, and therefore, can no longer be cost-effectively maintained"; the memo puts service life at 30 to 40 years and the parts ran out first. Mesa paid for the alternative: two 20-year-old 69 kV breakers failed at Fraser Substation in April 2019, 16,800 customers lost 48 MW of distribution capacity, and the city ratified [$292,224 of emergency purchases](https://legistar1.granicus.com/Mesa/attachments/39db8758-15cb-4eff-b141-91526fe11342.docx) including $121,621 of replacement breakers, three years after it had bought a fifth SF6 breaker as a spare for $38,531. For the low-voltage side, the OEM list is public. Schneider's US product pages carry list prices, and the recommended spares for a distribution switchboard read straight off them. | Reference | Description | List price, Sep 2026 | |---|---|---| | QO120 | QO 20 A 1-pole plug-in miniature breaker | $29.10 | | HDL36100 | PowerPacT H 100 A 3-pole 600 V, lugs, thermal-magnetic | $1,328 | | HGA36100 | PowerPacT H 100 A 3-pole 600 V, I-Line, thermal-magnetic | $1,772 | | JGA36250 | PowerPacT J 250 A 3-pole 600 V, I-Line, thermal-magnetic | $5,252 | | LGA36400U31X | PowerPacT L 400 A 3-pole 600 V, I-Line, MicroLogic 3.3 | $5,376 | *Source: [Schneider Electric US product pages](https://www.se.com/us/en/product/JGA36250/), list price shown per reference. CSV: [data/10-schneider-list-prices.csv](data/10-schneider-list-prices.csv)* The same 100 A H-frame is [$1,328 with lugs](https://www.se.com/us/en/product/HDL36100/) and [$1,772 as an I-Line plug-in](https://www.se.com/us/en/product/HGA36100/), a 33% premium for the bus connection, and the [400 A L-frame with a MicroLogic 3.3 electronic trip is $5,376](https://www.se.com/us/en/product/LGA36400U31X/), $124 more than the [250 A thermal-magnetic J-frame](https://www.se.com/us/en/product/JGA36250/). These are list prices, n=1 each, and a switchboard vendor's spares quote will show them at list or a small discount. The [Quote Check](https://open-factory.com/tools/quote-check/) tool returns the list for the reference so the discount is a number. What list does not tell you is what the retrofit costs when the trip unit family goes obsolete: the National Institutes of Health awarded [$3,358,750 in April 2026](https://www.usaspending.gov/award/CONT_AWD_75N98026C00030_7529_-NONE-_-NONE-) to replace "existing, obsolete GE circuit breaker trip units" in Building 10 with ABB Entelliguard TU units, sole source. ## PLC Cards: 0.2% Apart or 40% Apart The bid spread on a spares list is the margin signal. Where only the OEM channel can supply the part, the two quotes land on top of each other. Where distributors compete, they do not. Riverside's 2021 purchase of automatic voltage regulator spares for its two 2011 gas turbines drew [two responsive quotes, $109,497 and $109,716](https://riversideca.legistar1.com/riversideca/attachments/365a5c83-71f3-4123-853d-2e84170628b7.docx), 0.2% apart, with GE Packaged Power, GEXpro and a third vendor non-responsive; Mesa's 2017 bid for Schneider Modicon PLC replacement and spares at its water plants drew [$51,596.40 and $72,394](https://legistar1.granicus.com/Mesa/attachments/d8a540df-8369-4a3e-8790-6ce550482a39.docx), 40% apart, from two electrical distributors. Between them sit the breaker lots at 5.6% (n=4) and 7.6% (n=4), Mesa's six SF6 bids at 24%, and Phoenix's two bids for the identical annual genset service scope at [$153,215 and $202,955](https://webapi.legistar.com/v1/phoenix/matters/22406/texts/22509), 32%. A spares list that draws quotes 0.2% apart is not being competed; it is being confirmed. The [OEM Parts Network quote](https://riversideca.legistar1.com/riversideca/attachments/a0f47032-2a5e-4991-96ba-01be905e0824.pdf) itemizes an IS200ACIEH1B exciter card at $34,750, a power supply at $16,975, two I/O boards at $14,997 and $11,425, and three more cards between $9,375 and $11,250, all "subject to prior sale or while supplies last" with two-week delivery; the city's memo says the parts were bought so a November 2021 outage would not wait on them. That is a ten-year-old machine buying its controller spares from the secondary market at prices the OEM channel matched to the dollar. The federal record shows the endgame: the Corps of Engineers paid [$31,346.75 in August 2025](https://www.usaspending.gov/award/CONT_AWD_W912QR25PA022_9700_-NONE-_-NONE-) for ControlLogix 5580 processors "to replace end of life Rockwell Automation ControlLogix PLC processors" at McAlpine Locks, the Navy paid Rockwell [$329,840.16 sole source for ControlLogix EtherNet/IP modules](https://www.usaspending.gov/award/CONT_AWD_N6449825F4127_9700_N6449824G4001_9700) in March 2025, and NASA paid [$290,950 for 1756-L83E controllers](https://www.usaspending.gov/award/CONT_AWD_80KSC021F0045_8000_NNG15SD42B_8000) in 2021. None of the three awards publishes a quantity, which is the [Equipment Price Benchmark's](https://open-factory.com/models/equipment-price-benchmark/) reason to exist. The ControlLogix lead-time cycle that put those buyers on allocation in 2022 and 2023 is in [From 36 Weeks to Stock](https://open-factory.com/articles/controllogix-36-weeks-to-stock/). ## Obsolescence Has a Published Calendar Every major PLC vendor publishes a lifecycle status per catalog number. Schneider's US product pages print two dates on a discontinued reference, "Discontinued on" and "End-of-service on", and the gap between them is the window in which the OEM will still repair or supply the part before the aftermarket is all that is left. The [Modicon Quantum 140CPU65150 processor](https://www.se.com/us/en/product/140CPU65150/) was discontinued December 31, 2018 with end-of-service December 18, 2025 (substitute: an M580 BMEP582040); the [Quantum Ethernet module 140NOE77101](https://www.se.com/us/en/product/140NOE77101/) shares the 2018 discontinuation but runs to December 31, 2027; the [32-point DC input 140DDI35300](https://www.se.com/us/en/product/140DDI35300/) and the [power supply 140CPS11420](https://www.se.com/us/en/product/140CPS11420/) were discontinued December 31, 2022 and January 15, 2023 with end-of-service December 31, 2030; the [Premium CPU TSXP572634M](https://www.se.com/us/en/product/TSXP572634M/) got the shortest window, December 2, 2020 to November 19, 2024. So a Quantum rack installed in 2005 lost its CPU from the catalog after 13 years and from OEM service after 20, while its I/O stays serviceable until 2030. Mesa's 2017 memo describes exactly that rack: PLCs "15 plus years old" with "many failures in the past several months" that "cannot be reconfigured for future system enhancements and will be disposed of after cannibalizing any useable parts." The lesson is not that Schneider is slow to retire product. A seven-year service tail on a discontinued CPU is generous. The lesson is that the calendar is public and almost no purchase order references it. A spares clause that says "vendor shall give 12 months' notice of discontinuation and offer a last-time buy at contract pricing" costs nothing to write in 2026 and is worth the difference between the $31,346.75 the Corps paid for planned replacements and whatever Mesa's water plant paid in downtime while its filters could not backwash. The [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) tool takes the end-of-service date and the current lead time for the substitute and returns the last quarter in which the migration can still be planned rather than forced. ## PM Contracts at Dollars per Genset per Year Preventive maintenance is the recurring version of the spares list, and cities publish it with unit counts. Riverside's Fleet Division has maintained the same fleet of 45 stationary and portable generators through three contractors since 2018, which makes it the best public series we have. Collicutt Energy Services' 2018 agreement priced the base PM at [$106,904.49 a year for 45 generators](https://riversideca.legistar1.com/riversideca/attachments/49b13dba-4d31-4065-aa05-81c9126b882d.pdf), $2,376 each, plus up to $50,000 a year of on-call repairs; Bay City Electric Works' October 2020 proposal, adopted through a Sourcewell Kohler contract, priced [44 units at $143,902.65 a year](https://riversideca.legistar1.com/riversideca/attachments/69a6f1ef-0bd8-4316-92e2-3c6a5da981ef.pdf), $3,271 each; Duthie Electric's 2024 agreement, chosen from [eight proposals](https://riversideca.legistar1.com/riversideca/attachments/8bad0b0b-ac36-43f2-884d-86ee23bd5f91.pdf), priced the base at [$140,764.24 for 45 units](https://riversideca.legistar1.com/riversideca/attachments/a992e3cb-6af1-4de0-b118-1c2dec9e6f2c.docx), $3,128 each, plus $55,000 on-call. Phoenix's estimated [$217,920 a year](https://webapi.legistar.com/v1/phoenix/matters/22406/texts/22509) across the "more than 80 generators" its [2021 memo](https://webapi.legistar.com/v1/phoenix/matters/18777/texts/18880) counts is under $2,724 per unit; the Maritime Administration's [$113,782.07 over two years for 11 campus generators](https://www.usaspending.gov/award/CONT_AWD_693JF719C000020_6938_-NONE-_-NONE-) (nine offers, 2019) is $5,172 per unit per year; and Mesa's Facilities contract, which bundles interval and major repairs, rentals and refueling, runs [$404,000 a year for about 54 generators](https://legistar1.granicus.com/Mesa/attachments/3014fb0d-293d-4e6a-a620-809ac59412fa.docx), $7,481 each, with an escalator of up to 5% or CPI. Scope explains the 3x range: PM alone is $2,400 to $3,300 per genset per year on the Riverside series; add repairs and fuel and it is $7,500. The markup is on the same page as the base price and nobody reads it. Collicutt's worksheet lists "MATERIALS/PARTS MARKUP 35%", fuel at 35% over cost, labor at $120 an hour and $180 on call; Bay City's proposal lists labor at $169 to $278 and an automatic renewal "with a 2% increase per year to contract capping out at 10%." The base PM is competed; the parts pass-through is a percentage nobody bids against. It is the same mechanism our [2,439 public change orders](https://open-factory.com/articles/change-orders-212-invoices/) showed on construction contracts, and the same reason [Grainger's list prices](https://open-factory.com/articles/grainger-confession/) could sit at multiples of transaction prices for a decade. Collicutt's $106,904 was $36,540 of quarterly inspections, $41,079 of annual PM and coolant service, $25,296 of load testing and $3,989 of oil and fuel sampling, with transfer switch inspection "included"; Bay City's $143,903 was $69,258 of quarterly-plus-major service, $34,669 of load bank testing, $16,163 of ATS service, $16,253 of coolant service and $7,560 of fuel analysis. The 35% jump between them is mostly scope that moved from "included" to a line, which is the honest way to price it and the expensive way to buy it. Both contractors were dropped at the end of their base terms: the city's [2021 memo](https://riversideca.legistar1.com/riversideca/attachments/30f58d3d-0bf7-4c38-994f-581b42014a57.docx) says Collicutt's services "have been, and continue to be, unsatisfactory" and its [2024 memo](https://riversideca.legistar1.com/riversideca/attachments/a992e3cb-6af1-4de0-b118-1c2dec9e6f2c.docx) says Bay City's were "unsatisfactory regarding responsiveness, quality of workmanship, and repair estimate"; those are the city's judgments in its own council record, not measurements, and both firms bid again in 2024 and ranked fourth and seventh of eight. At utility scale one spare dwarfs all of this: Silicon Valley Power told its council in January 2025 that a 2022 overhaul and spare rotor exchange on one turbine generator cost ["over $1.4 million in unanticipated costs"](https://webapi.legistar.com/v1/santaclara/matters/24551/texts/14245), and with gensets on allocation through 2028 per [Cummins and Caterpillar Are Sold Out](https://open-factory.com/articles/gensets-sold-out-2028/), the spare rotor gets bought. ## Machining Centers: The Spindle Is the Spares List For a VMC the recommended spares list is short and the big item is the spindle. Against the [$70,995 Haas VF-2 list](https://open-factory.com/articles/haas-vf2-true-cost/) we priced in March, an integral spindle rebuild is [$8,000 to $25,000](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/) (11% to 35% of list), a ball screw regrind is $4,000 to $9,000 per axis, and a PM contract on one machining center is [$1,500 to $4,000 a year](https://www.in-housecnc.com/cnc-machine-repair-cost/) (2.1% to 5.6%). Hymson's labor ladder is $95 to $140 an hour for an independent technician against $150 to $225 for the OEM distributor. Those are vendor benchmarks, not awards. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has zero CNC awards with spares itemized, so the machining rows below lean on rebuild ranges. DMG Mori's first-half 2026 results give the OEM's view of the same number: [JPY 73.7bn of MRO, spare and engineering orders against JPY 335.2bn total](https://www.dmgmori.co.jp/corporate/en/ir/ir_library/pdf/fy2026_2shihanki_ex_e.pdf), 22% of the order book from machines already on the floor. ## When to Buy Spares With the Machine and When to Wait The rule that falls out of the awards is about channel, not equipment class. Buy with the machine whatever only the OEM channel can supply, because the price will not fall and the lead time will not shorten; defer whatever a distributor can bid, because the spread is 24% to 40% and the machine is already running. Three of the seven rows go on the award (trip units and relays, one spare breaker per frame at 15 kV and up, one PLC processor set) and three wait for a distributor bid (filters and coolant kits, spindles and ball screws on condition, standby-diesel controller boards); transformer bushings and relays go with the unit because the [$12,370 optional HV bushing](https://ocala.legistar1.com/ocala/attachments/c4a725ca-21c9-4c88-bbeb-7ecd55712a22.pdf) on Ocala's proposal costs more than that once the factory closes the order. What to do Monday: pull the recommended spares list out of the vendor's post-award quote and put it into the RFQ as a priced schedule, with the [Quote Check](https://open-factory.com/tools/quote-check/) list price next to each Schneider or Eaton reference and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) unit price next to each breaker; ask for the lifecycle status of every catalog number on the list, in writing, before the PO; and split the list into the sole-channel items you buy at award and the competitive items you send to two distributors in year two. Our [RFQ for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) has the schedule format. Our [switchgear quote anatomy](https://open-factory.com/articles/switchgear-quote-anatomy/) shows where the spares line hides among the other ten. The one clause that saves the most is spares at award pricing for five years, and Ocala's own quotes show what it is worth: **10.7% on a breaker in 24 months**, before tariffs, on a line the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) now flags for Section 232 adjustment. Behind the paywall: the four clauses, with model language a purchasing manager can paste, and the vendor replies to expect on each. --- # 800 VDC Will Strand Your White-Space Frame Agreement URL: https://open-factory.com/articles/800vdc-strands-frame-agreements/ Date: 2026-08-21 Electrical / Take NVIDIA 800 VDC shipping from Vertiv H2 2026, what changes SKU in 2027 (UPS, busway, PDU) and what does not, ABB's zero orders, how to write the change clause NVIDIA put a ship date on 800 VDC on August 11, 2026: the MGX power rack arrives [in the second half of 2026](https://blogs.nvidia.com/blog/800-vdc-power-architecture-ai-factory/). A 2 MW-per-row power center with an 800 VDC busway follows in 2027, and a DC power block that converts grid power in one step is the design for new facilities after that. Vertiv said the same fifteen months earlier: rectifiers, DC busway, rack DC-DC and DC backup, [scheduled for release in the second half of 2026](https://www.vertiv.com/en-us/about/news-and-events/corporate-news/vertiv-accelerates-ai-infrastructure-evolution-in-alignment-with-nvidia-800-vdc-power-architecture-announcement/), ahead of Kyber and Rubin Ultra. **A five-year white-space frame agreement signed this year on 415 V UPS, AC busway and AC PDUs puts 40 to 60% of its dollars into years when those are no longer the SKU the racks want.** What follows: the rack power roadmap by generation, which electrical SKUs change under 800 VDC and which do not, who has put a date on product, a 250 MW frame agreement laid against the transition year, and the change clause to write into the next one. Five generations on one axis: GB200 NVL72 at 120 to 132 kW, GB300 NVL72 at 142 kW, Vera Rubin NVL72 at 180 to 230 kW, Rubin Ultra NVL576 "Kyber" at 570 to 660 kW in the second half of 2027, and a 1 MW design target, against the 350 to 400 kW line where 415 VAC delivery runs out of copper and connector. ## The Rack Power Roadmap, Generation by Generation The GB300 NVL72 rack draws [up to 142 kW from eight 33 kW power shelves](https://docs.nvidia.com/enterprise-reference-architectures/nvl72-ai-factory/latest/components.html), fed at 415 or 480 VAC and converted to 54 VDC inside the rack. Vera Rubin NVL72, which Jensen Huang said was [in production on July 15, 2026](https://finance.yahoo.com/technology/ai/articles/huang-says-vera-rubin-production-140456434.html), runs at 180 to 220 kW on SemiAnalysis's count and 190 to 230 kW in the trade press; NVIDIA has not published the number. Kyber, the NVL576 rack for Rubin Ultra, is [specified near 600 kW for the second half of 2027](https://introl.com/blog/nvidia-vera-rubin-gpu-600kw-racks-2027), with NVIDIA's own generation table at 570 kW [as reported by StorageReview](https://www.storagereview.com/news/nvidia-moves-800-vdc-power-architecture-from-concept-to-production-just-dont-turn-off-ac-power-yet) and SemiAnalysis's Kyber Ultra estimate at 660 kW. The physics case is NVIDIA's from May 2025: 54 VDC in-rack distribution [hits its limits above 200 kW](https://developer.nvidia.com/blog/nvidia-800-v-hvdc-architecture-will-power-the-next-generation-of-ai-factories/), a 1 MW rack on 54 V needs up to 200 kg of copper busbar, and Kyber's power shelves would eat 64U. At 800 VDC the same conductor carries 85% more power and copper falls 45%, the gauge change the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) will register first. Vertiv's July 2026 engineering note puts the AC breakdown [between 350 and 400 kW per rack](https://www.vertiv.com/en-us/insights/articles/blog-posts/from-rack-to-data-hall-the-practical-path-to-800-vdc/) and hands the 8 to 16 rack units of power conversion back to compute. **Vera Rubin does not need 800 VDC; Kyber does.** A 200 kW rack runs on three-phase AC and a 415 V UPS without touching a limit, which is why SemiAnalysis calls the 2026 to 2027 sidecar phase [voluntary future-proofing](https://newsletter.semianalysis.com/p/inside-the-800vdc-revolution-part) led by Google and Meta. The 600 kW rack in the second half of 2027 is where the AC busway and the in-rack shelves stop being an option, and every vendor roadmap below is dated off that rack. ## What Changes SKU and What Does Not The buyer's question is which line items on the frame agreement survive. The answer splits along the gray-space and white-space line, and the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lists each line at $/MW with its lead time from the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). Eight SKUs across four phases: three gray-space rows show "stays" in all four columns; the central UPS goes from "stays" to "optional" to "exits at hyperscalers"; AC busway and AC PDUs both change in the 2028 to 2029 column; the in-rack PSU shelves leave the rack in the first phase. The [CSV is here](data/02-sku-changes.csv). The sidecar phase deletes nothing upstream. NVIDIA's August 2026 wording is that the power rack slots into existing AC infrastructure with [no changes to the building's electrical system](https://blogs.nvidia.com/blog/800-vdc-power-architecture-ai-factory/); Vertiv's is that customers "retain much of their upstream AC infrastructure." SemiAnalysis puts it in a buyer's words, same transformers, same UPS, same switchgear, same ATS, plus a row-level power rack worth [$400,000 to $500,000 per MW of added content](https://newsletter.semianalysis.com/p/inside-the-800vdc-revolution-part). Rob Bunger at Schneider told The Next Platform in May 2026 that sidecars will deliver [600 kW to 1 MW each](https://www.nextplatform.com/compute/2026/05/27/not-so-fast-on-that-charge-for-800-volt-datacenter-power/5246760), that centralized AC-to-800 V converters come in 2 to 5 MW chunks, and that by 2030 "maybe 10 percent of the new AI nodes" will require 800 VDC. The next two phases turn the white space over. When compute trays take 800 VDC natively (SemiAnalysis's 2027 to 2028 window, Vertiv's "2028 to 2029 timeframe"), the battery moves into the power rack and the central 415 V UPS becomes optional; SemiAnalysis carries [$1.2 million per MW of centralized UPS content exiting](https://newsletter.semianalysis.com/p/inside-the-800vdc-revolution-part) at hyperscalers, with colo operators keeping it for redundancy in the medium term. NVIDIA's row power center brings the 800 VDC overhead busway in 2027, and by the facility phase the AC busway with tap-offs gives way to a feeder-only DC busway, because an 800 V arc has no zero crossing to put itself out. Littelfuse's January 2026 investor slides already draw the new white space as "sidecar (1 MW+), PSU shelf, HVDC bus (±400/800 VDC), [DC PDU](https://www.sec.gov/Archives/edgar/data/889331/000162828026003710/q42025earningsslide.htm)." **The MV switchgear, the transformer and the genset on your frame agreement are safe through 2029. The UPS, busway and PDU lines are not.** The utility feed stays AC, so the 15 kV lineup we priced in [Why Your Switchgear Costs 50% More Than in 2021](https://open-factory.com/articles/switchgear-50-percent/) is unchanged, and the LV transformer stays until solid-state transformers ship at scale, which SemiAnalysis does not expect before early 2029. Eaton's CEO told analysts on July 31, 2026 that [$3.4 million per MW](https://finance.yahoo.com/markets/stocks/articles/eaton-corp-plc-etn-q2-230128644.html) "remains the right number" for its electrical content and that 800 VDC improves efficiency by 5%. The content does not shrink; it moves from the UPS room to the row. ## Who Has Put a Date on It Announcements are not ship dates. We read every 800 VDC release from the seven vendors plus NVIDIA and scored them the way the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) scores any product claim: is there a quarter or half-year in the text, or not. Eleven rows: NVIDIA dates three products (H2 2026, 2027, 2029), Vertiv dates its portfolio (H2 2026, pilots, ramp through 2027), Navitas dates its silicon (2H26 PSU ramp, native 800 V trays in 2H27), Schneider gives guidance (small deployments end 2026, centralized 2029), Delta's 2Q27 ramp is a Digitimes headline, and Flex, Eaton and ABB have announced products or architectures with no ship date. The [CSV is here](data/03-vendor-availability.csv). Vertiv is the most specific. Its May 19, 2025 release scheduled the portfolio for [the second half of 2026](https://www.vertiv.com/en-us/about/news-and-events/corporate-news/vertiv-accelerates-ai-infrastructure-evolution-in-alignment-with-nvidia-800-vdc-power-architecture-announcement/); its October 13, 2025 release repeated the date and said the reference architecture is [basis of design on several large AI factory projects](https://www.vertiv.com/en-us/about/news-and-events/corporate-news/from-vision-to-readiness-vertiv-collaborates-with-nvidia-to-advance-800-vdc-platform-designs-to-power-the-next-generation-of-ai-factories/); its July 2026 note names the PowerDirect 5000 sidecar at 400 to 900 kW and sequences commercialization in the second half of 2026, then pilots, then a ramp through 2027. Vertiv's $3.6 billion of customer deposits at June 30, 2026, sized in [Vertiv Stopped Reporting Its Backlog](https://open-factory.com/articles/vertiv-backlog-estimate/), is the cash behind those engagements, and the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) will show whether the date held when the FY2026 10-K prints backlog. ABB is the other end of the table. Its October 13, 2025 release says ABB [will support NVIDIA's planned 800 VDC architecture](https://uk.finance.yahoo.com/news/abb-develop-next-generation-ai-150100374.html) and names the HiPerGuard MV UPS and SACE Infinitus solid-state breaker; its June 1, 2026 release aligns "current portfolio and [future planned assets](https://finance.yahoo.com/sectors/technology/articles/abb-expands-collaboration-nvidia-integration-154600664.html)" with the NVIDIA DSX blueprint; its Q2 2026 results release schedules a [webcast on its direct-current portfolio for September 24, 2026](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf). None of the three names an 800 VDC product ship date or order. That is the measurement, the standard is the dated half-year Vertiv and NVIDIA gave, and ABB's reply runs here if it sends one. Eaton's October 13, 2025 release, [supercapacitors, ORV3 busbar and DC connectors](https://finance.yahoo.com/news/eaton-unveils-next-generation-architecture-150000822.html), has no date either, and Flex's March 16, 2026 power rack, [up to 880 kW per compute rack](https://finance.yahoo.com/news/flex-launches-800-vdc-power-203000964.html), has no delivery date. The dates also conflict. Digitimes reported on July 23, 2026 that NVIDIA's 800 V HVDC [enters production in 1Q27 with Delta ramping in 2Q27](https://www.digitimes.com/news/a20260723PD209/nvidia-production-forecast-2027-market.html), and that SemiAnalysis expects large-scale native 800 VDC deployment to slip beyond 2028; NVIDIA's blog three weeks later held to the second half of 2026 for the power rack. Both hold if the power rack ships in 2026 and native compute ramps in 2027, the sequence the SKU table assumes. ## The Frame Agreement Against the Transition Year The public multi-year agreements show the shape. Generac's Q2 2026 8-K describes a global supply agreement whose "product specific terms" were finalized separately and [committed nearly $700 million of volume for 2027](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm), plus a second hyperscaler framework signed June 24, 2026 with 2027 and 2028 volumes still open. Modine took a [$165.0 million up-front deposit](https://www.sec.gov/Archives/edgar/data/67347/000110465926066795/mod-20260331x10k.htm) in March 2026 against more than $4 billion of 2027 to 2029 sales, and its 10-K says its multi-year agreements "typically do not contain contractually-guaranteed volumes." Vertiv's 10-K says its largest customers "require more favorable terms and conditions" on [large, multi-year projects to support artificial intelligence](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm). The deposit mechanics are in [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/). We built the exposure the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) way: a 250 MW critical-load frame at 50 MW a year from 2026 to 2030, N+1 UPS at 1.25 times critical load, and the $700/kW installed price from four federal UPS awards used in [When a Megapack Beats a Static UPS](https://open-factory.com/articles/bess-as-ups/) ($400 low, $1,050 high, January 2026). That is $43.75 million of 415 V UPS a year and $218.75 million over the term. Busway and PDU dollars are excluded because no public $/MW exists for them, so the number is a floor. Five tranches of $43.75 million: 2026 and 2027 land before any transition; 2028 is exposed if native 800 VDC racks arrive on SemiAnalysis's 2027 to 2028 schedule ($131.25 million, 60% of the frame); 2029 and 2030 are exposed on Schneider's 2029 date as well ($87.5 million, 40%). At SemiAnalysis's $1.2 million per MW of UPS content the same shares apply to a $300 million frame. The [assumptions are in the CSV](data/assumptions.csv). **Every dollar in the 2029 and 2030 columns is committed today to a product the buyer's own rack roadmap says it will not want.** The lead time that makes buyers sign early is real, [30 to 48 weeks for a 500 to 1,500 kVA static UPS and 40 to 60 for modular 2,000 kVA-plus](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) in Terrapin's June 2026 survey, carried by the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) as a published lead time. A 48-week lead time justifies ordering the 2027 tranche in 2026. It does not justify fixing the 2030 SKU in 2026, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) shows the 2029 tranche can be released as late as Q1 2028 and still make energization. ## What to Do Monday Split the agreement the way the SKU table splits the room. Sign the five-year frame on the gray-space lines that do not change, MV switchgear, transformers, gensets and paralleling gear, and run their prices through [Quote Check](https://open-factory.com/tools/quote-check/) against the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands. Sign the white-space lines, UPS, busway, PDU, STS, for two years of firm volume with the 2028 to 2030 tranches as reserved capacity, not committed SKUs. Ask every vendor for the sentence Vertiv already wrote in public, that its architecture "doesn't force you to replace what you already have," in the contract rather than the blog, and ask ABB and Eaton for the ship date their releases do not carry. The rectifier and power-rack content moving into the row is largely Taiwanese (Delta, LITEON, Flex), so run the new SKUs through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) before a substitution price is agreed. Behind the paywall: the three clauses, SKU substitution at equivalent price, volume flexibility and technology-refresh rights, written out, and which of the six vendors' public postures and agreement structures support each, including the **two-layer structure that already commits $700 million of 2027 volume** while leaving the product terms open. --- # What 50 MW Costs in 15 US Markets URL: https://open-factory.com/articles/what-50mw-costs-15-markets/ Date: 2026-08-21 Power / Deep dive Dominion $1.5M per MW and 14 years, AEP Ohio 72.5% minimum at 50 MW, Texas $50,000 per MW only above 75 MW, Georgia 15-year contracts, JLL 2.5 to 7 years to power, the three markets still live in 2028, Large-Load Cost Table v0 A 50 MW connection requested in Northern Virginia this quarter energizes in 2033, after posting $75 million of collateral and signing for 14 years. The same 50 MW in Houston posts no rule-mandated security, signs no minimum-take contract, and can be live in 2028. Between those two ends sit thirteen other markets, each with its own threshold, minimum bill, term and exit fee, and until now nobody had put the seven numbers a buyer needs on one page. This piece does: the [Open Factory Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) v0, fifteen utilities side by side, every cell tied to a tariff sheet, a docket or a dated company statement. We cover the table itself, the 75 MW line that decides which Texas regime you land in, the Ohio and Virginia take-or-pay model, what security actually costs in dollars, how contract terms converged on 12 years, the industrial rate underneath each tariff, nineteen months of rule changes, realistic energization years, and, behind the wall, the three markets where 2028 is still on the calendar. Fifteen rows, and the energization column runs from 2028 in Houston, Phoenix under SRP and northern Indiana to 2033 in Northern Virginia, while the security column runs from nothing at all below Texas's 75 MW threshold to $1.5 million per MW at Dominion. ## Fifteen Markets, Seven Cells The table records seven things per market: study fee, security or collateral per MW, minimum bill, contract term, exit fee, CIAC exposure, and the realistic year a 50 MW load requested in Q3 2026 takes service. Each cell in [data/large-load-cost-table.csv](data/large-load-cost-table.csv) names its source URL and date, and a verification flag. Where a utility publishes a number we use it. Where only the structure is public (Georgia's customer-specific contracts, APS's pending collateral formula) the cell says so, and the market is marked partial. | Market (utility) | Threshold | Minimum bill | Term | Security | 50 MW live | |---|---|---|---|---|---| | N. Virginia (Dominion GS-5) | 25 MW | 85% T&D, 60% gen | 14 yr | $1.5M/MW | 2033 | | Columbus (AEP Ohio DCT) | 25 MW | 72.5% at 50 MW | up to 12 yr | 50% of term min. charges | 2031 to 2033 | | Atlanta (Georgia Power) | 100 MW | yes, contract-specific | up to 15 yr | contract-specific | 2031 | | Dallas (Oncor) | 75 MW | none | none | $50k/MW + $50k/MW fee (proposed), 75 MW+ | 2029 | | Houston (CenterPoint) | 75 MW | none | none | deposits, ~$64k/MW implied | 2028 to 2029 | | W./S. Texas (AEP Texas) | 75 MW | none | none | as Oncor | 2029 | | Carolinas (Duke HLF, proposed) | 50 MW at 80% LF | 75% | 10 or 15 yr | cash or LC for min. bills | 2029 to 2030 | | Phoenix (APS XHLF, proposed) | n/p | yes, level n/p | long-term | collateral, formula n/p | 2029 | | Phoenix (SRP E-67) | 20 MW | 80% of forecast | n/p | LCIP upgrades upfront | 2028 to 2029 | | Salt Lake City (Rocky Mountain Power) | 100 MW (SB 132) | n/p | n/p | n/p | 2028 to 2029 | | Louisiana (Entergy) | n/p | minimum charges | 5 yr min. | n/p | 2029 | | Colorado (Xcel, proposed) | 50 MW | 80% | 15 yr | $120k deposit + 6 mo. bills | 2029 to 2030 | | Chicago (ComEd TSAs) | transmission level | revenue commitments | n/p | ~$250k/MW implied | 2031 | | Bay Area (PG&E Rule 30) | 50 to 230 kV service | none | none | 100% facility advance | 2031 to 2032 | | N. Indiana (I&M) | 70 MW | n/p for Indiana | 12 yr + 5 yr ramp | collateral, formula n/p | 2028 to 2029 | *Full table with study fees, exit fees, CIAC exposure, docket numbers and sources: [data/large-load-cost-table.csv](data/large-load-cost-table.csv). n/p = not published.* **The first thing the table shows is that 50 MW is the wrong size for half of these tariffs, and that is the buyer's advantage.** The LBNL and Brattle review of 55 large-load tariffs, published August 10, 2026, found thresholds from 0.3 MW to 150 MW with a median of 25 MW and 75% of tariffs between 5 and 100 MW ([LBNL brief, p. 3](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)). A 50 MW plant or colocation hall sits above AEP Ohio's and Dominion's 25 MW lines and Xcel Colorado's 50 MW line, and below Texas's 75 MW, Indiana Michigan Power's 70 MW, Georgia's 100 MW and Utah's 100 MW. Below the line you get the utility's ordinary industrial tariff and line-extension policy; above it you get the minimum bill, the term and the collateral. The [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool applies this threshold test before it computes anything else, because nothing else in the table matters until you know which side you are on. ## The 75 MW Line in Texas Texas wrote the threshold into statute. SB 6, signed June 20, 2025, applies to new loads of 75 MW or more: a minimum $100,000 study fee, disclosure of duplicate requests, site control, a uniform security requirement, and ERCOT's right to curtail in an emergency ([Greenberg Traurig summary of the PUCT proposal, March 2026](https://www.gtlaw.com/en/insights/2026/3/texas-senate-bill-6-update-what-data-centers-large-load-customers-should-know-about-proposed-interconnection-standards)). The PUCT's implementing rule, proposed 16 TAC §25.194 in Project 58481, was published March 12, 2026 with comments due April 17. It would take $50,000 per MW of financial security at the intermediate agreement, of which 80% is forfeited on withdrawal, a further non-refundable $50,000 per MW interconnection fee at the interconnection agreement, study fees of $100,000 for 75 to 250 MW and $300,000 above 250 MW, and 100% CIAC for radial lines and substations with no revenue credit ([ERCOT to Senate B&C Committee, April 1, 2026, slide 9](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf)). In the documents we could open the rule was still a proposal in August 2026. At 50 MW, none of that applies. What applies is the TDSP's standard facilities-extension process and, since June 18, 2026, the shadow of Batch Zero, ERCOT's batch study for loads of 75 MW and up, approved by the PUCT that day with 438 GW of requests pending and a final transmission plan promised for fall 2027 ([ERCOT release, June 18, 2026](https://www.ercot.com/news/release/06182026-puct-approves-ercots)). Then on August 3 Governor Abbott directed the PUCT and ERCOT to audit every data center in a queue he put at 474 GW, "more than five times Texas' record peak electricity demand," before approving any more ([Governor's letter, August 3, 2026](https://gov.texas.gov/uploads/files/press/Thomas_Gleeson_Pablo_Vegas_Data_Centers_Directive_Letter_to_PUCT_ERCOT_August_2026_.pdf)). ERCOT told the PUCT on August 20 it is aiming for a December 10 audit report, that the Batch Zero study will not be done by April 9, 2027, and that it will also run a community-impact review on facilities of 25 MW and above ([Utility Dive, August 21, 2026](https://www.utilitydive.com/news/ercot-texas-puc-data-center-audit/828472/)). The queue grew from 63 GW at the end of 2024 to 233 GW in December 2025, 410 GW on March 26, 438 GW on June 18 and 474 GW on August 3, 2026, while only 17 loads totaling 6.6 GW had cleared every ERCOT gate except energization by August 20. We wrote up the funnel from 474 GW requested to 3.9 GW drawing power in [The 1.6% Funnel](https://open-factory.com/articles/ercot-funnel/) in July. The point for a 50 MW buyer is narrower. **Below 75 MW you are outside SB 6, outside Batch Zero and outside the $100,000-per-MW stack, but not outside the audit, which reaches down to 25 MW.** Oncor told analysts on August 6 that of its 298 GW of requests about 44 GW met the Batch Zero criteria and that none of that load is in its $47.5 billion capital plan before 2027 ([Utility Dive, August 7, 2026](https://www.utilitydive.com/news/fate-of-oncors-nearly-300-gw-load-pipeline-unclear-following-texas-data-ce/827303/)). CenterPoint said on July 28 that 14 GW of its 17 GW of Batch Zero submissions should qualify, backed by signed facility extension agreements and about $900 million of customer cash commitments and deposits, which works out to roughly $64,000 per MW ([Utility Dive, July 28, 2026](https://www.utilitydive.com/news/centerpoint-energy-14-gw-texas-batch-zero-large-load-earnings/826329/)). ## Ohio and Virginia: The Take-or-Pay Model AEP Ohio's Schedule DCT is the tariff everyone else copied, and it is the only one of the fifteen where every cell is on the utility's own web page. PUCO approved it July 9, 2025 after a May 2024 filing ([AEP release, July 9, 2025](https://www.aep.com/news/stories/view/10327/)). It applies at 25 MW, aggregated across affiliates. The load study fee is $10,000 for 25 to 50 MW, $50,000 for 50 to 100 MW and $100,000 above. The ramp is at most four years at 50, 65, 80 and 90% of contract capacity, followed by eight years. Collateral is 50% of the total minimum charges over the full term unless you carry A- and A3 ratings and ten times the requirement in cash; surety bonds are not accepted. The exit fee is 36 months of minimum charges, available only after the fifth year following the ramp, and before energization a Letter of Agreement makes you liable for 100% of the buildout if you cancel or slip more than 12 months ([AEP Ohio Data Center Tariff page](https://www.aepohio.com/company/about/rates/data-center-tariff/)). The headline 85% is not what a 50 MW customer pays. The minimum demand for 25,001 to 75,000 kW is 15,000 kW plus 85% of capacity above 25,000 kW, so at 50 MW the floor is 36,250 kW, or 72.5% of contract, and the 85% cap only binds above 75 MW ([AEP Ohio Data Center Tariff page](https://www.aepohio.com/company/about/rates/data-center-tariff/)). The tariff did what it was built to do: of 30,000 MW that had expressed interest, 13,022.7 MW paid for a study and 5,642 MW signed binding contracts by February 12, 2026, on top of 12,219 MW signed before the tariff, against a system peak of 8,000 to 10,500 MW ([AEP Ohio, February 13, 2026](https://www.aepohio.com/company/news/view?releaseID=10753)). Those contracts come online "progressively through 2035," and the next study tranche opens October 1, 2026 for applications received by September 30. Dominion's Schedule GS-5 is the harder version. The SCC created the class for loads of 25 MW and up in its November 25, 2025 order in the biennial review (PUR-2025-00058), with a 14-year minimum term, a four-year ramp at no less than 20% of contract capacity a year, minimum demand charges of 85% of contract for distribution and transmission and 60% for generation, and collateral of $1.5 million per MW of contracted capacity ([LBNL brief, pp. 4 to 6 and 13](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)). On July 31, 2026 the SCC went further in PUR-2026-00056 and ordered a mandatory CIAC for "direct connect" transmission, meaning the substations and lines that connect a large load to the bulk system, net of the transmission revenue the customer is expected to produce, with Dominion's amended line-extension policy due within 90 days. The order also floated assigning higher-order transmission costs to the GS-5 class outright ([SCC Final Order, July 31, 2026](https://www.scc.virginia.gov/docketsearch/DOCS/8dt001!.PDF)). For a 50 MW contract, Dominion's 85% floor with its published ramp commits 510 MW-years over 14 years, AEP Ohio's 72.5% formula with its 50/65/80/90 ramp commits 386 MW-years over 12, Duke's proposed 75% commits 375 over 10, and Xcel Colorado's proposed 80% over 15 years commits 600, the largest of the four. At a demand charge of $10 per kW-month, an assumption you should replace with your own rate, 510 MW-years is about $61 million of demand charges due whether the plant runs or not, which is why the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool asks for your expected load factor before it asks for anything else. LBNL's median minimum billing demand across the tariffs it reviewed is 80%; Oregon's Schedule 96 and I&M's proposed Michigan tariff sit at 90%, Duke's proposal at 75%, and Dominion's generation component at 60% is the lowest published floor among the approved tariffs ([LBNL brief, p. 5](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)). ## Security: $75 Million or Nothing The security cell is where the fifteen markets diverge most, and where the fewest utilities publish a number. Dominion's $1.5 million per MW is a tariff figure: $75 million for 50 MW, posted before a kilowatt-hour flows. ComEd does not publish a rate, but Exelon told analysts on July 30 that about 4 GW of data center load with signed Transmission Security Agreements had posted $1 billion of collateral, an implied $250,000 per MW, or $12.5 million for 50 MW ([Utility Dive, July 31, 2026](https://www.utilitydive.com/news/exelon-data-center-load-new-jersey-battery-earnings/826686/)). Those TSAs, approved by FERC in February and March 2026, carry a defined ramp, a facility readiness obligation, committed revenue contributions with shortfall payments and a termination-fee schedule ([Utility Dive, March 11, 2026](https://www.utilitydive.com/news/ferc-comed-exelon-data-center-service-agreement/814414/)). Dominion's $75.0 million is six times ComEd's implied $12.5 million, ten times the $7.5 million a 75 MW Texas project would post and forfeit under the proposed rule, 23 times CenterPoint's implied $3.2 million of deposits, and 125 times Xcel Colorado's proposed $600,000 of deposit and studies. **Collateral, not the minimum bill, is the line item that decides whether a 50 MW project pencils in Virginia, because $75 million of letters of credit costs a developer roughly $1 to 2 million a year to carry before the first demand charge.** AEP Ohio's collateral is a formula, 50% of term minimum charges, that steps down by one year of minimum charges for each year of on-time payment ([LBNL brief, p. 7](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)). PG&E's Rule 30 takes no collateral as such. It takes an Advance equal to the project-specific cost estimate for service facilities, interconnection upgrades and network upgrades at 50 to 230 kV, refunds it over time under rules the CPUC will finalize in Application 24-11-007, and forfeits all remaining refunds if you terminate after the facilities are in service ([PG&E Electric Rule 30, effective December 4, 2025](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_RULES_30.pdf)). ## Terms Are Converging on 12 Years Contract length moved faster than any other cell. LBNL found the median minimum term was five years for tariffs proposed before 2025 and 12 years for tariffs proposed since, with a range from one to 20 years ([LBNL brief, p. 4](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)). Georgia's PSC raised the allowed term for 100 MW customers from 5 to 15 years on January 23, 2025, added minimum bills, required contracts to be filed 30 days before execution, and made those customers pay for upstream generation, transmission and distribution "as construction of the data centers progresses" ([Georgia PSC release, January 23, 2025](https://psc.ga.gov/site/assets/files/8617/media_advisory_data_centers_rule_1-23-2025.pdf)). Georgia Power's TOU-SC-16 sheet now bars any customer with an expected peak of 100 MW or more from the standard price option unless it contracted before February 1, 2025 ([Georgia Power TOU-SC-16, June 2026](https://www.georgiapower.com/content/dam/georgia-power/pdfs/tariffs/tou-sc.pdf)). Entergy Louisiana's Large Load High Load Factor rate keeps a five-year minimum, Duke's proposed HLF runs 10 or 15 years by load size, AEP Ohio 12, I&M 12 after a ramp of up to five, Dominion 14, Xcel Colorado's proposal 15, and Portland General's Schedule 96 scales from 10 years at 20 MW to 30 years at 220 MW under Oregon's May 7, 2026 order ([OPUC Order 26-154](https://apps.puc.state.or.us/orders/2026ords/26-154.pdf)). Indiana Michigan Power's settlement, approved February 19, 2025, is the model for how term and exit interact: loads of 70 MW at a site or 150 MW across a company sign for at least 12 years after a ramp of up to five, may cut contract capacity by up to 20% after year five with 42 months' notice at no charge, and pay an exit fee for anything larger, with reductions over 20% now needing IURC approval ([Utility Dive, November 25, 2024](https://www.utilitydive.com/news/indiana-michigan-power-aep-amazon-google-microsoft-data-center-interconnect/733850/); [AEP, February 19, 2025](https://www.aep.com/news/stories/view/10037/)). Duke's version, filed in late June 2026 for a 2027 start, is lighter: 75% minimum, security for the minimum bills, and an exit penalty of 25% of remaining bills, which Latitude Media's analyst called "pretty light-touch" against Dominion ([Latitude Media, July 8, 2026](https://www.latitudemedia.com/news/duke-energys-light-touch-large-load-tariff-could-be-a-problem/)). ## The Rate Underneath the Tariff None of these instruments sets the price of energy. The industrial rate does, and it now varies 3.2 to 1 across the fifteen markets. EIA's May 2026 table puts Texas industrial power at 6.33 cents per kWh, Louisiana at 6.88, Georgia at 7.04, Arizona at 7.54, North Carolina at 7.78, Utah at 8.10, Indiana at 9.08, Colorado at 9.47, Ohio at 9.87, Illinois at 10.20, Virginia at 10.53 and California at 20.20, against a US average of 8.71 ([EIA Electric Power Monthly, July 2026, Table 5.6.A](https://www.eia.gov/electricity/monthly/archive/july2026.pdf)). Ohio's industrial rate rose 29.7% in twelve months, from 7.61 to 9.87 cents, Illinois 14.3% and Virginia 14.0%, while Texas fell 3.9% and Arizona 8.6%. The three risers are PJM states, and the driver is the capacity charge: $333.44 per MW-day for 2027/28 is about $122,000 per MW-year, and we walked through how that lands on a plant's bill in [A Factory Owner's Guide to the PJM Capacity Charge](https://open-factory.com/articles/pjm-capacity-factory-bill/) in May. For a 50 MW load at full contract demand the [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) puts the 2027/28 capacity line near $6.1 million a year before any tariff minimum, and PJM's board decided on July 27 to exclude new large loads that do not bring their own supply from future auction forecasts and to make them curtailable near emergencies from June 2027 ([Utility Dive, July 28, 2026](https://www.utilitydive.com/news/pjm-board-backstop-capacity-auction-data-center-curtailment/826347/)). Rate and tariff pull the same way: the three markets with the longest terms and largest collateral also have the most expensive and fastest-rising energy. ## Nineteen Months of Rule Changes Every cell in the table has a date on it because most of them changed between January 2025 and August 2026. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) includes the running list. The eighteen entries that touch these fifteen markets are below. Eighteen actions in nineteen months, eleven of them since March 2026: Georgia's rule (January 23, 2025), Indiana's I&M order (February 19, 2025), Utah SB 132 (March 2025), Texas SB 6 (June 20, 2025), AEP Ohio's DCT (July 9, 2025), PG&E Rule 30 (July 2025), Dominion GS-5 (November 25, 2025), the PUCT's forecasting rule (March 1, 2026) and proposed interconnection standards (March 12), Xcel Colorado's filing (April 2), Pennsylvania's model tariff (April), Oregon's Order 26-154 (May 7), Batch Zero and FERC's show-cause orders (June 18), Duke's HLF filing (late June), PJM's board decision (July 27), the SCC's CIAC order (July 31), Abbott's directive (August 3), the LBNL brief (August 10) and the PUCT's good-cause exception for ERCOT (August 20). Utah's law is the outlier in direction: SB 132 defines a large load as 100 MW within five years, gives Rocky Mountain Power 90 days to evaluate, and if the utility cannot serve without major investment lets the customer contract with separate generation, paying the studies, upgrades and transmission rate itself ([Latitude Media, April 15, 2025](https://www.latitudemedia.com/news/utah-is-taking-a-different-approach-to-new-data-center-load/)). Xcel's Colorado filing goes the other way: 50 MW threshold, a $120,000 deposit plus studies for about $600,000 upfront, a 15-year contract at 80%, security of six months of operations, and an exit fee equal to every remaining minimum monthly bill ([Colorado Sun, April 2, 2026](https://coloradosun.com/2026/04/02/data-centers-xcel-energy-colorado-rate-tariff-consumers/)). ## Energization: 2028 to 2033 The last column is the one buyers ask about first and utilities publish last. JLL's 2026 outlook is the only source we found that charts average grid-connection lead time specifically for a new 50 MW data center by market: about 2.5 years in Dallas and Phoenix, about 5 in Atlanta and Chicago, about 7 in Northern Virginia, with the primary-market average above four years ([JLL 2026 Global Data Center Outlook, p. 18](https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-reports/26-research-global-data-center-outlook-new.pdf)). A CPS Energy planning presentation at PowerGen 2026 puts Columbus at 84 months, Silicon Valley at 72 and Pittsburgh, Chicago, Houston and Dallas at 36 ([as reported by Avanza Energy, 2026](https://avanzaenergy.substack.com/p/data-centers-are-killing-the-grid)). We combined those with utility statements to set a range per market. Houston, northern Indiana and Salt Lake City sit at 2 to 3 years from a Q3 2026 request, Phoenix under SRP at 2.5 to 3, Dallas, AEP Texas, Phoenix under APS and Louisiana at 2.5 to 3.5, the Carolinas and Colorado at 3 to 4, Atlanta and Chicago at 4.5 to 5.5, the Bay Area at 5 to 6, Columbus at 5 to 7 and Northern Virginia at 6.5 to 7.5. Behind the Virginia number is a Dominion queue of 70 GW against an all-time peak of 24.6 GW, growing by about 10 requests a month ([former FERC Chairman Christie, Utility Dive, February 2026](https://www.utilitydive.com/news/states-ferc-large-load-tariff-doe-christie/816672/)). Behind the Columbus number is 17,861 MW of signed contracts scheduled through 2035. Behind the Houston number is a utility with 3.5 GW under construction, 8 GW of data center load it expects to energize by 2029, and a pipeline in which 90% of projects are 500 MW or less and sited near existing substations ([Utility Dive, April 23, 2026](https://www.utilitydive.com/news/centerpoint-energy-data-center-load-earnings/818293/)). **The years in that column are gated by equipment as much as by paperwork.** A 50 MW service at 138 kV needs a 60 to 75 MVA transformer, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists 128 weeks as the published lead time for power transformers and 144 weeks for GSUs, numbers we traced back to Butler Works and one steel mill in [Transformer Lead Times Hit 160 Weeks](https://open-factory.com/articles/transformer-160-weeks/). The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tool will tell you that a utility promising 2028 energization must have ordered the transformer before this article was published, and the [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool will tell you what to budget if you are asked to buy it yourself under a CIAC. ## What to Do Monday Size the request against the threshold. At 50 MW you are inside AEP Ohio, Dominion, Xcel Colorado and SRP's tariffs and outside Texas, Indiana, Georgia and Utah's. A 49 MW request in Columbus saves $40,000 on the study fee and nothing else, while a 74 MW request in Dallas avoids the entire proposed $100,000-per-MW stack and Batch Zero. Ask the utility for three numbers in writing before you sign a letter of intent: the collateral formula in dollars for your load, the minimum demand in kW for your contract capacity (not the headline percentage), and the exit fee in months. Write the energization date into the ESA with a term adjustment if the utility misses it; AEP Ohio's contract term starts on the actual energization date and lets you petition PUCO if infrastructure is late, and that clause is worth copying into every other market. Price the minimum bill at your realistic load factor, not the contract capacity, and if that number exceeds what a 50 MW bridge of engines or fuel cells would cost, run the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool and read [Recips vs Turbines vs Fuel Cells](https://open-factory.com/articles/turbines-recips-fuel-cells/) before you sign for 14 years. And if you are in a PJM state, budget the capacity charge as a separate line, because no tariff in this table waives it. Every market above 4 years on the energization chart is one where a 2028 plant needs a bridge, and we covered what bridges cost and how long they take in [Behind the Meter, In Front of the Queue](https://open-factory.com/articles/behind-meter-90gw/) and the slot deposits behind them in [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/). The three markets below 3 years are a different problem: there the question is which tariff mechanics get a 50 MW load to 2028 without a 12-year floor. Behind the paywall: the three markets where a 50 MW buyer can still energize in **2028**, the specific threshold, deposit and study rules that make it possible in each, and the one clause in each that can take the date away, all drawn from the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). --- # Liquidated Damages at 0.5% a Week: Why Caps Sit at the Vendor's Margin URL: https://open-factory.com/articles/liquidated-damages/ Date: 2026-07-20 Procurement / How-to Cap norms by equipment class, what a week of delay costs per MW, negotiating above the cap, LD vs schedule bonus, real clauses from utility bid tabs A week of delay on a 50 MW datacenter hall costs about $2.7 million. The liquidated damages clause in the transformer contract behind that hall pays back $74,000 for the same week, and stops paying after ten weeks. On a 2 MW plant the ratio is worse: $649,000 of lost value added, diesel and carrying cost against $13,300 of LDs. We pulled every liquidated damages clause we could find in public transformer, switchgear and genset contracts, priced a week of delay for both buyers from the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/), and compared the cap to what the vendor actually earns. This piece covers the clauses as written, the 0.5% a week and 5% cap norm, the cost of a week for a plant and a hall, the caps against OEM margins, what courts do with these clauses, and, behind the paywall, how to negotiate above the cap. The 2 MW plant loses $649,000 a week and recovers $13,300. The 50 MW hall loses $2.73 million and recovers $74,000, so the clause covers 2.0% and 2.7% of the weekly loss respectively. ## What the Public Clauses Actually Say Liquidated damages are a pre-agreed sum for a breach, usually late delivery, that replaces proof of actual loss. Under [UCC 2-718](https://www.law.cornell.edu/ucc/2/2-718) they are enforceable "only at an amount which is reasonable in the light of the anticipated or actual harm," and "a term fixing unreasonably large liquidated damages is void as a penalty." That sentence is the reason every clause is written low. The seller's lawyer argues that anything above their margin is a penalty. The buyer's lawyer, who cannot prove the future loss on a plant that does not exist yet, takes what is offered. We read the eleven positions below in full. They come from City of Ocala's two power transformer contracts (2023 and 2024), BrightRidge's March 2026 transformer bid, the FAR, and the standard terms that Hitachi Energy, Eaton and Cummins attached to public quotes. Every one is in the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) with its URL. [CSV: data/01-ld-clauses.csv](data/01-ld-clauses.csv) Of the eleven, three buyers secured a rate and a 5% cap, two federal clauses set a per-day figure with no cap in the clause text, three OEM standard terms exclude liquidated damages entirely, and three seller clauses excuse or waive delay. **The 0.5% a week, 5% cap norm is real, and it is the seller's number, not the buyer's.** When PTI Transformers bid $8,070,340 for two 105/140/175 MVA autotransformers for Ocala's Dearmin substation in June 2023 (Virginia Transformer bid $16,168,895 for the same two units), [PTI's proposal](https://ocala.legistar1.com/ocala/attachments/556334d5-416c-4ed0-8509-a742afa8fd8e.pdf) fixed the buyer's remedy at "0.5% per week to a maximum of 5% of the purchase price of the unit, the delivery of which has been delayed," and then the lesser of that and "the actual liquidated damages suffered by Purchaser," and then, in the document's own words, "No harm no foul." The clause only fires for delay "solely attributable to the negligence or willful act of Seller," and PTI's force majeure clause in the same proposal excuses "inability to obtain necessary labor, materials, manufacturing facilities, or supplies," which in a [143-week transformer market](https://open-factory.com/articles/transformer-160-weeks/) covers most delays a buyer will ever see. The city's own paper, in the [executed agreement](https://ocala.legistar1.com/ocala/attachments/ea734d92-0505-4dd3-a9ea-63a9aed98f38.pdf), sets $1,742 per calendar day for late substantial completion, about 0.30% a week on a $4.0 million unit, with no stated cap. A year later Ocala bought three 25/33.3/41.6 MVA transformers from Virginia Transformer at $1,892,556 each. The [contract](https://ocala.legistar1.com/ocala/attachments/e3d930d9-cbb6-4618-be9e-c40cd7b5929a.pdf) requires delivery within 72 to 76 weeks against a stated "manufacturer's standard lead times (96 to 136 Weeks)," sets LDs at "$500 PER DAY for the 25/33.3/41.6 MVA power transformer(s)," and caps them at "FIVE PERCENT (5%) of the offending unit contract value." That is 0.18% a week. The same contract lets Virginia Transformer charge the city "a re-scheduling fee of THIRTY PERCENT (30%) of the specific order value" if the city's delay loses the production slot, and pass through up to 6% of cost increases. The seller's remedy for buyer delay is six times the buyer's remedy for seller delay. [CSV: data/02b-ld-rates-and-caps.csv](data/02b-ld-rates-and-caps.csv) Weekly rates run from 0.18% (Virginia Transformer, $500 a day on a $1.89 million unit) to 0.30% (Ocala's own $1,742 a day) to 0.50% (PTI), and every stated cap is 5%. BrightRidge, the Johnson City, Tennessee utility, put "$250 per day, not to exceed five percent (5%) of the purchase price of the unit" in its [March 2026 bid form](https://brightridge.com/wp-content/uploads/2026/03/81023.8.pdf) for a 30/40/50/56 MVA, 69:13 kV transformer, with the trigger at "Certification" ready for energization and the right to set the damages off against milestone payments. The federal government's own clause, [FAR 52.211-11](https://www.acquisition.gov/far/52.211-11), leaves the daily figure blank for the contracting officer and makes it payable "in place of actual damages"; [FAR 11.501](https://www.acquisition.gov/far/11.501) tells that officer the rate "must be a reasonable forecast of just compensation" and that LDs "are not punitive." Then there is the paper the OEMs send. [Hitachi Energy's terms](https://ocala.legistar1.com/ocala/attachments/c7ce8f7a-37ba-4f38-a21a-498de85a6eca.pdf), attached to a 2024 WESCO quote for a 2,500 kVA pad-mount, say the company shall in no event "be liable for penalties or penalty clauses of any description." [Eaton's terms](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf) on Silicon Valley Power's switchgear orders exclude "LOSS OF PROFITS, LOSS OF REVENUE, LIQUIDATED DAMAGES, OR LOSS OF USE" and cap liability at the purchase price; we walked through the rest of that document in [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/). [Cummins' Sourcewell terms](https://ocala.legistar1.com/ocala/attachments/febd0aa7-2119-4a8b-8b6d-e9119499e587.pdf) on two 500 kW mobile gensets exclude "LIQUIDATED, OR CONSEQUENTIAL DAMAGES OF ANY KIND." If your PO went out on the vendor's quote form, this is your clause, and the [RFQ checklist](https://open-factory.com/articles/rfq-capital-equipment/) exists to stop that. ## What a Week of Delay Costs a 2 MW Plant The 2 MW case is the plant from [The 2 MW Plant Service Upgrade](https://open-factory.com/articles/2mw-plant-service-upgrade/): a 2,500 kVA pad-mount and a 15 kV metal-clad lineup, the lineup priced at Riverside's $2,561,551.88 SUB-871 award and the pad-mount at about $98,000, a $2.66 million package. At 0.5% a week the LD is $13,300. The 5% cap is $133,000 and is reached in week 10. The [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) puts the lineup at the end of the critical path, so a late lineup is a late plant. The loss has four lines, all in [data/02-cost-of-delay-weekly.csv](data/02-cost-of-delay-weekly.csv). Lost output is the big one. We have no plant-specific margin, so we use the national average: BEA puts manufacturing value added at [9.4% of GDP](https://fred.stlouisfed.org/series/VAPGDPMA) in Q4 2025, or $2,954 billion on [GDP of $31,423 billion](https://fred.stlouisfed.org/series/GDP), against EIA's [1,042,217 GWh of 2025 industrial retail sales](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_1), which is $2.83 of value added per industrial kWh. A 2,000 kW plant at 60% load factor uses 201,600 kWh a week, so ~$571,000 of value added a week stands idle. An energy-intensive plant will be well under that figure. A machine shop or a pharma line will be over it. Bridge power is the second line. A 2,000 kW rental unit listed at [$33,945 a month at Ziegler](https://www.zieglerrental.com/equipment/lights-generators/generators-diesel/2000kw-diesel-generator/) is $7,800 a week bare; a full-service package in Houston runs [$75,000 to $110,000 a month](https://stagrentals.com/understanding-the-cost-of-renting-a-2000-kw-generator-what-you-need-to-know/). Fuel dominates: the [Cat XQ2000 burns 77.7 gallons an hour at 50% load](https://ohiocat.com/wp-content/uploads/2023/06/XQ2000-Ohio-CAT-Spec-Sheet-2020.pdf), 13,054 gallons a week, and at EIA's [$5.134 a gallon for the week of July 20, 2026](https://fred.stlouisfed.org/series/GASDESW) that is $67,000 a week. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool runs the same math at any load. Carrying cost on the stranded package at [Moody's Baa yield of 6.19%](https://fred.stlouisfed.org/series/BAA) (July 2026) is $3,200 a week. If the electric service agreement has already started billing contract demand, AEP Ohio's [GS primary demand charge of $8.08/kW plus the $10.79/kW transmission rider](https://open-factory.com/articles/pjm-capacity-factory-bill/) adds $8,700 a week for nothing delivered. We leave that line out of the total because most ESAs start at energization. **Total: $649,000 a week against $13,300 of liquidated damages, a 49 to 1 ratio, and the clause stops paying at week 10.** Even a buyer who throws out the value-added line and counts only diesel, rental and carrying cost is at $78,000 a week, six times the LD. ## What a Week of Delay Costs a 50 MW Hall The 50 MW case uses the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) inputs and the electrical package from Riverside's [SUB-872 award](https://riversideca.legistar1.com/riversideca/attachments/c3dcbbeb-623f-4217-a27a-c3909abd87f2.pdf): three substation power transformers, two of them 45 MVA, from Delta Star at $14,878,322, bid opened April 16, 2026, awarded July 10, 2026, against Pennsylvania Transformer's $15,175,278. At 0.5% a week that package yields $74,400 of LDs. The 5% cap is $744,000, the 10% cap $1.49 million. Lost rent is the first line. Digital Realty's [Q4 2025 supplemental](https://www.sec.gov/Archives/edgar/data/1297996/000110465926010887/dlr-20260205xex99d1.htm) reports new Americas leases above 1 MW signed at $186 per kilowatt per month of annualized GAAP base rent, so a 50,000 kW hall forgoes $9.3 million a month, or $2.15 million a week. Carrying cost is the second: Turner & Townsend's [2025 Data Centre Cost Index](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends) puts Phoenix and Columbus at $9.8 a watt, so a 50 MW build is $490 million, and at 6.19% that is $583,000 a week of interest on a building that cannot bill. The [MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes the same per-component view with each line's lead time from the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). **Total: $2.73 million a week against $74,400 of LDs, a 37 to 1 ratio.** Bridging the hall on rental diesel instead of waiting does not change the order of magnitude: 25 rented 2 MW units are $196,000 a week bare and burn $1.68 million a week of diesel at 50% load, before permits, which is why the [gas genset and turbine bridge market](https://open-factory.com/articles/gensets-sold-out-2028/) exists. [CSV: data/04-cumulative-50mw.csv](data/04-cumulative-50mw.csv) At ten weeks late the hall has lost $27.3 million and collected $744,000; at twenty weeks it has lost $54.6 million and collected $1.49 million under a 10% cap, or still $744,000 under a 5% one. The minimum-bill tariffs we mapped in [What 50 MW Costs in 15 Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/) make this worse, not better: AEP Ohio's 85% minimum and Dominion's GS-5 floor bill contracted capacity on a schedule the transformer vendor does not sign. The vendor's exposure is 5% of its own price. The buyer's is the tariff, the lease and the loan. ## Why the Cap Sits Where It Does The folk explanation for the 5% to 10% cap is that it equals the vendor's margin: a vendor that pays the full cap on a contract has worked for free, and no rational vendor accepts a downside beyond zero profit. The [procurement research file](https://open-factory.com/models/bid-tab-price-book/) records the EPC version of the rule: contractors cap total LDs at 5% to 10% "because that is the contractor's margin." That was true for the people who wrote the first versions of these clauses. It is not true for the OEMs now signing them. [CSV: data/03-oem-margins-vs-caps.csv](data/03-oem-margins-vs-caps.csv) Eaton's Electrical Americas segment earned 29.9%, Cummins Power Systems 21.7%, Hubbell Utility Solutions 21.5%, Vertiv 20.4%, Powell 19.7%, Generac 17.0% and GE Vernova Electrification 14.9%, against EMCOR's 10.1% and caps of 5% and 10%. The filings are unambiguous. Eaton's Electrical Americas segment posted [$3,972 million of operating profit on $13,276 million of sales in 2025](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000002/etn12312025exhibit99.htm), a 29.9% margin, with backlog up 31%. Powell, which builds the switchgear lineups in these bid tabs, reported [operating income of $217.9 million on $1,104.3 million](https://www.sec.gov/Archives/edgar/data/80420/000008042025000152/powl-20250930.htm) in fiscal 2025, 19.7%, and its 10-K says plainly that "many of our customer contracts have schedule and performance obligation clauses that, if we fail to meet, could subject us to penalty provisions, liquidated damages or claims against us." Hubbell's Utility Solutions segment ran a [21.5% GAAP operating margin](https://www.sec.gov/Archives/edgar/data/48898/000162828026004800/exhibit991_02032026.htm) (24.1% adjusted). Cummins' Power Systems segment posted [EBITDA of 21.7% of sales](https://www.sec.gov/Archives/edgar/data/26172/000002617226000004/cmi2025q48-kex99.htm) in Q4 2025 on data center demand. GE Vernova's Electrification segment, the grid equipment business, went from [9.0% to 14.9% EBITDA margin](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000012/gevpressrelease4q25.htm) in a year with equipment backlog up 53%. The one company on the chart near the 10% line is EMCOR, an electrical contractor, at [10.1%](https://www.sec.gov/Archives/edgar/data/105634/000010563426000027/eme-ex991_20251231xq4.htm). **A 5% cap on a 20% to 30% margin business means the vendor that delivers a year late still clears 15 to 25 points on your order.** The cap did not move when the margins did, because the buyer who negotiates it is reading the vendor's paper, not the vendor's 10-K. The [Book-to-Bill Table](https://open-factory.com/models/book-to-bill-table/) puts the segment margin next to the backlog for every public OEM, which is the number to have open when the vendor says the cap is all they can bear. The sellers' side of the same asymmetry is visible in the clauses themselves: PTI's cancellation schedule charges the buyer 5% at order, 40% once materials are ordered and 100% after vapour phase; Virginia Transformer charges 30% for a lost slot; Hitachi Energy's quote bills [$375 to $425 a week per transformer](https://ocala.legistar1.com/ocala/attachments/c7ce8f7a-37ba-4f38-a21a-498de85a6eca.pdf) for storage if the buyer is not ready. Every one of those is uncapped or effectively so. [CSV: data/06-remedy-asymmetry.csv](data/06-remedy-asymmetry.csv) In the same two Ocala contracts the seller's charges for buyer delay are 30% (lost slot), 40% (cancellation after materials are ordered) and 100% (after vapour phase), while the buyer's cap on seller delay is 5%, and the three OEM standard terms allow 0%. ## What Courts Do With the Clause Three things a buyer should know before signing, none of which the vendor's form will tell you. First, a clause written as liquidated damages is enforced as written, even when the actual loss is smaller. In [California & Hawaiian Sugar v. Sun Ship](https://www.courtlistener.com/opinion/472662/california-and-hawaiian-sugar-company-v-sun-ship-inc/) (9th Cir. 1986, 794 F.2d 1433) a shipyard owed $17,000 a day for a late barge and paid it under Pennsylvania's version of UCC 2-718 although the buyer's provable loss was far below the total. In [Wahlcometroflex v. Westar Energy](https://www.courtlistener.com/opinion/2756496/wahlcometroflex-inc-v-westar-energy-inc/) (10th Cir. 2014, 773 F.3d 223), an equipment supplier to a Kansas power plant argued the utility had to prove that its "late delivery of equipment" caused the damages. The court held that under the clause Westar "did not need to establish" any such thing. In [DJ Manufacturing v. United States](https://www.courtlistener.com/opinion/719833/dj-manufacturing-corporation-v-united-states/) (Fed. Cir. 1996, 86 F.3d 1130) the government withheld LDs from a supplier of field packs during Desert Storm and the Federal Circuit let it. The clause is a bargain, and courts hold both sides to it. Second, the clause is usually the only remedy. FAR 52.211-11 says the damages are paid "in place of actual damages." PTI's clause is the "full aggregate amount." Once LDs are the sole remedy for delay, the lost rent and the diesel are gone as a claim, which is why the cap matters more than the rate. Third, [UCC 2-719(3)](https://www.law.cornell.edu/ucc/2/2-719) lets a seller exclude consequential damages "unless the limitation or exclusion is unconscionable," and between two businesses it almost never is. Eaton's and Cummins' exclusions will hold. A buyer who signs the vendor form has agreed that a $2.7 million week costs the vendor nothing. The practical corollary sits in a Riverside staff report. In January 2026 Prolec GE Waukesha was the apparent low bidder on two 66 kV grounding transformers at $2,749,330.28, but [conditioned its offer with a 30-day price hold and "exceptions to the City T&C's"](https://riversideca.legistar1.com/riversideca/attachments/e9dd393f-dfed-48b6-a244-0d3cd719674c.docx), confirmed both when asked, and was ruled non-responsive; Niagara Power Transformer took the $2,762,537 award. A public buyer with mandatory terms can hold the line for $13,207. A private buyer can too, but only if the terms are in the RFQ before the quote arrives. ## What to Do Monday Put the LD clause, the cap, the trigger and the sole-remedy language in the RFQ, not in the PO negotiation, and make acceptance of the terms a condition of a responsive bid, the way Riverside did. Compute your own week of delay before you open bids: lost output or rent, bridge power at the [Genset TCO](https://open-factory.com/tools/genset-tco/) rate, carrying cost at your loan rate, and any minimum bill from the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). Write that number into the recitals as the anticipated harm; under UCC 2-718 it is what makes a higher rate "reasonable" rather than a penalty. Run the vendor's quoted lead time through the [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) so the delivery date you are protecting is one the market can meet. Then price the terms against the quote with [Quote Check](https://open-factory.com/tools/quote-check/): a vendor who will not accept 0.5% a week and a 10% cap on a 20% margin business is telling you where the schedule risk really sits. Behind the paywall: the four ways to get above the 5% cap that vendors with 20% to 30% margins have accepted in public contracts, model language for each, and the order to redline them in so the **30%** rescheduling fee in the vendor's paper becomes the number that anchors yours. --- # How to Write an RFQ for Capital Equipment in 2026 URL: https://open-factory.com/articles/rfq-capital-equipment/ Date: 2026-07-14 Procurement / How-to Ocala's $1.1M PPI escalator, LDs at $250 a day capped at 5%, 25/35/30/10 milestones, FAT as the payment trigger, IEEPA refund allocation, 156-week pad-mounts, the six templates The City of Ocala, Florida agreed to pay [$8,070,340 for two 105 MVA autotransformers in August 2023](https://webapi.legistar.com/v1/ocala/Matters/3941) and approved [$9,170,340 for the same two units on April 28, 2026](https://webapi.legistar.com/v1/ocala/Matters/10210), seven months before the first one arrives. The $1.1 million difference is one clause: a PPI escalator the vendor wrote into its proposal and the city accepted because the alternative was a bid twice the price. Every capital equipment RFQ written this year has to answer the same six questions Ocala's did, and most of the answers now sit in public procurement files where anyone can read them. This piece covers what changed in 2026, six template sections with model language from real utility, municipal and federal RFQs, and, behind the wall, the full templates and the redlines vendors send back. Five real transformer payment schedules run from 0% paid before delivery (Lake Placid, Mount Pleasant) to 60% paid before the factory test (PTI Transformers on Ocala's order) and 90% at shipment (Virginia Transformer on Ocala's next order). The [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds the documents behind every line. ## What Changed Between February and July 2026 **Four legal events and one lead-time number moved the RFQ this year, and none of them are in the template your plant inherited.** On February 20, 2026 the Supreme Court held in Learning Resources v. Trump that IEEPA does not authorize tariffs; on March 4 Judge Eaton at the Court of International Trade [ordered CBP to liquidate entries without IEEPA duties and refund roughly $165 billion](https://www.skadden.com/insights/publications/2026/04/insights-april-2026/tariff-refund-mechanism-takes-shape) across 53 million entries and 330,000 importers, through a new CAPE module that was 45% to 80% built as of March 19. The refund is not automatic. The importer of record files the claim, which means the party that paid the surcharge in a 2025 purchase order is not the party CBP pays. Skadden's advice to clients on May 20 was blunt: start ["memorializing refund-allocation mechanics in agreements and purchase orders with commercial counterparties"](https://www.skadden.com/insights/publications/2026/05/us-trade-court-strikes-down-section-122-tariffs) because disputes "are already emerging among importers of record, downstream purchasers and end-users, including class actions." Our [IEEPA refund piece](https://open-factory.com/articles/ieepa-refund-fight/) walks the surcharge trail; the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) lists the instrument and CSMS citation for every rate below. The second event is Section 232. [Proclamation 11021 of April 2, 2026](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states) (91 FR 18201) applies the steel, aluminum and copper duty "to the full customs value of the imported product, regardless of metal content" from April 6, with tiers of 50% for metal articles, 25% for listed derivatives, 10% for products made from US-melted metal, and a transitional 15% combined rate for Annex III grid equipment, transformers and switchgear included, that runs "until 11:59 p.m. eastern standard time on December 31, 2027." Clause (7) of the same proclamation says that from January 1, 2028 the Annex III goods pay the clause (3) rate. A transformer ordered in July 2026 at the 156 to 158 weeks Hitachi Energy quoted Ocala in 2024 enters in 2029, after the 15% tier expires. The third is Section 122: a 10% surcharge from February 24, struck by the CIT on May 7 for three plaintiffs only, expiring by statute on July 24. Eight instruments, each touching a clause a 2022 template does not have. The fourth event is the lead time. ISM's June report, released July 1, put the [average commitment lead time for capital expenditures at 171 days, production materials at 84 and MRO at 48](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-53-3-june-2026-ism-manufacturing-pmi-report-302814991.html), with the Supplier Deliveries Index at 57.4 (slower for the seventh month) and electrical components "in short supply" for the twelfth month running. Our [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes Terrapin's June 2026 slot-reservation ranges: pad-mounts 40 to 65 weeks, 5 to 50 MVA substation units 75 to 110, 15 kV switchgear 52 to 80, and 1,500 to 3,000 kW diesel gensets 50 to 78. The [171-day debrief](https://open-factory.com/articles/171-day-capex-lead-time/) explains why that number understates what an electrical buyer sees. ## Template 1: Specification and Acceptance Write a performance spec and evaluate on total cost, not first cost. BrightRidge, the Johnson City, Tennessee public power utility, issued its [30/40/50 MVA 69:13 kV power transformer specification in March 2026](https://brightridge.com/wp-content/uploads/2026/03/81023.8.pdf) with guaranteed no-load and load losses on the bid form, and the contract amount adjusts if "transformer test losses exceed the guaranteed losses." Its award criteria put evaluated losses first; spare parts prices "will not be used in the evaluation." Mount Pleasant Municipal Utilities in Iowa, buying two 345/69 kV autotransformers with Ameren in June 2024, [required proof of six similar units built in ten years](https://cdn.ymaws.com/members.iamu.org/resource/resmgr/classifieds/ntb/mpmu_345-69kv_autotransforme.pdf) before reading a bid. For machine tools the federal form is the cleanest. The Merchant Marine Academy's [CNC vertical machining center RFQ 6923G224Q000016](https://sam.gov/opp/00ce1872446f4bcab6b680f2299afc35/view) (August 2024) invoked FAR 52.211-6 brand name or equal, listed the salient characteristics an "equal" had to meet, required delivery 185 days after award FOB destination, eight hours of OEM training for up to four instructors, and a one-year warranty, then awarded lowest price technically acceptable. Norfolk Naval Shipyard's [Hurco VMX42i solicitation of July 6, 2026](https://sam.gov/opp/f28262587257429d90e1af9f81bbcea0/view) asked for 24 months on the control, servomotors and spindle, foundation drawings within 60 days, three operators trained for three eight-hour days, and acceptance at destination. Acceptance language matters because Eaton's standard terms, as filed by the City of Santa Clara in January 2026, say [acceptance occurs "when Buyer fails to reject within ten (10) days after delivery"](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf) and that the buyer "waives any right to revoke acceptance." Write acceptance as SAT sign-off with findings closed, and say that delivery and milestone payments are not acceptance. Run the spec through [Quote Check](https://open-factory.com/tools/quote-check/) first so the kVA, sections or kW map to a price band with an n. ## Template 2: Milestones and Payment The research norm is 30% at order, 40% at a middle milestone and 30% at delivery or SAT. Real documents are wider than that. PTI Transformers' proposal on Ocala's autotransformers, incorporated as [Exhibit G to contract ELE/230196](https://ocala.legistar1.com/ocala/attachments/c4a725ca-21c9-4c88-bbeb-7ecd55712a22.pdf), reads: "25% net 30 days at submission of first set of Drawings for approval. 35% net 30 days at time of receipt of Core Steel and Winding Conductor at our works. 30% net 30 days at time of successful completion of testing at our works. Balance 10% payment after the final acceptance, but not to exceed 30 days after delivery." Ocala's next contract, [ELE/240044 with Virginia Transformer](https://ocala.legistar1.com/ocala/attachments/e3d930d9-cbb6-4618-be9e-c40cd7b5929a.pdf) for three 25/33.3/41.6 MVA units at $1,892,556 each (March 2024), pays "20% of the purchase order with the initial order, 30% after receipt of approved drawings, and 40% at confirmed shipment (post-factory test)," with 10% retained "until all required submittals have been approved, not to exceed thirty days from date of delivery." Virginia Transformer had proposed 90% at shipment; the city's paper won. **Every milestone in a good RFQ names the document that releases it, and the 60% a buyer has paid before the factory test is the number to negotiate down.** Utilities with the upper hand pay nothing until the unit is on the pad: Mount Pleasant pays "95% of the Proposal value, upon delivery, assembly, and testing of the Transformers at the Owner's site," with the 5% balance within 30 days of energization and no later than 180 days after delivery, and invites bidders who "require staged progress payments" to attach a schedule. Lake Placid Village Electric's [March 2026 distribution transformer IFB](https://cms2.revize.com/revize/northelba/LPVED%20March%202026%20Distribution%20Transformers%20Bid.pdf) is one line: "Payment terms are 100% at delivery of equipment to site, net 30 days." The mirror of the milestone table is the cancellation table: PTI's runs 5% at order, 20% at engineering complete, 40% at materials ordered, 75% at winding complete and 100% after vapour phase; Irby's 2025 stocking contract with Ocala [reaches 100% two weeks after factory release](https://ocala.legistar1.com/ocala/attachments/b8ee6b52-bc63-41eb-b36a-6b439cc0f25b.pdf). Put both tables in the RFQ so bidders fill them in rather than attach their own. ## Template 3: Schedule and Liquidated Damages The public record shows a consistent cap and a wide spread of rates. BrightRidge sets "$250 per day, not to exceed five percent (5%) of the purchase price of the unit," deducted from money due, after the sentence "Time is of the essence in order for the Owner to comply with established construction schedules." Ocala's Virginia Transformer contract sets "$500 per day," capped at "five percent (5%) of the offending unit contract value," for a unit priced at $1,892,556; at that rate the cap takes 189 days to reach, which is longer than most delays. Ocala's own contract paper on the PTI order carries $1,742 per calendar day for late substantial completion, while PTI's proposal offers "0.5% per week to a maximum of 5% of the purchase price of the unit," followed by the words "No harm no foul." The federal clause, [FAR 52.211-11](https://www.acquisition.gov/far/52.211-11), leaves the daily figure blank, and [FAR 11.501](https://www.acquisition.gov/far/11.501) requires "a reasonable forecast of just compensation" with "a maximum amount or a maximum period." Of ten public documents, two set a per-day rate with a 5% cap, one sets a rate with no cap, one vendor offers 0.5% a week to 5%, and six have no liquidated damages at all. Fayetteville PWC's [February 2024 pad-mount IFB](https://www.faypwc.com/wp-content/uploads/2024/02/bid-PWC2324061.pdf) says "Timely delivery of the Goods is of the essence" and lets the utility terminate; Seguin, Texas [made bidders state "Number of days required to deliver" per item](https://www.seguintexas.gov/DocumentCenter/View/3823/AF-2023-38-Transformers-PDF) and reserved the right to buy elsewhere and charge the difference; Eaton's terms state "Time is not of the essence" and exclude liquidated damages from the seller's liability; Cummins' quote terms on Ocala's [two 500 kW mobile gensets ($594,416, February 2025)](https://ocala.legistar1.com/ocala/attachments/febd0aa7-2119-4a8b-8b6d-e9119499e587.pdf) say delivery dates "are estimated and not guaranteed." **A 0.5%-a-week rate with a 5% cap is exhausted in ten weeks, and the vendor's margin, not your loss, is what set the cap.** Our [liquidated damages piece](https://open-factory.com/articles/liquidated-damages/) prices a week of delay per MW against the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). Ocala's warehouse staff collected four quotes for one 2,500 kVA three-phase pad-mount in March 2024: [a Prolec GE unit nearing completion at Irby for $163,750 in 4 to 6 weeks](https://webapi.legistar.com/v1/ocala/Matters/5600), a JST Power Equipment unit made in China for $102,566 in 32 weeks, the ERMCO unit under its Gresco stocking contract at $69,647 in 156 weeks "with no guarantee of slot availability," and a Hitachi Energy unit through WESCO at $159,905 in "156-158 Weeks after the acceptance of order." A 4-to-158-week spread for one nameplate is why the delivery clause needs two parts: weeks after approved drawings, and a calendar date. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) turns the second number into an energization month. ## Template 4: Tariff and Material Escalation Ocala's autotransformer contract is the cleanest public record of an escalator working exactly as written. PTI's proposal: "Transformer ex-work price to be adjusted based on the FRED Electric Power and Specialty Transformer Manufacturing Index (PCU335311335311)... [Current Index / Base Index] * Quoted Transformer Price where Based Index is the value of FRED in the month of quote, Current Index is the value of FRED 6 months prior to the Delivery Date. Price adjustment upwards or downwards will be applied to the last payment milestone." The base was May 2023 at 415.858. The city [locked the index in February 2026 at 458.462](https://webapi.legistar.com/v1/ocala/Matters/10074), a 10.2% rise, which took the $7,270,000 manufacturing portion to $8,014,800. With freight, oil and field-service escalation the calculated adjustment was $1,015,723.10, and council approved $1,100,000 including an $84,276.90 contingency. The escalator alone added $744,800, or 9.2% of the original contract, on a unit whose delivery slipped from Q2 2026 in the proposal to November 2026. Ocala took the clause because [two of 543 notified firms bid](https://ocala.legistar1.com/ocala/attachments/d53573e5-ff7c-4f3a-a6dc-0bd089e89006.pdf): PTI at $8,070,340 and Virginia Transformer at $16,168,895, after an earlier round came in over budget and the city "resolicited outside of North America." The lesson is to write the index yourself: series, base month, measurement month, deadband, cap, symmetric. Virginia Transformer's Ocala contract does that for inputs other than copper and oil: "abnormal price increases, that raise the overall cost of the unit by two percent (2%) or more" may be passed through but "shall not exceed six percent (6%) of the total cost of the unit," with evidence. Hitachi Energy's 2024 quote had [a price adjustment clause weighted core steel 22%, aluminum 7%, copper 5%, oil 10%, fabricated steel 11% and labor 35%, with no change inside plus or minus 2%](https://ocala.legistar1.com/ocala/attachments/c7ce8f7a-37ba-4f38-a21a-498de85a6eca.pdf). Ocala's 2025 stocking contracts cap adjustments at 1.5% a quarter on the PPI (Irby) and 3% a quarter on Eaton's commodity index (Gresco). The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) has a calculator for the copper leg. Tariffs are the leg most 2024 templates miss. PTI's terms already say that "where there shall be any change in the Taxes after the date of submission of the Seller's tender... payment to the Seller shall be adjusted to reflect such change," a one-way clause. Brookhaven's [15 kV vacuum breaker RFQ 467353](https://sam.gov/opp/f33365932fcd47418ca89bea9d121f01/view) (August 2025) goes the other way: prices "including any import duties, customs, and associated costs." Neither says who gets a refund. **Write four things: the HTS codes and countries of origin the bid price assumes, the duty rate on the bid date, that rate changes flow through at documented cost in both directions, and that any refund belongs to whoever bore the duty.** For Section 232 goods add the melt-and-pour certificate, because the 10% tier depends on it, and for anything entering after December 31, 2027 price the Annex III sunset. A bonded warehouse defers the rate to the withdrawal date. It does not lower a 232 duty. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) gives the baseline rate with its citation for the clause. ## Template 5: Spares and Service Spares and field service are quoted after award at list, which is where the margin the bid gave up comes back. Ocala found this out on the Virginia Transformer order: the contract's [optional pricing](https://ocala.legistar1.com/ocala/attachments/c07cf413-ae0e-407c-81a0-b76486889463.pdf) carried field service testing and assembly at $39,597 per unit as an "estimate only," one witness-test day included and $10,000 a day after, and payment 90% at shipment. On February 6, 2025 the vendor's change request [raised field service to $86,597 and added a $9,000 offloading fee](https://webapi.legistar.com/v1/ocala/Matters/8160) with less than ten days' notice before the first delivery, so the CFO authorized the $56,000 as an emergency payment. PTI priced its required spares lot at $63,340 and an optional list (HV bushing $12,370) at bid, and published its field rates: $2,050 per diem, $190 an hour, trailer detention CAD 250 a day per tank. **Price every spare and every service rate in the bid, hold them for 24 months, and state that spares are not evaluated in the award so bidders cannot buy the job with a cheap list.** Lake Placid's warranty is the buyer's version: 18 months from arrival or 12 months after energization, "whichever is longer," covering freight both ways, removal and reinstallation. Eaton's is the vendor's: the shorter of 12 months from installation or 18 from shipment, removal labor capped at $10,000 per switchgear assembly. Ocala's [Cummins 400 kW natural gas set and 2,000 A transfer switch, $283,376](https://webapi.legistar.com/v1/ocala/Matters/4799) (November 2023) priced the five-year, 2,500-hour warranty as a $4,872 line item, which is how a buyer wants it; the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool includes that into a ten-year cost. ## Template 6: Drawings, FAT and SAT Drawings gate fabrication; factory tests gate shipment; site tests gate the last payment and start the warranty. Mount Pleasant's specification says "Equipment fabrication shall not proceed unless the Bidder has received 'Status-1 Accepted' or 'Status-2 Accepted as Noted' drawings," revised drawings come "two weeks after receipt of MPMU's comments," test plans arrive "at least four (4) weeks prior to factory testing," and the witnessed FAT is "a hold point." BrightRidge gives its engineer "two (2) weeks to review factory test results before shipping" and requires "FOR CONSTRUCTION" drawings six weeks before shipment. Ocala's Virginia Transformer contract adds a line every buyer with a foreign vendor should copy: "Testing shall take place within the United States of America to prevent international travel for personnel." PTI's drawing package arrives "50 - 60 weeks prior to ship date," so design freezes a year and a half before delivery. Across the three documents, only BrightRidge sets a per-day LD, only Mount Pleasant makes the FAT a hold point, and only the Ocala contract carries an index. Each buyer wrote the clause it had been burned on. **The SAT report with findings closed should be the trigger for the last 10% and the warranty clock, and the FAT report the trigger for the shipment milestone, because a payment tied to a document is a payment you can withhold.** Storage belongs here too: Mount Pleasant made bidders price "cost per week for storage at their facility"; Virginia Transformer's contract passes title when a unit enters storage at the city's request and charges a 30% rescheduling fee if the city loses a slot; Cummins bills 1.5% a month after 30 days. Our [switchgear quote anatomy](https://open-factory.com/articles/switchgear-quote-anatomy/) finds the same lines in an MV lineup quote, and the [$60 million deposit piece](https://open-factory.com/articles/slot-economy/) explains why the slot is the asset the clause protects. ## What to Do Monday Fifteen clauses, each already written by a public buyer, with the citation to send a vendor who says "nobody agrees to that." Monday, pull the last equipment PO your plant signed and check it against three lines: does a payment fall before drawings are approved, does the LD cap exceed 5% and does any clause name an HTS code. Then set bid validity at 90 days (Mount Pleasant) rather than the 30 PTI offered, because Section 232 tiers have changed twice since April and a 30-day bid is a 30-day option the vendor holds. Run the spec through [Quote Check](https://open-factory.com/tools/quote-check/) for the band, the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) for the delivery month you should write in, and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) for the baseline rate the escalation clause needs. For a transformer, [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) gives the kVA-to-price band before the engineer over-specifies. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) says which vendors publish lead times and list prices, and those are the ones a clause can be written against. Behind the paywall: the six templates in full as pasteable text, and the **13 redlines** vendors sent back in the public files, from Eaton's "time is not of the essence" to Virginia Transformer's 30% rescheduling fee, with the counter for each, drawn from the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). --- # The 2% Funnel: 474 GW Requested, 9 GW Approved, 5.8 GW Energized in ERCOT URL: https://open-factory.com/articles/ercot-funnel/ Date: 2026-07-08 Power / Take Oncor's 298 GW of requests, Batch Zero, Abbott's August pause and the December audit, who the 3.9 GW are and what they did differently Governor Abbott's [August 3, 2026 letter](https://gov.texas.gov/uploads/files/press/Thomas_Gleeson_Pablo_Vegas_Data_Centers_Directive_Letter_to_PUCT_ERCOT_August_2026_.pdf) to the PUCT and ERCOT put the number at "approximately 474 gigawatts of requests to connect to the Texas grid," more than five times ERCOT's 85,508 MW peak record, about 90% of it data centers. ERCOT's own status table tells the other half: as of [March 26, 2026](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf), 410.6 GW requested, 9.0 GW approved to energize, 5.8 GW energized. **About 2% of what Texas has been asked for has permission to draw power, and about 1.4% draws it.** This piece walks the funnel by status, by utility and by dollar of collateral, then the pause and what the December audit cuts. The queue was [63 GW at the end of 2024](https://www.ercot.com/files/docs/2025/12/02/16.2-System-Planning-and-Weatherization-Update_Revised.pdf), 225.8 GW on November 18, 2025, 410.6 GW on March 26, 2026, [438 GW when the PUCT approved Batch Zero on June 18](https://www.ercot.com/news/release/06182026-puct-approves-ercots), and 474 GW when the Governor wrote his letter. The data center share went from 73% to 87% to about 90%. ERCOT's intake chart shows why: 198 requests arrived in Q1 2026 alone, almost all 750 MW or larger, against 53 in total before 2023. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/), [launched in October 2025](https://open-factory.com/articles/large-load-cost-table-launch/), carries the ERCOT queue as one of its fifteen market legs. The [15-market comparison](https://open-factory.com/articles/what-50mw-costs-15-markets/) put Texas last on security and near the top on energization year for the same reason. ## The Funnel in ERCOT's Own Table Slide 3 of ERCOT's [April 1, 2026 Senate deck](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf) is the whole story in six rows: of 410,618 MW requested through 2030, 293,651 MW (72%) had no studies submitted, 3,241 MW was approved to energize but not operating, and 5,778 MW was observed energized. From November 18, 2025 to March 26, 2026 the no-studies row grew 165 GW while observed energized grew 476 MW. **Energized capacity has grown about 1 GW a year since 2022 (2,634 MW in 2022, 4,834 MW by 2024, 5,778 MW by March 2026) while requests grew 350 GW in sixteen months.** Approved-to-energize capacity is the bottom two rows, 9,019 MW: 2.2% of the March queue, 1.9% of the Governor's 474 GW. ERCOT's planners discount further: its 2025 long-term forecast found [average peak consumption per data center site was 49.8% of the requested MW](https://www.ercot.com/files/docs/2025/04/08/2025_LTLF_Report.docx), 55.4% of officer-letter loads promised for 2024 had energized, and projects ran 180 days late. Those factors are the Texas realization default in the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool. The generation side of the same grid clears at ten times the rate: Berkeley Lab's [Queued Up 2025](https://eta-publications.lbl.gov/publications/queued-2025-edition-characteristics) counts 63.5 GW of ERCOT generation filed in 2000 to 2019 that reached operation against 286 GW filed, 22%. A load queue that clears at 1 to 2% is a survey of intent with a $100,000 entry fee. ## Oncor's 298 GW Oncor told the SEC on [August 6, 2026](https://www.sec.gov/Archives/edgar/data/1193311/000119312526336875/d171086dex991.htm) that its transmission-level large commercial and industrial queue held 737 requests: "approximately 282 gigawatts from data centers and over 16 gigawatts of load from various other industrial sectors," about 298 GW, up from 289 GW at Q1 per the Sempra call as [reported by Utility Dive](https://www.utilitydive.com/news/fate-of-oncors-nearly-300-gw-load-pipeline-unclear-following-texas-data-ce/827303/). ERCOT's by-TSP chart, read to within 2 GW, shows Oncor at about 192 GW of the 410.6 GW tracked in March, roughly 47%. The 106 GW gap is requests Oncor holds that were not yet submitted to ERCOT, which a [January LLWG presentation](https://www.ercot.com/files/docs/2026/01/21/Batch-Zero-Loads-Presentation.pptx) complained about: "Oncor did not submit many of their loads in the process." After Oncor at ~192 GW come AEP Texas at ~40 GW, LCRA TSC at ~32 GW, WETT and CTT at ~24 GW each, CenterPoint at ~23 GW and Lonestar at ~20 GW. Seven other TSPs share about 50 GW. The buyer's numbers are Oncor's next ones: of its 298 GW, [approximately 44 GW](https://www.sec.gov/Archives/edgar/data/1193311/000119312526336875/d171086dex991.htm) is expected to be eligible for Batch Zero, 27 GW as base load and 17 GW as studied load, and that 44 GW includes about 8 GW of existing interconnected load still ramping to its authorized capacity. Oncor holds about $2 billion of collateral against those projects, roughly $45,000 per MW. CenterPoint's numbers rhyme: [17 GW submitted, 14 GW expected eligible](https://www.utilitydive.com/news/centerpoint-energy-14-gw-texas-batch-zero-large-load-earnings/826329/) (10 GW base, 4 GW studied), backed by about $900 million of customer cash and deposits, about $64,000 per MW, on a Houston system whose peak is 21 GW. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) includes both utilities' collateral-per-MW beside Dominion's, AEP Ohio's and Georgia Power's. ## Batch Zero and the $50,000 Per Megawatt Batch Zero is ERCOT's answer to a process built for 40 to 50 large loads at a time. [PGRR145 and NPRR1325](https://www.ercot.com/mktrules/issues/PGRR145) replace the one-at-a-time study with one system-wide study of every 75 MW-plus request that meets maturity criteria. The Board approved it June 2, 2026, the PUCT June 18, effective July 11. Developers filed Form X (bring your own generation) or Form W (accept curtailment) by [July 10](https://www.ercot.com/services/rq/large-load-integration), classification as base, studied or excluded load was due August 7, the security commitment deadline is June 2027, the transmission plan lands fall 2027 and Batch 1 opens summer 2027. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) holds each date with its filing. The cash is set by a different document. The PUCT's proposed [16 TAC 25.194](https://www.gtlaw.com/en/insights/2026/3/texas-senate-bill-6-update-what-data-centers-large-load-customers-should-know-about-proposed-interconnection-standards) (Project 58481, published March 12, 2026, the interconnection-standards leg of the [SB6 roadmap in Project 58317](https://interchange.puc.texas.gov/Documents/58317_5_1526835.PDF)) asks a 75 MW-plus load for a $100,000 study fee (75 to 250 MW) or $300,000 (above 250 MW), $50,000 per MW of security at the intermediate agreement, a second $50,000 per MW that becomes a non-refundable fee at the interconnection agreement, 100% CIAC for every radial line and substation, a site lease running five years past the contracted peak date, and disclosure of every duplicate request by the same company or affiliate anywhere in Texas. Withdraw, or miss a milestone by six months, and 80% of the security goes to the utility's rate base; run five years at contracted demand and the balance comes back. A 75 MW load posts $7.6 million before energization, 100 MW posts $10.1 million, 250 MW posts $25.1 million, 500 MW posts $50.3 million and a 1 GW campus posts $100.3 million, all before a dollar of CIAC. **A 50 MW plant posts nothing, because the rule starts at 75 MW, and a 74 MW plant that later adds 10 MW crosses the line and owes the full schedule.** The [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool prices both sides of that threshold. Under Texas Government Code 2001.027 a proposed rule is withdrawn if not adopted within six months, so adoption or re-proposal is due by September 12, 2026. ## Abbott's Pause and the December Audit The Governor's letter directed the PUCT and ERCOT to "conduct a comprehensive verification and audit of all data centers advancing through ERCOT's interconnection process before any additional data centers are approved to move forward," collecting each project's tax incentives, on-site generation, water, community mitigation and ownership. The trigger he named was the PUCT's water-and-power survey, to which [28 of 377 notified companies responded](https://www.texastribune.org/2026/08/03/texas-data-center-project-audit-greg-abbott/). ERCOT issued a market notice the same day pausing the Batch Zero study and pausing approvals to energize for data centers and crypto facilities of 75 MW or more; medium loads of 25 to 74.9 MW and non-data-center large loads [continue to energize](https://interchange.puc.texas.gov/Documents/59142_52_1675838.PDF). The timeline came at the [August 20 PUCT open meeting](https://www.utilitydive.com/news/ercot-texas-puc-data-center-audit/828472/): RFIs to TSPs and developers from late August through November, site inspections for base loads energizing before April 1, 2027, a verification report filed December 10, a commission open meeting December 17. About [250 to 300 projects totaling roughly 200 GW](https://www.texastribune.org/2026/08/14/texas-data-center-approval-pause-ercot-power-grid/) are in scope. Two numbers from the same deck size the near-term pipeline: 17 large loads with 6,608 MW of five-year peak demand have cleared every gate except approval to energize, and of 290 dynamic models submitted, 18% passed first review. The April 9, 2027 study deadline will slip; ERCOT said so in writing. BNEF, [as reported by Utility Dive](https://www.utilitydive.com/news/texas-data-center-pause-puts-us-pipeline-risk-of-delay-bnef/827185/), puts 49.8 GW under the pause, 20% of the US pipeline, and expects 1.2 GW of ERCOT data center additions through Q1 2027 and 8.25 GW more through 2030, a total above 17 GW. The physical constraint sits under the political one. TSP-reported 2030 load was 208 GW in April against ERCOT's adjusted 138 GW; Ascend Analytics, [via Utility Dive](https://www.utilitydive.com/news/supply-constraints-will-limit-ercot-peak-demand-growth-report/827677/), models 120 GW at a 55.4% success rate and names gas turbines, EPC capacity and high-voltage gear as the limits. We track the same limits: turbine slots gone through 2029 in [the slot economy](https://open-factory.com/articles/slot-economy/), power transformers at [128 to 160 weeks](https://open-factory.com/articles/transformer-160-weeks/) in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/), [90 GW behind the meter announced against 2 GW running](https://open-factory.com/articles/behind-meter-90gw/). A Batch Zero allocation is worth nothing without the GSU. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) puts both dates on one page. ## What to Do Monday If you are under 75 MW, stay there on paper until the rule is adopted, and size the service so a 10 MW addition does not cross the line. The [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool sizes a substation for 74 MW firm with a spare bay. If you are above 75 MW and not in Batch Zero, you are in Batch 1, applications summer 2027, with no data center approval to energize before December 17 at the earliest; put that date in the schedule and price a bridge with the [Genset TCO](https://open-factory.com/tools/genset-tco/) and the [recips vs turbines vs fuel cells](https://open-factory.com/articles/turbines-recips-fuel-cells/) comparison. If you buy capacity rather than build it, the scarce asset in Texas is an energized interconnection, and its owners are mostly bitcoin miners. Behind the paywall: the seven owners of the energized and approved gigawatts, named from their own filings, what each did differently, and our estimate that the December audit leaves **~90 to 110 GW** of the ~200 GW under review standing, with ~20 to 30 GW drawing power by 2030. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) lists the per-site rows. --- # CloudNC and Limitless Labs: AI Is Coming for the CNC Programming Bottleneck URL: https://open-factory.com/articles/cloudnc-limitless-labs/ Date: 2026-07-01 Machinery / Deep dive CloudNC $20M and 1,000 shops, Lockheed as investor and customer, Limitless $20M Series A with Blue Origin and Cadillac F1, 4,790 programmers for 16,619 shops, CAM Assist inside Mastercam and NX, what a seat should save CloudNC and Limitless Labs raised $40 million between them in the twelve weeks from 16 June to 8 September 2026, and both are aiming it at the same 4,790 people. That is how many [CNC tool programmers work in America's machine shops](https://www.bls.gov/oes/2023/may/naics5_332710.htm) in the last industry-level BLS survey, against 64,080 machinists and [16,619 shops](https://www.ibisworld.com/united-states/number-of-businesses/machine-shop-services/645/). CloudNC, founded in London in 2015, says its CAM Assist software now runs in [more than 1,000 machine shops](https://techcrunch.com/2026/09/08/cloudnc-raises-20m-to-automate-manufacturings-most-pressing-bottlenecks/), four in five of them in the United States. Limitless Labs, founded in Tel Aviv in 2024, has taken its CAM Agent from a first pilot to [production at Blue Origin, Cadillac Formula 1 and ISCAR](https://www.calcalistech.com/ctechnews/article/hjod1pr11me) in about eighteen months. This piece covers the bottleneck in numbers, how each company got here, what the two products do inside Mastercam, Siemens NX, Fusion, GibbsCAM and Cimatron, what shops report, what a seat should be worth to a buyer, and behind the wall, the pilot plan, the seat math by shop size and the questions to put to both vendors. A programmer at the [BLS median of $68,120](https://www.bls.gov/ooh/production/machinists-and-tool-and-die-makers.htm) with a 30% load costs $88,556 a year, or $42.58 an hour. If half of those hours are spent in CAM, the CAM hours cost $44,278, and Limitless Labs' production figure of a 50% cut returns $22,139 a year per programmer. CloudNC's published figure of up to 80% returns $35,422. Both assumptions, the load and the CAM share, are marked (A) in the CSV and are the two numbers a buyer should replace with their own. ## The 4,790-Person Bottleneck The shortage that both companies describe is visible in one BLS table. In May 2023 the machine-shop industry, NAICS 332710, [employed 265,030 people](https://www.bls.gov/oes/2023/may/naics5_332710.htm): 64,080 machinists at a median $23.62 an hour, 32,440 CNC tool operators at $23.40, 14,670 welders at $23.76 and 4,790 CNC tool programmers at $31.90. The programmer is the highest-paid production role in the building and the scarcest. Across all industries the programmer's median reached [$68,120 in May 2025](https://www.bls.gov/ooh/production/machinists-and-tool-and-die-makers.htm), and BLS projects [about 30,400 openings a year for machinists and tool and die makers](https://www.bls.gov/ooh/production/machinists-and-tool-and-die-makers.htm) through 2035, most of them replacements for people retiring. 64,080 machinists, 32,440 CNC operators, 14,670 welders, 4,790 programmers. Divide the programmers by IBISWorld's 16,619 shops and the average shop has 0.29 of one. In most shops the owner or the lead machinist programs between setups, and the machine waits while they do. Limitless Labs cites [409,000 unfilled positions and a gap of 1.9 million by 2033](https://www.prnewswire.com/news-releases/limitless-labs-raises-20m-series-a-to-expand-its-physical-ai-foundation-model-and-platform-for-precision-manufacturing-302801451.html) for the wider manufacturing workforce, with a quarter of it aged 55 or older. The [171-day capex lead time](https://open-factory.com/articles/171-day-capex-lead-time/) that ISM reports is partly this: machines arrive before the people who can program them, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) counts the machines, not the programmers. 16,619 shops, 4,790 programmers in them, 30,400 machinist openings a year. **The programmer's seat, not the spindle, is the constraint on a 16,619-shop industry, and two companies have now raised $155 million between them to put software in it.** The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) lists CAM automation alongside robot tending and pallet pools as the three ways a shop adds spindle hours without adding a shift, and [the robot cell piece](https://open-factory.com/articles/robot-61k-cell-150k/) priced the tending option at $144,432 for a cobot cell. ## CloudNC: From a Chelmsford Test Factory to 1,000 Shops Theo Saville came to the problem from metal 3D printing research at Warwick Manufacturing Group, where, [as the company tells it](https://www.cloudnc.com/about-us), he got used to the simplicity of running a printer and was struck by how little of that automation existed in CNC machining. He has an M.Eng in manufacturing and mechanical engineering from Warwick and a background in defence. His co-founder Chris Emery, now chief scientist, has an M.Eng in computing from Imperial College London, worked as a software engineer at Google, and built a Mac application, EventBox, that Realmac Software acquired within a year of its release. Both made the Forbes 30 Under 30 list for manufacturing in 2019, and Saville the MIT 35 Under 35 in 2023. The company they started in 2015 did something unusual for a software firm: it built its own machine shop in Chelmsford, Essex, to test the software on real parts, and [that shop supplies precision components to the UK Atomic Energy Authority](https://www.themanufacturer.com/articles/pulling-your-punches-the-greatest-leap-in-cnc-cam-programming-since-punch-cards/) for fusion diagnostic windows. The money followed the factory. In June 2022 CloudNC closed a [$45 million Series B led by Autodesk](https://www.prnewswire.com/news-releases/cloudnc-raises-45m-to-deliver-autonomous-manufacturing-301571740.html), with Lockheed Martin, British Patient Capital, Atomico, Episode 1 Ventures and QVentures. CAM Assist went into beta in January 2023 and [into the Autodesk App Store in September 2023](https://www.cloudnc.com/news-room/cam-assist-software-release-fusion-360) as a plug-in for Fusion 360. By February 2024 the company had signed [DSI, the largest US reseller of Fusion](https://www.prnewswire.com/news-releases/dsi-to-sell-cloudncs-cam-assist-ai-software-to-north-american-cam-cad-programmers-302065284.html), and CAM Assist was in the hands of hundreds of programmers. On 8 September 2025 [CAM Assist 2.0 shipped](https://www.cloudnc.com/news-room/cam-assist-now-used-by-over-1-000-machine-shops-globally-to-accelerate-cnc-programming-with-ai) and the company announced the 1,000-shop mark. On 8 September 2026, a year later to the day, it announced the [$20 million extension led by Nimble Ventures](https://techcrunch.com/2026/09/08/cloudnc-raises-20m-to-automate-manufacturings-most-pressing-bottlenecks/) with Calculus Venture Capital, Entrepreneur First and LM Ventures, Lockheed Martin's venture arm, taking lifetime funding to $128 million. Headcount is 80. $45 million for CloudNC in 2022 and $20 million in 2026, $4.1 million for Limitless Labs in 2025 and $20 million in 2026. Lockheed Martin is the connecting thread: an investor in CloudNC since 2022 and, [per Tech.eu](https://tech.eu/2026/09/09/cloudnc-lands-20m-to-expand-its-ai-software-for-precision-manufacturing), a customer alongside Major Tool and Machine, the Indianapolis aerospace shop. The company is working with Knox Systems toward FedRAMP authorization, which is what a defence prime's IT department asks for before a cloud tool touches a part file. In [the disclosure scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) both CloudNC and Limitless Labs publish more about their integrations and security posture than most machine-tool builders publish about lead times. ## What CAM Assist Does Inside Mastercam and Siemens NX CAM Assist reads a 3D model and proposes the machining strategy: the operations, the order, the tools from the shop's own library, and physics-based feeds and speeds, with a cycle-time estimate and what the company calls [instant machinability feedback](https://www.cloudnc.com/mastercam). The programmer reviews it, suppresses or edits operations, and hands off to the CAM package's own toolpath engine. It covers 3-axis and 3+2 milling today, not lathes or mill-turns. The company's claim is that it completes [up to 80% of a CAM program in minutes](https://www.cloudnc.com/), takes an hour to learn, and installs the same day. The reach across CAM systems is the part that took four years. CAM Assist runs inside Autodesk Fusion, Mastercam 2025 through 2027, Siemens NX 2306 through 2512, GibbsCAM 2025 and 2026 and SolidCAM from 2025 SP4, with PTC Creo announced as coming. Certifications are SOC 2 Type 2, ISO 27001 and Cyber Essentials Plus, with an AWS GovCloud option for the Mastercam, GibbsCAM and NX versions. Two adjacent tools ship with it: a free Soft Jaw Designer and a Cutting Parameters Explorer. The next product, Quote Agent, applies the same engine to estimating, so that a shop can price a part from the model before anyone programs it. [Quote Check](https://open-factory.com/tools/quote-check/) does the same job for a buyer reading a switchgear quote, turned around here to face the shop that has to write one. Six systems for CloudNC (five shipping, Creo coming), three for Limitless Labs (Mastercam and NX shipping, Creo coming) plus the Cimatron agent it built with Sandvik. Mastercam and Siemens NX, the two most common systems in American aerospace and job shops, are covered by both. A shop on Fusion, GibbsCAM or SolidCAM has one option today, a shop on Cimatron has the other, and a shop on Mastercam or NX can run a head-to-head on its own parts. ## Limitless Labs: The Physics of Cutting as a Foundation Model David Priev and Assaf Peleg are both graduates of Unit 81, the Israeli military's technology unit, where, [per Calcalist](https://www.calcalistech.com/ctechnews/article/hjod1pr11me), they built engineering tools and plug-ins for complex 3D software. Their third co-founder, Shahaf Finder, is a doctoral researcher in machine learning and deep neural networks at Ben-Gurion University. They founded the company in 2024 as LimitlessCNC and came out of stealth on 24 April 2025 with a [$4.1 million seed from Grove Ventures and Meron Capital](https://metrology.news/ai-startup-limitlesscnc-emerges-from-stealth-to-transform-cnc-programming/), 15 employees and one Tier-1 customer running more than 250 CNC machines. The technical bet is a foundation model trained on three things at once: the physics of metal cutting, the geometry of design files, and the constraints of real machines and tool libraries. The agent sits inside the CAM system the shop already owns, reads the part, selects tools from the shop's library, sequences the operations and generates the toolpaths, and hands the result to the programmer for review. Priev's framing, in the [IMTS release](https://www.prnewswire.com/news-releases/enterprise-manufacturers-cut-cnc-programming-time-by-up-to-50-with-limitless-cam-agent-debuting-at-imts-2026-302861954.html), is that the goal "is not to remove programmers from the floor. It is to take the knowledge of your best programmer and scale it across every new part." Fourteen months after the seed, on 16 June 2026, Limitless Labs closed a [$20 million Series A co-led by Dell Technologies Capital and Square Peg](https://www.prnewswire.com/news-releases/limitless-labs-raises-20m-series-a-to-expand-its-physical-ai-foundation-model-and-platform-for-precision-manufacturing-302801451.html), with Grove Ventures, Meron Capital and Kinetica returning, for $27.3 million in total. The customer list had moved from pilots to production: Blue Origin, Cadillac Formula 1, ISCAR and Sandvik in the release, with [GM, Elbit Systems and Samson AG](https://www.limitless-labs.ai/) on the company's site. The production figure is a programming-time cut of up to 50%, and the release gives it in minutes: programs that took two hours done in under ten, multi-operation programs in under five, jobs that took three to five days done in hours. The Sandvik relationship became a product: Cimatron 2027 ships a [CAM Agent built with Limitless Labs](https://dailycadcam.com/cimatron-2027-ai-powered-cam-agent-to-be-debuted-at-imts-2026/) that generates complete 3-axis toolpaths from part geometry and the tool library, in front of Cimatron's roughly 30,000 users, debuting at IMTS in Chicago on 14 to 19 September 2026 at booth 133255, with Limitless itself at booth 237605. The company is ITAR compliant, SOC 2 Type II and ISO 27001 certified, deployable on AWS GovCloud, has opened a US office in Oradell, New Jersey, and plans to double headcount within the year. ## What the Shops Report The published numbers from users are consistent in shape: an hour or two of programming per part becomes ten to twenty minutes, with the programmer spending the saved time on the parts that need judgment. CloudNC's case study of [MSP Manufacturing](https://www.cloudnc.com/case-studies) is the most precise: a part that took the programmer 1.5 to 2 hours got 80% of the way in 7 minutes, and 15 minutes of fine-tuning finished it. Xeon NC reports [an hour becoming 12 minutes](https://www.cloudnc.com/). Dreamz Machining puts it in money: two days a month of programming saved, which it values at $50,000 a year. FJH Group says the software pays for itself in three days. Avalanche Energy, the fusion startup, says it saved three months of programming time. Titan Gilroy of Maki Precision Machining, whose Titans of CNC channel is watched by a good share of the trade, says it cut programming time by 80%. 105 minutes to 22 at MSP Manufacturing, 60 to 12 at Xeon NC, 120 to under 10 at the Limitless enterprise customer. Andy Soos of Bedford CNC, a UK three-axis shop that has used CAM Assist for six months, describes the second-order effect: the [3+2 features let him take on jobs he would have turned away](https://www.cloudnc.com/case-studies/andy-soos-cam-assist), and a guided tool-library import ended the manual library management that every small shop does badly. [The machine-tool piece](https://open-factory.com/articles/machine-tool-record-fewer-units/) found the same pattern from the other side: the average machine sold in 2026 is dearer and more capable than the one it replaces, and the shop's ability to feed it programs is what decides whether the fourth axis ever gets used. ## What a Seat Should Save Both companies price per seat and neither publishes a list price, which is normal for CAM: Mastercam has always sold through resellers, and the public reference points are Autodesk's, at [$680 a year for Fusion and $1,465 a year for its Manufacturing Extension](https://www.autodesk.com/products/fusion-360/extensions). A buyer should ask each vendor for the seat price in writing, and then do the arithmetic in the first chart with their own wage and their own CAM share. A programmer at $88,556 loaded who spends half the year in CAM and gets half of that back returns $22,139, so a seat priced anywhere near the Fusion extension pays for itself in the first month and a seat priced at a fifth of the saving still returns four to one. The value is larger than the wage line suggests, because the programmer is usually the person the machine is waiting for. In [the VF-2 total-cost piece](https://open-factory.com/articles/haas-vf2-true-cost/) we excluded the CAM seat and the operator on purpose, and put the ten-year cost of owning and feeding one machine at $172,523, from the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) machine-tool worked example. A machine idle for a shift while a part is programmed is burning the whole of that line and none of the revenue. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) takes the seat price a shop is quoted and the minutes it measures, so the next shop's quote has a comparison. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) is where we watch machine deliveries. The programmer is the lead time after the machine lands. The wider machinery market makes the case stronger. Machine imports from Japan and Germany [land with about 15% of tariff](https://open-factory.com/articles/tariff-schedule-2026/) on top of list under the current schedule, which the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) applies to a specific quote, so a new spindle is dearer than it was and the software seat that keeps the existing one cutting is comparatively cheaper. Used machines are moving too, as [the auction index](https://open-factory.com/articles/used-cnc-auction-index/) shows, and every used machine that changes hands needs someone to program it in its new shop. Companies like [Nox Metals](https://open-factory.com/articles/nox-metals/) are cutting the wait for material from days to a day and a half, and [SendCutSend](https://open-factory.com/articles/sendcutsend/) turns sheet parts in two to four days. As the material arrives faster, the program becomes the longest step between the drawing and the chip. ## What to Do Monday Pick the ten parts your shop programmed most recently, with the time each one took, and run five through each vendor's trial. Time them with the same programmer, on the same machine, with the shop's real tool library imported, and count the finishing minutes as well as the generation minutes, the way MSP did. Ask both vendors for the seat price in writing, per named user and per floating seat, and for the data terms: where the part files go, whether GovCloud or on-premise deployment is available, and what happens to the learned strategies if you leave. Ask CloudNC when Quote Agent ships and whether your trial can include it. Ask Limitless Labs whether the Creo integration has a date and whether the Cimatron agent will be sold by Sandvik or by them. If you are at IMTS between 14 and 19 September, both are on the floor, and the [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) is the tool to have open when a machine-tool builder at the next booth quotes a delivery. Behind the paywall: the 30-day pilot plan with the measurements that make a trial a decision, the seat math for a **one-programmer, three-programmer and ten-programmer shop**, and the eleven questions to put to both vendors, with the answers we would want to hear. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) form is where the measured minutes go. --- # The R-454B Summer: What the Refrigerant Switch Did to Chiller Prices URL: https://open-factory.com/articles/refrigerant-transition-prices/ Date: 2026-06-30 Cooling / Take AIM Act steps to 60% and 30%, cylinders at $345 to $2,000, Honeywell's 42% surcharge, Chemours pricing up 11%, 563 lb in a 500-ton chiller, EPA's May 2026 reconsideration, the purchase-order clauses A 20 lb cylinder of R-454B cost [$345 in 2021](https://aciq.com/aciq-dealer-program/2025-epa-refrigerant-phase-out/) and traded at [$700 to $2,000 in July and August 2025](https://aristotleair.com/r-454b-refrigerant-shortage-update/). On May 26, 2026 EPA finalised a rule ([91 FR 31284](https://www.federalregister.gov/documents/2026/05/26/2026-10387/), effective July 27) that lets the last R-410A houses be installed and leaves every comfort-cooling chiller on the 700 GWP limit it has had since January 1, 2025. The switch raised prices 10% at list on a residential split, 10% on a Carrier applied chiller in each of March 2024 and March 2025, and 42% on a Honeywell cylinder in April 2025. It did not make the gas a large part of a chiller: the [563 lb charge in a 500-ton Trane RTAF](https://elibrary.tranetechnologies.com/public/commercial-hvac/Literature/Product%20Catalog/RLC-PRC049K-EN_04022022.pdf) is 1.2% of the machine at the $6/lb wholesale HFC price Hudson Technologies reported in May 2026. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists 50 to 70 weeks for a water-cooled chiller. This is about the invoice, not the calendar: the allowance staircase, the 2025 cylinder squeeze and who priced it, six years of OEM list letters, the charge math on a 500-ton machine, what the May 2026 rule left alone, and behind the wall the purchase-order clauses and the ledger of who absorbed and who passed. ## The Step-Down That Made the Market **HFC supply fell 40% on January 1, 2024 and falls another half on January 1, 2029. Every price below sits on that staircase.** The AIM Act caps US production and consumption of bulk HFCs at [90% of baseline for 2022 and 2023, 60% for 2024 through 2028, 30% for 2029 through 2033](https://www.epa.gov/hfcs/frequent-questions-phasedown-hydrofluorocarbons), then 20% and 15%. EPA issued exactly [181,522,990 MTEVe of consumption allowances](https://www.federalregister.gov/documents/2025/11/20/2025-20439/) for 2026 (90 FR 52391, November 20, 2025), the same figure as 2024 and 2025: flat at the 60% step for three years, thirty months before it halves. The chart shows 273.5 MMTEVe of consumption in 2023, 181.5 in 2024 through 2028, 90.8 from 2029, 45.4 from 2036, with production 26% above consumption because exports count. Allowances cover bulk gas, not the charge inside an imported machine, which is why equipment kept shipping while cylinders did not. The producers priced the staircase. Chemours' Thermal and Specialized Solutions segment reported price down 6% in Q1 2025, then [up 4% in Q2](https://www.sec.gov/Archives/edgar/data/1627223/000095017025103204/cc-ex99_1.htm), [up 11% in Q3](https://www.sec.gov/Archives/edgar/data/1627223/000119312525269744/cc-ex99_1.htm) on "stronger Opteon Refrigerant blends aftermarket pricing," [up 10% in Q4](https://www.sec.gov/Archives/edgar/data/1627223/000119312526059564/cc-ex99_1.htm) with "opportunistic Freon refrigerant sales," and [up 11% in Q1 2026](https://www.sec.gov/Archives/edgar/data/1627223/000162722326000010/cc-ex99_1.htm) (May 5, 2026), while Opteon volume grew 65%, 80%, 37% and 12%. Solstice Advanced Materials, spun out of Honeywell on October 30, 2025, reported refrigerant sales [up 22%, 20% and 19%](https://www.sec.gov/Archives/edgar/data/2064953/000206495326000045/exhibit991-earningsrelease.htm) in the three quarters to March 2026. Hudson Technologies, the reclaimer that saw "continued low refrigerant market price" in Q1 2025, said on May 6, 2026 that [HFC prices were "firming above $6 per pound"](https://www.sec.gov/Archives/edgar/data/925528/000110465926056260/tm2613612d1_ex99-1.htm). Legacy gas is rising from a low base. The new gas is falling from a spike. The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) does not yet carry the two producers; it should. ## The Cylinder, Not the Molecule **The 2025 shortage was a shortage of 20 lb steel cylinders with left-hand A2L valves, and it priced R-454B at two to six times list for one season.** In March 2025 a Virginia supply house paid [$730 a jug on the spot market against $310 from its regular distributor](https://www.achrnews.com/articles/164334-contractors-should-plan-ahead-to-obtain-r-454b) after a six-week wait, and Chemours told ACHR News it had "lead times of approximately 10 weeks for new orders." Honeywell had already raised R-454B 15% in February and 8% in March; on April 11 it added a [42% surcharge on every order placed on or after February 15, 2025, plus $4 per pound from April 9](https://www.achrnews.com/articles/164475-honeywell-announces-42-surcharge-on-r-454b), with the surcharge applied to the new price, and Chemours was [reported at $2.85 per pound](https://www.coolingpost.com/world-news/honeywell-adds-42-tariff-surcharge-on-r454b/). Contractors were "already paying upwards of $500 for a cylinder" that week. By July the reported range was [$700 to $2,000 per cylinder](https://aristotleair.com/r-454b-refrigerant-shortage-update/), with one region going "from $485 to $1,050 in weeks"; all of these are single reported quotes, not a distribution, and we say so. The chart shows $17 a pound in 2021, $16 list against $36 spot in March 2025, $25 in April, $24 to $52 in May and June, $35 to $100 in July and August, against a legacy HFC wholesale marker above $6 in May 2026. The mechanism was industrial. Worthington told ACCA's June 18, 2025 panel that OEMs had ["probably 10x'ed them, 5x'ed them"](https://hvac-blog.acca.org/r-454b-shortage-crisis-whats-being-done-and-when-will-it-end/) on cylinder forecasts, one batch failed testing, Honeywell raised blending and down-packing capacity 44%, Arkema entered as a third filler in mid-May, and Worthington [added two crews in Columbus and a third shift in Paducah](https://www.coolingpost.com/world-news/us-industry-surprised-by-r454b-demand/). Watsco said in April that ["everyone in the industry is currently on allocation"](https://www.achrnews.com/articles/164518-r-454b-shortage-creates-challenges-for-distributors) with R-454B at 20 to 25% of Q1 residential sales; by Q2 the [A2L mix was 60%](https://www.sec.gov/Archives/edgar/data/105016/000119312525168635/d941916dex991.htm) and Watsco was converting "over $1 billion of inventory" across 650 branches. HARDI declared [the crisis over on October 13, 2025](https://www.achrnews.com/blogs/17-opinions/post/165373-hardi-on-r-454b-the-crisis-is-over), prices "still elevated" but falling month over month. Volume went the other way. AHRI's December 2025 release puts 2025 central AC and heat pump shipments at [7.75 million, down 20.0% from 9.68 million](https://www.ahrinet.org/news-events/news/ahri-releases-december-2025-us-heating-and-cooling-equipment-shipment-data), air conditioners down 26.2% to 4.10 million. Carrier's read is that the industry averaged 9.7 million units a year from 2020 to 2024, a ["cumulative overage" of about 3.5 million](https://www.achrnews.com/articles/165803-carriers-residential-sales-plunge-nearly-40-in-q4) pulled forward ahead of the switch, 45% of it absorbed in 2025. Volume down a fifth while price rose 10% is the signature of a regulatory transition, not a demand cycle, and the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) treats rooftops and small chillers that way. ## What OEM List Prices Did **Four OEMs announced 46 list increases between January 2021 and April 2026, and the 2025 round was 6 to 10% before tariffs, on top of an A2L platform premium of about 10%.** The premium was declared in advance. Carrier told analysts in October 2024 that ["the base price of R-454B will be 10% higher than the base price of R-410A"](https://www.achrnews.com/articles/163729-equipment-prices-keep-rising-as-r-410a-inventory-decreases) and 15 to 20% over two years; Lennox said "over 10% more"; Trane guided to the "high-single-digit range" and "margin neutral." Then the letters came: Carrier at [6% residential, 8% light commercial, 10% commercial applied from March 1, 2025](https://www.achrnews.com/articles/164568-hvac-price-increase-list), the same three numbers as March 4, 2024; Daikin, Goodman and Amana up to 8 to 10% from April 1, 2025, "not including the impact of recent tariffs"; Lennox with [two increases early in Q2 2025 that lifted its price gains to 7%](https://www.achrnews.com/articles/164566-oems-address-r-454b-shortage-tariffs-in-earnings-calls) plus surcharges of 2 to 10% by series; Trane with increases and surcharges from February to April to offset a $250 to $275 million tariff bill "dollar-for-dollar." The chart shows Trane's largest letter was [up to 18% on commercial equipment effective May 16, 2022](https://www.achrnews.com/articles/146567-trane-technologies-commercial-brands-announce-price-increase), Lennox's 14% commercial in August 2022 and 13% twice in late 2021, Carrier's 12% in January and July 2022, and the 2025 and 2026 letters cluster at 5 to 10%. Seventeen of the 46 rows are ACHR News' own reporting; 29 come from a [class action filed March 20, 2026](https://www.achrnews.com/articles/166007-lawsuit-alleges-oem-coordination-behind-hvac-price-increases) in the Eastern District of Michigan that alleges Trane, Carrier, Daikin, Bosch, Lennox, Rheem and AAON coordinated increases from 2020. Those are allegations, the defendants deny them, and Lennox's CEO said on its [April 29, 2026 call](https://www.achrnews.com/articles/166148-lennox-pushes-back-on-price-fixing-claims-in-q1-call) that equipment is 30 to 40% of an installed job. We use the complaint for dates and percentages only, marked secondary in the CSV. What reached the distributor is in Watsco's numbers: "double-digit pricing realization primarily attributable to the A2L product transition" in Q2 2025, an [11% higher average selling price in Q4 2025](https://www.sec.gov/Archives/edgar/data/105016/000119312526053048/wso-ex99_1.htm) and [9% in Q1 2026](https://www.sec.gov/Archives/edgar/data/105016/000119312526183549/wso-ex99_1.htm). Lennox's light-commercial segment booked ["9% revenue growth from price and mix, driven by the full transition of light commercial products to the new R-454B refrigerant"](https://www.achrnews.com/articles/166148-lennox-pushes-back-on-price-fixing-claims-in-q1-call) in Q1 2026. That is the rooftop answer: the switch added about 10% at list, the OEMs kept it, and the 2026 letters (Lennox 5% commercial and 7% residential, Daikin 7%, Carrier 5 to 6%, Trane 5%) are copper, aluminum and steel. The [Quote Check](https://open-factory.com/tools/quote-check/) spec field for rooftops and chillers includes the refrigerant because a band that mixes R-410A and R-454B quotes is 10% wide before anything else is measured. ## The Chiller Math **A 500-ton chiller holds about 1.1 lb of refrigerant per ton, and at any price the market has printed the charge is a rounding error against the machine.** Trane's Sintesis RTAF catalog lists the 500/520-ton air-cooled screw unit at [275.1 lb on circuit one and 287.8 lb on circuit two](https://elibrary.tranetechnologies.com/public/commercial-hvac/Literature/Product%20Catalog/RLC-PRC049K-EN_04022022.pdf), 562.9 lb of R-134a or R-513A. At Hudson's $6/lb that is $3,377. At the 2021 R-454B cylinder price of $17.25/lb it would be $9,710; at the March 2025 spot of $36.50/lb, $20,546; at the summer 2025 high of $100/lb, $56,290. Against a $275,000 machine, the midpoint of the [$300 to $800 per ton](https://latestcost.com/chiller-cost-per-ton-price-ranges-drivers/) retail cost-guide band for large chillers (a guide, not a bid tab; the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has no 500-ton chiller award yet and says so), the shares are 1.2%, 3.5%, 7.5% and 20.5%. The chart shows 1.2% at $6/lb, 3.5% at the 2021 cylinder price, 7.5% at March 2025 spot, 20.5% at the summer 2025 high; the RTAF does not run R-454B, and we apply those prices to its charge only to show what the spike would have meant at chiller scale. So the [10% Carrier applied increases of March 2024 and March 2025](https://www.achrnews.com/articles/154001-carrier-plans-march-price-increases) and Trane's 18% commercial letter of 2022 are not refrigerant; they are the platform (new compressors, leak detection, controls), copper, steel, tariffs and a backlog that let the OEM choose its price. The refrigerant exposure on a chiller sits in the field: the trim charge at commissioning, top-ups over 20 years, and the reclaim market in 2029. Carrier's own number is that [R-513A costs 2 to 3% of IPLV](https://www.achrnews.com/articles/153191-chillers-go-green-with-low-gwp-refrigerants) against R-134a, which over a chiller's life is worth more than the charge. The May 26, 2026 rule did not touch this class. It [removed the installation deadline](https://www.federalregister.gov/documents/2026/05/26/2026-10387/) for residential and light commercial systems built or imported before January 1, 2025, raised supermarket and remote condensing unit limits to 1,400 GWP until January 1, 2032, and moved the 2026 and 2028 dates to January 1, 2030 for industrial process chillers only where they serve semiconductor manufacturing with a charge of 100 lb or less. Comfort-cooling chillers stayed at 700 GWP, so a new machine runs R-513A (A1, near drop-in), R-1234ze or R-454B (A2L, machinery-room ventilation and detection), or R-1233zd and R-514A (A1, low pressure, centrifugals), and JCI's chiller director put the retrofit rule bluntly: ["safety standards do not permit retrofitting from a Group A1 refrigerant to an A2L refrigerant."](https://www.achrnews.com/articles/153191-chillers-go-green-with-low-gwp-refrigerants) AHRI's president said the rule ["works against basic supply and demand"](https://www.achrnews.com/articles/166226-epa-removes-r-410a-installation-deadline) and legacy refrigerant prices "are likely to rise"; HARDI put the added cost from the commercial-refrigeration extensions near $8 billion. The [Cooling Roadmap](https://open-factory.com/roadmaps/cooling-roadmap/) tracks which chiller platform ships on which refrigerant, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) counts a published charge table in pounds per circuit as disclosure; Trane's RTAF catalog passes. ## What to Do Monday Price the gas separately from the machine. On a rooftop or small chiller quote, ask for the A2L platform premium as its own line and check it against the [Quote Check](https://open-factory.com/tools/quote-check/) band. The OEMs told analysts 10%, and a quote that cannot show where the 10% went is carrying a tariff or metal surcharge inside it, which the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) prices by HTS code. On a chiller, write the refrigerant, the GWP, the factory charge in pounds and the field trim charge into the purchase order, priced at delivery, because the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) shows 40 to 60 weeks on an air-cooled machine and 50 to 70 on a water-cooled one. If the machine is A2L, put machinery-room ventilation and leak detection in the mechanical contractor's scope in the same [RFQ](https://open-factory.com/articles/rfq-capital-equipment/) as the chiller, not a change order at commissioning, under the same [0.5% a week liquidated damages](https://open-factory.com/articles/liquidated-damages/) clause. On a hyperscale hall, buy the chiller owner-furnished as [OFCI vs Design-Build](https://open-factory.com/articles/ofci-vs-design-build-chillers/) lays out, with the trim-charge cylinders on the OEM's order, not the installer's truck. Budget the 2029 step from Hudson's $6/lb floor. The charge-cost row in the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) is chart 4 per ton at that price. Behind the paywall: the nine purchase-order clauses that put refrigerant type, charge quantity, price at delivery, A2L handling and the retrofit path on paper, and the ledger of who absorbed and who passed the 2025 cost, from the OEM that itemised **$50 a unit** for extra gas to the one that says its cost did not move, with the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) row for each. --- # OFCI vs Design-Build: How Hyperscalers Buy Chillers and Colo Developers Do Not URL: https://open-factory.com/articles/ofci-vs-design-build-chillers/ Date: 2026-06-26 Cooling / How-to Owner-furnished 78 weeks out, the 15% markup in public contracts, deposit and storage carry, sales-tax exemptions in three states, warranty clocks, Modine's $165M deposit, the clause set and the reservation playbook A water-cooled chiller over 1,000 tons quoted at [50 to 70 weeks](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026) in May 2026. The developer who lets the mechanical subcontractor buy it after the guaranteed maximum price pays about 15% more for a machine that arrives 20 weeks late. The route the hyperscalers use, owner-furnished and contractor-installed, puts the chiller purchase order on the owner's desk 78 weeks before mechanical completion, strips the two markup layers that public construction contracts print in black and white, and hands the owner the deposit, the storage bill, the warranty clock and the rigging risk in exchange. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) prices both routes for a 20 MW liquid-cooled hall below: what they cost, what public contracts say the markup is, what the owner carries instead, the sales-tax angle in Virginia, Texas and Florida, the warranty and single-point-of-responsibility trade, capacity reservations as the hyperscaler version of OFCI, the calendar, and, behind the wall, the nine clauses that make it work. The long-lead cooling package for a 20 MW hall is $25.9 million at OEM price in both bars; through the mechanical subcontract it becomes $32.8 million after a 15% subcontractor markup ($3.9 million) and a 10% general contractor fee ($3.0 million), while owner-furnished it becomes $27.9 million after $0.4 million of deposit carry, $0.4 million of one month's vendor storage and $1.2 million of owner staff and installer productivity, a saving of $4.9 million or 14.9% of the contractor-furnished price. Every input is in [data/assumptions.csv](data/assumptions.csv) with its source. The ones that are ours are labelled ours. ## Two Routes to the Same Machine **Design-build and CM-at-risk buy the chiller after the price is fixed; OFCI buys it before the design is finished, and the 50 to 70 week lead time is what makes the difference worth $4.9 million.** Under the contractor-furnished route the GMP is set when drawings are 60 to 90% complete, the mechanical subcontract is bought out after that, and the sub places the chiller order. Under OFCI the owner's engineer freezes the basis of design (tonnage, W-class, redundancy, refrigerant), the owner issues the RFQ and the PO in its own name, carries the deposit, and assigns delivery, rigging and installation to the mechanical contractor by a coordination clause. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes Origin Partners' Q2 2026 ranges for the three cooling classes: 40 to 60 weeks for air-cooled chillers over 500 tons, 50 to 70 for water-cooled over 1,000 tons, 30 to 50 for CDUs. Public owners have bought this way for decades. Riverside Public Utilities bought a 15 kV switchgear lineup from Avail Switchgear Systems for [$2,561,551.88](https://riversideca.legistar1.com/riversideca/attachments/e8afd5a4-06af-4cfa-a0d7-1d18104ff259.docx) in June 2024 on its own purchase order, with delivery "scheduled for the first quarter of 2026", a 10% contingency, and $82,293 of the utility's own project management and engineering charged to the work order; construction, factory testing and commissioning were a separate contract, which is the pattern [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/) took apart line by line. The City of Ocala bought three power transformers for [$5,726,568](https://ocala.legistar1.com/ocala/attachments/e3d930d9-cbb6-4618-be9e-c40cd7b5929a.pdf) in March 2024 with 20% down, 30% at approved drawings and 40% at shipment, and a clause under which title passes to the city the day a unit goes into the vendor's storage. The federal government wrote the general rule into [FAR 45.102](https://www.acquisition.gov/far/part-45): "Contractors are ordinarily required to furnish all property necessary to perform Government contracts", and agencies may furnish property only when "the overall benefit to the acquisition significantly outweighs the increased cost of administration". GSA and the Corps of Engineers buy under that clause. A 70-week chiller on a 52-week construction schedule is the textbook exception. ## What the Markup Is, From Public Contracts **Negotiated subcontract markups are private, but the markups that public owners print as ceilings for extra work run 15 to 21% on materials and equipment, 7 to 12% on subcontracted work, and 0% on owner-furnished property.** Caltrans' 2018 Standard Specifications pay force-account work at cost plus [35% on labor, 15% on materials and 15% on equipment rental](https://dot.ca.gov/-/media/dot-media/programs/design/documents/f00203402018stdspecs-a11y.pdf), with "an additional 10 percent markup" when a subcontractor performs the work. WSDOT's 2024 book pays [29% on labor plus 2% for sick leave, 21% on materials and 21% on equipment](https://wsdot.wa.gov/publications/manuals/fulltext/M41-10/SS2024.pdf), and marks up subcontractor work at 12% up to $50,000, 10% to $200,000 and 7% above that. Turner & Townsend's 2025 cost index puts [GC/GRs and GC fees at 10%](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends) of a US data center's construction cost, air-cooled or liquid-cooled. Twelve disclosed rates run from 35% (Caltrans labor) to 0% (FAR 52.245-1, where a contractor that receives Government-furnished property gets "an equitable adjustment" for late delivery and no markup on the property itself). The two materials rates are 15% and 21%, and the three subcontract-layer rates are 7%, 10% and 12%. The trade's rule of thumb, that OFCI [saves about 15% of equipment cost](https://www.irecruit.co/insights/ofci-vs-cfci-procurement-owner-vs-contractor-furnished-equipment-data-centers) at a cost of up to 6% in installer productivity on unfamiliar gear, comes from a recruiter's compilation of practitioner estimates and is used here only because it lands where the public schedules land. The MEP Cost Table uses Caltrans' 15% and 10% as the base case and WSDOT's 21% as the high case. The negotiated buyout markup on a $25.9 million package is the number no one publishes, and the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) exists to collect it. The package itself is the mechanical share of a liquid-cooled hall at Turner & Townsend's $9.8 per watt for Phoenix and Columbus, 33% of $196 million, or $64.7 million, of which we assume 40% is owner-furnishable long-lead equipment: five 1,600-ton machines (four plus a standby), towers or dry coolers, 18 CDUs at 1.3 MW, pumps, plate exchangers and glycol. The chiller line alone at 8,000 installed tons and $1,000 per ton is $8.0 million, inside the [$1,000 to $2,100 per ton](https://latestcost.com/chiller-cost-per-ton-price-ranges-drivers/) equipment range a retail guide printed in December 2025; that is a guide range, not a quote distribution, and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has no chiller award at datacenter scale yet. ## The Owner's Side of the Ledger **Owning the machine early costs $0.52 million a month while it sits in storage, and the deposit interest that everyone worries about is the smallest line.** The OFCI PO goes out at 78 weeks before mechanical completion. A mechanical sub buying after the GMP would place it around week 40. A 30% deposit, the top of the [10 to 30% range](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) manufacturers hold slots for, is $7.8 million out 38 weeks early, and at the [6.75% prime rate](https://fred.stlouisfed.org/graph/fredgraph.csv?id=DPRIME) on June 26, 2026 that is $383,000 of interest. Storage is the expensive line: Cummins' terms in a February 2025 Ocala contract charge "a minimum storage fee in the amount of [one and one-half percent (1.5%) per month](https://ocala.legistar1.com/ocala/attachments/febd0aa7-2119-4a8b-8b6d-e9119499e587.pdf) of the total quoted amount" once the customer fails to take delivery within 30 days, and Eaton's terms make the buyer "reimburse Seller for all costs of storage and handling incurred by Seller after the date that Seller is prepared to make shipment". On $25.9 million that is $388,000 a month, plus $131,000 of interest on the 90% already paid. Carrying cost starts at $0.4 million (the deposit carry) and climbs $0.52 million a month, crossing the $2.6 million a 10% GC fee is worth at month 5 and the $6.9 million the two Caltrans layers are worth at month 13. OFCI pays for itself as long as the machine does not live in the vendor's yard, so a delivery-date clause with a free deferral window matters more than the deposit percentage. The other two owner costs are staff and installer friction: Riverside charged 3.2% of the switchgear price to its own engineering and project management, $832,000 on $25.9 million, and the 6% productivity hit on installation labor of a quarter of equipment value is $388,000. ## The Sales Tax Angle **Three states exempt data center chillers from sales tax, and each one attaches the exemption to the purchaser of record, which is the owner under OFCI and the contractor under design-build.** Virginia's [58.1-609.3(18)](https://law.lis.virginia.gov/vacode/title58.1/chapter6/section58.1-609.3/) exempts "computer equipment or enabling software ... including chillers and backup generators" for a data center that makes "a new capital investment ... of at least $150 million" and creates 50 jobs at one and a half times the prevailing wage, through June 30, 2035, and "applies to the data center operator and the tenants of the data center if they collectively meet the requirements". Texas exempts "electricity; an electrical system; a cooling system; an emergency generator" and "a mechanical, electrical or plumbing system that is necessary to operate any tangible personal property" for a [qualifying data center](https://comptroller.texas.gov/taxes/data-centers/) of 100,000 square feet with $200 million invested and 20 jobs, for 10 years at $200 to $249 million and 15 years above $250 million; the buyer presents Form 01-929 carrying both the facility's and its own registration number. Florida's [212.08(5)(r)](https://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0200-0299/0212/Sections/0212.08.html) exempts property used "to construct, outfit, operate, support, power, cool, dehumidify, secure, or protect a data center" once the owners and tenants have invested $150 million and the site carries 100 MW of critical IT load. | State | Instrument | Threshold | Exempt (quoted) | Who claims | Period | |---|---|---|---|---|---| | Virginia | Va. Code 58.1-609.3(18) | $150M investment, 50 jobs at 1.5x prevailing wage ($70M and 10 jobs in distressed localities) | "computer equipment and enabling software ... including chillers and backup generators" | Operator and tenants, collectively | Jul 1, 2010 to Jun 30, 2035 | | Texas | Tax Code 151.359; Form 01-929 | 100,000 sq ft, $200M over 5 years, 20 jobs; large project 250,000 sq ft, $500M, 40 jobs | "an electrical system; a cooling system; an emergency generator; ... a mechanical, electrical or plumbing system" | Registered owner, operator or occupant with its certificate | 10 years ($200M to $249M), 15 years ($250M+), 20 years large project | | Florida (private) | Fla. Stat. 212.08(5)(r) | $150M cumulative investment, 100 MW critical IT load, 1 MW per tenant, within 5 years | "to construct, outfit, operate, support, power, cool, dehumidify, secure, or protect a data center" | Certificateholder (owner or tenant) | No sunset in the text | | Florida (public works) | Fla. Stat. 212.08(6)(b); Rule 12A-1.094; DOR GT-800067 | Entity buys direct: its PO, its invoice, its payment, title on delivery, risk of loss at purchase | Materials and supplies that become part of a public facility | Governmental entity with a Certificate of Entitlement; contractor purchases taxable | Current | *Three states exempt data center cooling gear, and each ties the exemption to the purchaser of record. Source: Va. Code 58.1-609.3; Texas Comptroller; Fla. Stat. 212.08; Florida DOR GT-800067. Data: [05-sales-tax-by-state.csv](data/05-sales-tax-by-state.csv)* Florida's public-works rule is the cleanest statement of why the purchaser matters. Statute [212.08(6)(b)](https://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0200-0299/0212/Sections/0212.08.html) says the government exemption "does not include sales of tangible personal property made to contractors employed directly to or as agents of any such government", and the Department of Revenue's [GT-800067](https://floridarevenue.com/Forms_library/current/gt800067.pdf) spells out the owner-direct-purchase test: the entity issues its own purchase order and exemption certificate, pays the vendor directly, is invoiced directly, "must take title to the materials upon delivery to the jobsite" and "must assume the risk of loss of the materials at the time of purchase". That is OFCI defined by a tax collector, and at Florida's [6% state rate](https://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0200-0299/0212/Sections/0212.05.html) it is worth $1.55 million on a $25.9 million package before county surtax. A developer holding a Texas or Virginia certificate should read the same test into its design-build contract: the exemption is the registrant's, and a subcontractor's PO in the subcontractor's name is a form the state may or may not look through. ## Warranty, Rigging and the Single Throat **Every OEM warranty we opened runs from start-up or from shipment, whichever comes first, so a chiller bought 78 weeks out and stored for a season can arrive at commissioning with half its warranty gone.** Carrier's commercial terms cover defects "within [one (1) year from the date of initial operation or eighteen (18) months from date of shipment](https://www.shareddocs.com/hvac/docs/1003/Public/0B/CCS_TERMS_CONDITIONS_SEPT18.pdf), whichever is earlier". Eaton's switchgear terms are the shorter of 12 months from installation or 18 from shipment. Hitachi Energy quoted Ocala 18 months from delivery or 12 from commissioning, whichever first. Cummins is the outlier a buyer should copy: its "3-year warranty begins at the successful completion of startup and testing in lieu of acceptance or substantial completion". Under design-build the shipment clock is the sub's problem inside a one-year building warranty; under OFCI it is the owner's, and Ocala's transformer contract adds the trap that the warranty is void for "storage ... not in accordance with Contractor's specifications". Rigging is the second place the risk moves. Cummins' terms say "offloading, handling, and placement of Equipment and crane services are the responsibility of Customer"; PTI Transformers' proposal to Ocala allows [two free hours for offloading](https://ocala.legistar1.com/ocala/attachments/c4a725ca-21c9-4c88-bbeb-7ecd55712a22.pdf) and then charges CAD 150 an hour, CAD 250 a day per tank in trailer detention and CAD 1,300 a day in layover. A 1,600-ton centrifugal outside a hall whose pad is not cured is those numbers times the trucks. The single-point-of-responsibility argument for design-build is real and it is this: when an OFCI item "arrives late, damaged, or incomplete, the responsibility lies with the owner, not the contractor", as one contractor put it in the iRecruit compilation, and the federal clause says the same thing in regulatory prose, the installer gets "an equitable adjustment" when Government-furnished property misses its date. Integration risk follows the same line. The mechanical contractor still owns piping, flush, glycol fill, controls integration and commissioning, and a coordination clause that names the OEM start-up prerequisites (Cummins requires "a signed site check sheet confirming readiness" and two weeks' notice) is what keeps two throats from becoming none. ## Capacity Reservations: OFCI on the Factory **The hyperscaler version of OFCI skips the machine and reserves the factory: Modine took a $165 million deposit in March 2026 against more than $4 billion of cooling products for 2027 to 2029, a 4.1% deposit on three years of a plant's output.** Modine's 10-K describes "a long-term capacity agreement with one of its strategic data center customers" and "a [$165.0 million up-front deposit](https://www.sec.gov/Archives/edgar/data/67347/000110465926066795/mod-20260331x10k.htm) from the customer to support the Company's investments necessary to meet the planned sales volume", in the same filing that admits "demand for certain key components is currently outpacing supplier capacity". [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/) covered the gas-turbine version in May: GE Vernova ended Q1 2026 with [44 GW of firm backlog and 56 GW of slot reservation agreements](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000063/gevpressrelease1q26.htm), signed 19 GW of new reservations in the quarter, and tells shareholders it makes "capacity expansion decisions and supply commitments based on demand forecasts, orders, slot reservation agreements, and deposits". | Buyer or seller | Date | Terms | Source | |---|---|---|---| | Modine, one strategic data center customer | Mar 2026 | $165.0M up-front deposit on more than $4B of cooling products, CY2027 to 2029 (4.1% floor) | Modine 10-K FY2026, May 27, 2026 | | GE Vernova gas turbines | Q1 2026 | 44 GW firm backlog plus 56 GW slot reservation agreements; 19 GW of SRAs signed in the quarter | GE Vernova 8-K, Apr 22, 2026 | | Vertiv, deferred revenue (current) | Mar 31, 2026 | $2,461.8M vs $1,095.1M a year earlier (+125%); 0.93x quarterly sales | SEC XBRL; Vertiv 10-Q | | Munters, US colocation customer | Dec 31, 2025 | SEK 2.1B of CRAHs, CDUs and chillers; deliveries Q4 2026 through Q1 2028 | Munters press release | | Switchgear, transformer and genset OEMs | Jun 2026 | Slots held for 10 to 30% of equipment cost; reservation compresses apparent lead time 8 to 16 weeks | Terrapin Construction Group | | Utility transformer buyer (City of Ocala) | Mar 2024 | 20% with order, 30% at approved drawings, 40% at shipment; 30% rescheduling fee; title passes into storage | Ocala agreement ELE/240044 | *Capacity reservations are OFCI with the deposit on the factory instead of the machine. Source: [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) (compiled from SEC filings, Munters, Terrapin, City of Ocala; as of June 2026). Data: [06-capacity-reservations.csv](data/06-capacity-reservations.csv)* The cooling OEMs' balance sheets show who is paying early. Vertiv's current deferred revenue, the customer cash it holds ahead of shipment, was [$2,461.8 million on March 31, 2026](https://data.sec.gov/api/xbrl/companyconcept/CIK0001674101/us-gaap/ContractWithCustomerLiabilityCurrent.json) against $1,095.1 million a year earlier, up 125%, and equal to 0.93 quarters of sales; its 10-K says the $15.0 billion backlog "is expected to be shipped within the next [12 to 18 months](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm)". Munters booked a [SEK 2.1 billion](https://www.munters.com/en-us/news-media/press-releases/2025/munters-wins-record-orders-for-data-center-equipment-with-a-total-value-of-2.1-bsek) order for CRAHs, CDUs and chillers from a US colocation company on December 31, 2025 with deliveries "scheduled to begin in Q4 2026 and continue through Q1 2028"; [Chillers Are the 60-Week Problem](https://open-factory.com/articles/chiller-60-week-problem/) laid out the five Munters windows and the Trane, Carrier and JCI backlogs behind them in March, and the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) tracks the quarterly ratios. A colocation developer cannot write a $165 million check for a factory. It can write the three things the check buys: a named slot, an indexed price, and a delivery window it controls, and the paid section says how. ## The Calendar **Counting back from mechanical completion with the Monitor's published ranges, the chiller RFQ goes out 66 to 86 weeks ahead, the PO and deposit at 58 to 78, and the CDU PO at 36 to 56, all of them before a GMP signed at week 40.** The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) builds the calendar from the published range plus a submittal, FAT and freight buffer of 8 weeks for chillers and 6 for CDUs, the same buffers the sibling article used for its 78-week headline. The milestone windows inside the PO come from the public specifications above: drawing submittals [4 to 6 weeks after order](https://ocala.legistar1.com/ocala/attachments/c7ce8f7a-37ba-4f38-a21a-498de85a6eca.pdf) in Hitachi Energy's quote, a test plan four weeks before FAT and FAT as "a hold point" in [Mount Pleasant's](https://cdn.ymaws.com/members.iamu.org/resource/resmgr/classifieds/ntb/mpmu_345-69kv_autotransforme.pdf) transformer spec, and two weeks for the engineer to review test results before shipping in [BrightRidge's](https://brightridge.com/wp-content/uploads/2026/03/81023.8.pdf). Nine windows: RFQ 66 to 86 weeks out, PO 58 to 78, approval drawings back 50 to 74, CDU PO 36 to 56, CDU drawings 30 to 52, FAT 8 to 14, ship 6 to 8, chillers set 4 to 6, CDUs set 2 to 6, against a GMP at roughly week 40 that a design-builder cannot honestly sign earlier because the drawings are not done. The three chiller windows and both CDU windows open before the contractor who would buy the equipment exists as a party to the job. The developer can still have the design-builder's single throat for everything after week 40, and not for the chiller. The [Cooling Roadmap](https://open-factory.com/roadmaps/cooling-roadmap/) does not show the 50 to 70 week range closing before 2028. ## What to Do Monday First, run the hall through the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) with the Monitor's 50 to 70 weeks for water-cooled chillers and 30 to 50 for CDUs, and if the chiller RFQ is not dated before the land closes, the energization month is wrong. Second, split the mechanical scope in the CM-at-risk or design-build agreement now, before the GMP: chillers, towers or dry coolers, CDUs and the pump-and-exchanger kit as owner-furnished, with the contractor's fee on that equipment written at the FAR's 0% or a stated handling percentage, not left to the buyout. Third, write the RFQ with the [capital-equipment template](https://open-factory.com/articles/rfq-capital-equipment/): slot week on the face of the PO, deposit no higher than the 20% Ocala paid, a PPI index clause on the last milestone instead of a firm price the OEM will not give, and [liquidated damages at 0.5% a week](https://open-factory.com/articles/liquidated-damages/) against the slot. Fourth, put the deferral window in: 30 days free, then a stated dollar-per-week storage rate under the OEM's own storage instructions, title and the owner's builder's-risk attaching on the storage date, and a warranty that starts at commissioning. Fifth, register for the state exemption before the PO, not after, and run any imported machine through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) and the quote through [Quote Check](https://open-factory.com/tools/quote-check/) before the deposit leaves. Behind the paywall: the nine-clause set for the owner's chiller and CDU purchase order and the mechanical subcontract, each with the public contract it is copied from, and the five terms a 20 MW developer can lift from the hyperscaler capacity-reservation playbook, starting with the **4.1%** deposit Modine's customer paid for three years of a factory, all built on the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # Grainger's 2017 Confession: The Year a Distributor Admitted Its List Prices Were Fiction URL: https://open-factory.com/articles/grainger-confession/ Date: 2026-06-21 Procurement / Deep dive 450,000 web-priced SKUs, 1.7 point U.S. margin hit, 811 Hot List SKUs at 50% off, Fastenal 62% through vending and eBusiness, Amazon Business $35B, how to read a contract price file On April 18, 2017, W.W. Grainger told investors that the first quarter "[clearly fell short of our expectations, driven primarily by the stronger than anticipated customer response to our U.S. strategic pricing actions](https://www.sec.gov/Archives/edgar/data/277135/000027713517000013/gww8kex991q12017.htm)." It had cut list prices across the board, put new web prices on about 450,000 SKUs, and watched U.S. gross margin fall 1.7 points in one quarter because "customers with access to lower pricing bought more than company expectations." **A $10 billion distributor had discovered that its own list prices were the reason customers were leaving, and the only fix was to admit they had never been real.** This piece walks through how an MRO price is actually built (list, web, contract tier, vending), what the 2017 reset showed about the spread between those prices, what has happened to Grainger's and Fastenal's gross margins in the nine years since, what 143 public contract discount lines and an 811-SKU Sourcewell price file say about where the fiction is thickest, and how to read the contract price file your own distributor sends you. Company gross margin was [42.4% in 2015](https://www.sec.gov/Archives/edgar/data/277135/000027713517000007/gww20161231-10k.htm), 40.6% in 2016, [39.3% in 2017](https://www.sec.gov/Archives/edgar/data/277135/000027713518000007/gww20171231-10k.htm), 38.7% in 2018 and 38.3% in 2019, then 35.9% in the pandemic year and back to [39.1% in 2025](https://www.sec.gov/Archives/edgar/data/277135/000027713526000011/gww-20251231.htm); the first quarter of 2026 printed [40.0%](https://www.sec.gov/Archives/edgar/data/277135/000027713526000053/gww-20260331.htm), the first 40 since 2016. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) lists these series next to the public contract discounts they were built to defend, and the sibling piece on the [50-SKU price basket](https://open-factory.com/articles/mro-price-basket/) measures the spread between the four channels on live SKUs. Here the job is the history and the structure. ## How an MRO Price Is Built Grainger's CEO described the model in March 2017 at his own trade show, to a table of trade editors: "[We've had a very high list, less discount model](https://web.archive.org/web/20191113171336/https://www.inddist.com/home/blog/13773803/grainger-ceo-details-changes-to-salesforce-strategy-pricing). Large customers particularly valued getting discounts off of lists. We will continue to have that model, but our list prices are too high right now." That is the whole architecture in three sentences. A broadline MRO distributor carries one published list price per SKU (Grainger calls the current version the Catalog Reference Price, or CRP, in its [OMNIA Partners contract documents](https://cdn.sanity.io/files/h427pqeg/production/2d08d01ad8dd1821c9f3c165c38590b683ef6f8e.pdf)), sells to walk-in and small accounts near that list, and sells to everyone else at list minus a percentage that is negotiated per customer, per product category and often per SKU. The tiers stack like this. List is the anchor nobody pays. Web price, introduced by Grainger on about [450,000 SKUs in early 2017](https://www.sec.gov/Archives/edgar/data/277135/000027713517000013/gww8kex991q12017.htm) and on the entire assortment from [August 1, 2017](https://www.sec.gov/Archives/edgar/data/277135/000027713517000020/gww063020178kex991.htm), is a second published number set to be, in Macpherson's words, "in the ballpark" of Amazon and McMaster-Carr. Contract price is list minus a category discount for accounts with a signed agreement. The public versions of these agreements run from 5% to 57% off, and we get to them below. Under the contract sit SKU-specific "market basket" prices for the 300 to 1,500 items a plant actually buys repeatedly, which is where the real money moves and which no cooperative publishes. And under all of it sits the vending machine or the KeepStock bin, whose contents are priced by a contract that gets reviewed annually if the buyer remembers to ask. *[Data: distributor-revenue-gross-margin.csv](data/distributor-revenue-gross-margin.csv)* Seven channels, roughly $80 billion of annual MRO volume between them, and one of them (McMaster-Carr) shows one price per SKU to everybody. Grainger booked [$17.9 billion at a 39.1% gross margin in 2025](https://www.sec.gov/Archives/edgar/data/277135/000027713526000011/gww-20251231.htm); Fastenal [$8.2 billion at 45.0%](https://www.sec.gov/Archives/edgar/data/815556/000081555626000009/fast-20251231.htm); Genuine Parts' Industrial segment, which is Motion, [$8.9 billion](https://www.sec.gov/Archives/edgar/data/40987/000004098726000003/gpc-20251231.htm) with no margin disclosed and a separation from the automotive business announced February 17, 2026; Applied Industrial [$4.56 billion at 30.3%](https://www.sec.gov/Archives/edgar/data/109563/000010956325000080/ait-20250630.htm) for the June 2025 fiscal year; MSC [$3.77 billion at 40.8%](https://www.sec.gov/Archives/edgar/data/1003078/000100307825000123/msm-20250830.htm). Amazon Business reported "[over $35 billion in annualized gross sales](https://press.aboutamazon.com/2025/8/from-solopreneurs-to-fortune-500-amazon-business-now-powers-8-million-organizations-providing-even-more-value-selection-and-convenience-to-its-customers)" and more than 8 million organizations in August 2025, a gross-merchandise number rather than revenue. The spread between Applied's 30.3% and Fastenal's 45.0% is not a spread in what a bearing costs. It is a spread in how much of the bill is service, stocking and the pricing opacity that pays for both, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades exactly this kind of silence on the manufacturer side. ## The Reset, Quarter by Quarter The pricing actions were "[first described in November 2016](https://www.sec.gov/Archives/edgar/data/277135/000027713517000013/gww8kex991q12017.htm)" and "primarily implemented in January and February" of 2017. The release lists three: "adjusting list prices across the board to make it easier for large customers to consolidate their purchases"; "introducing new web prices on about 450,000 SKUs to drive medium and large noncontract customer acquisition and growth"; and "negotiating large customer contracts with more competitive pricing for infrequently purchased items," with the note that "most large customers already receive very competitive pricing on routine items through their contracts." That last clause is the confession inside the confession: the accounts with contracts were already paying something close to market on the items they bought weekly, and the list-price fiction was being charged to the small customer and to the large customer's infrequent items. *[Data: grainger-us-segment-quarterly.csv](data/grainger-us-segment-quarterly.csv)* U.S. segment price was a 4 point drag in [Q1 2017](https://www.sec.gov/Archives/edgar/data/277135/000027713517000013/gww8kex991q12017.htm) against 4 points of volume, 4 against 5 in [Q2](https://www.sec.gov/Archives/edgar/data/277135/000027713517000020/gww063020178kex991.htm), 5 against 7 in [Q3](https://www.sec.gov/Archives/edgar/data/277135/000027713517000036/gww093020178kex991.htm) and 5 against 11 in [Q4](https://www.sec.gov/Archives/edgar/data/277135/000027713518000003/gww123120178kex991.htm), and U.S. gross margin fell 1.7, 1.3, 1.9 and 1.8 points in those quarters. The company cut its 2017 EPS guide from $11.30 to $12.40 down to $10.00 to $11.30 in April and pulled the pricing work scheduled for 2018 into the third quarter, because "the data provided confidence that the pricing actions were successful." Volume answering price at better than one for one, inside one year, is the cleanest published evidence that MRO buyers knew they were overpaying and were shopping elsewhere in proportion. **The 2017 10-K puts the full-year cost of honesty at 1.7 points of U.S. gross margin and $17 million of company gross profit dollars on $10.4 billion of sales.** The filing's language is careful: the actions "[included adjusting list price and introducing new lower web prices on the entire business assortment, which drove faster growth in 2017 through share gains with existing customers and acquisition of new customers](https://www.sec.gov/Archives/edgar/data/277135/000027713518000007/gww20171231-10k.htm)," and U.S. gross margin fell "primarily driven by price deflation exceeding higher volume in response to pricing actions." The risk-factor section added a sentence that had not been there in 2016: "To remain competitive, the Company must be willing and able to respond to market pressures, including pricing, whether widely available or negotiated under a contract." In March 2017 Macpherson said the company "[started its pricing moderation process in January, and plans to be finished by March 2018](https://web.archive.org/web/20191113171336/https://www.inddist.com/home/blog/13773803/grainger-ceo-details-changes-to-salesforce-strategy-pricing)," with 400,000 items already carrying a web price that existing customers opted into "with a two-question survey." The tail of the reset shows up in the next two 10-Ks. U.S. gross margin fell another [0.7 points in 2018](https://www.sec.gov/Archives/edgar/data/277135/000027713519000010/gww20181231-10k.htm), 0.5 of it an accounting reclassification of KeepStock costs into cost of goods, with price at minus 2 points in the first quarter and plus 1 point by the third and fourth. Then [2019](https://www.sec.gov/Archives/edgar/data/277135/000027713520000007/gww20191231-10k.htm): U.S. gross margin down 0.4 points "reflecting the impact of contract renegotiations and customer mix." That sentence is the second act. Once the web price existed, every contract customer could see the gap between its negotiated discount off list and the number a stranger got on the website, and the renegotiations ran through 2019. The [Rockwell pricing letters](https://open-factory.com/articles/rockwell-250bps/) describe the same dynamic from the manufacturer side: the discount is a rebate structure, and once the list moves, every downstream number has to be re-based. ## What the Reset Revealed About Dispersion The reset did not publish a dispersion table, but public contracts do, and a distributor's own public price files do. Sourcewell, the Minnesota cooperative whose contracts are usable by every U.S. public agency, awarded Grainger contract [091422-WWG](https://www.sourcewell-mn.gov/cooperative-purchasing/091422-wwg) in late 2022 (Grainger scored 871 of 1,000 points and ranked first of 20 proposers in the [proposal evaluation](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/Solicitation%20Documents/Prop.%20Eval.-Facility%20MRO%20091422.pdf); Motion was second, Fastenal third). The contract lets the supplier publish a "[Hot List](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/00004572/Contract%20Documents/Grainger%20Contract%20091422.pdf)" of items "at discounts greater than those listed in the Contract," and Grainger's current Hot List file, dated December 2025, prints two prices for each of 811 SKUs: a "US Contract Reference Price" and a Hot List price. *[Data: grainger-sourcewell-hotlist.csv](data/grainger-sourcewell-hotlist.csv)* We parsed all 811 lines ([data](data/grainger-sourcewell-hotlist.csv)): the median Hot List price is 50.4% below the reference price, the mean 48.7%, and the range runs from 4.9% (a disposable food tray at $57.82 against $60.86) to 89.3%. A Dayton A49 V-belt lists at $12.69 and sells to a Sourcewell member at $3.51; a 20-yard roll of vinyl electrical tape lists at $1.64 and sells at $0.80; a Fluke single-dot IR thermometer lists at $178.19 and sells at $148.35. Rule of thumb from the file: the reference price on a commodity consumable is roughly double what a public agency pays, and the reference price on a branded instrument is about 20% above it. *[Data: grainger-sourcewell-hotlist-by-brand.csv](data/grainger-sourcewell-hotlist-by-brand.csv)* **Grainger's private labels sit furthest under their own reference price: Dayton motors and belts at a 74% median discount across 41 SKUs, Condor safety at 67% across 23, Tough Guy janitorial at 64% across 49, against 3M at 44% and Georgia-Pacific at 31%.** Private label was [22% of U.S. sales in 2017](https://www.sec.gov/Archives/edgar/data/277135/000027713518000007/gww20171231-10k.htm), and the brands on which the distributor controls both the cost and the list are the brands where the list is most elastic. That is not a criticism of the Hot List, which is a real discount a real agency receives. It is a measurement of how much room the reference price was built with. A national-brand item has a manufacturer's suggested list that a distributor cannot inflate without looking odd next to a competitor. A Dayton belt has no such constraint. *[Data: mro-contract-discounts.csv](data/mro-contract-discounts.csv)* Category discounts are the layer above the Hot List, and four public schedules give 143 lines of them ([data](data/mro-contract-discounts.csv)). Fastenal's [Sourcewell schedule](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/00004574/Additional%20Documents/091422-FAS%20Pricing%20Information.pdf) runs 23% to 57% off catalog across 37 categories: fasteners 57%, wipers and rags 40%, filters, liners and hand protection 35%, electrical, plumbing, lighting and safety 30%, most tool categories 25%, corded and cordless power tools 23%. Grainger's OMNIA Partners contract through the [City of Tucson](https://cdn.sanity.io/files/h427pqeg/production/2d08d01ad8dd1821c9f3c165c38590b683ef6f8e.pdf) (240078-01, January 2025 to December 2027) runs 5% to 40% off CRP: air filters 40%, fasteners 32%, electrical 23%, cleaning and lighting 22%, plumbing 20%, motors, safety and power transmission 19%, and a floor of 5% on hardware, hydraulics, office supplies and raw materials. The [Region 4 contract](https://cdn.sanity.io/files/h427pqeg/production/bf98a2598ff46a2309a5243fb0676712ad00245d.pdf) (240808) is 2 to 3 points thinner in most categories, and the [University of California contract](https://cdn.sanity.io/files/h427pqeg/production/f3f29a8f6b1ea8a18464497e42a4a844ca40afe5.pdf) (2018.000207, running to June 2028) is thinner still, with 24 of 35 categories at exactly 5%. Every one of the three Grainger documents describes itself as a "full catalog contract with discounts on entire product portfolio, minimum discount of 5%." A 57% fastener discount at Fastenal and a 5% fastener discount at Grainger UC are not evidence that Fastenal is eleven times cheaper on a bolt. They are evidence that the two catalogs' list prices for a bolt are unrelated numbers, which is why the [50-SKU basket](https://open-factory.com/articles/mro-price-basket/) prices the bolt and not the discount. The spread within one distributor is as wide as the spread between them: Grainger gives Tucson 32% off fasteners and the University of California 5%, on the same reference price, under contracts a few years apart. The UC schedule also shows what a contract buys beyond the percentage: Fastenal's companion [UC agreement](https://cdn.sanity.io/files/h427pqeg/production/bf420121e42aca79a30b98a0bacfa139143116b6.pdf) adds a 3% Onsite incentive, a 5% growth incentive on prior-year sales, a 1% to 2% e-commerce rebate, 2% for early payment and up to 10% more on "Green" or exclusive-brand items. Five rebates stacked on a category discount is how a 25% contract becomes a 40% net, and none of it appears on the invoice. ## Vending Is a Pricing Contract Fastenal's answer to the transparency problem was not a web price. It was to put the store inside the customer's building and price the contents on a contract. The company reports "weighted FMI devices," machine-equivalent units of FASTVend vending machines and FASTBin scale bins, and the installed base went from [83,951 at the end of 2020](https://www.sec.gov/Archives/edgar/data/815556/000081555622000003/ex_99112312021earningsrele.htm) to 92,874, 102,151, 113,138, 126,957 and [136,638 at the end of 2025](https://www.sec.gov/Archives/edgar/data/815556/000081555626000003/ex_99112312025earningsrele.htm), then 137,702 at [March 31, 2026](https://www.sec.gov/Archives/edgar/data/815556/000081555626000019/ex_99103312026earningsrele.htm). Signings were 25,892 in 2025 against a 2026 goal of 28,000 to 30,000. The older device counts run further back on a different definition: [40,775 installed machines at the end of 2013](https://www.sec.gov/Archives/edgar/data/815556/000081555614000016/fast1231201310-k.htm), 55,510 at the end of 2015, "over 86,000 devices in the field" at the end of [2017](https://www.sec.gov/Archives/edgar/data/815556/000081555618000010/fast1231201710-k.htm) and about 110,700 at the end of [2020](https://www.sec.gov/Archives/edgar/data/815556/000081555621000008/fast-20201231.htm), about 15,000 of them lockers for one retail customer. *[Data: fastenal-vending-digital.csv](data/fastenal-vending-digital.csv)* Fastenal began reporting its "Digital Footprint," sales through FMI devices plus the eBusiness sales that are not FMI billings, in the first quarter of 2021 at [34.8% of sales](https://www.sec.gov/Archives/edgar/data/815556/000081555622000003/ex_99112312021earningsrele.htm) (39.1% on the later definition); it was 46.4% in the fourth quarter of 2021, 52.6% a year later, 58.1% in Q4 2023, [62.2% in Q4 2024](https://www.sec.gov/Archives/edgar/data/815556/000081555625000048/ex_99112312024earningsrele.htm) and $5,094.0 million or 62.1% of sales for the full year 2025, with a stated goal of [66% during 2026](https://www.sec.gov/Archives/edgar/data/815556/000081555626000009/fast-20251231.htm). FASTBin and FASTVend sales alone went from $1,064 million in 2020 to $2,675 million in 2025, 18.8% to 32.6% of the company. MSC is on the same road with [29,611 vending machines and 411 in-plant programs at August 30, 2025](https://www.sec.gov/Archives/edgar/data/1003078/000100307825000123/msm-20250830.htm), up from 27,003 and 342 a year earlier, and 63.8% of its orders placed digitally. **A vending machine is a pricing contract with a lock on it: the price of every item inside was set when the machine was signed, and the buyer's next look at a competitor's price happens when the contract renews, not when the glove is dispensed.** Fastenal's own 10-K explains the margin logic without saying so; its gross margin has fallen every year but one since 2016, from [49.6% to 45.0%](https://www.sec.gov/Archives/edgar/data/815556/000081555626000009/fast-20251231.htm), and each year's filing attributes the slide to "a continued shift toward larger customers, which typically generate higher volume at lower gross margins." The devices are how a 45% distributor holds onto large accounts that a 39% distributor would otherwise take, and the price inside the machine is where the 6 points of difference live. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) accepts vending contract price files for exactly this reason: the machine's price list is the only place a plant's true MRO net price is written down. ## What Has Happened to Margins Since The nine years since the reset divide into three periods on the filings: two years of digestion, a pandemic mix hole, and an inflation round that gave the margin back with interest. Grainger was at 40.6% in 2016 and 39.1% in 2025; Fastenal at 49.6% and 45.0%. Fastenal's decline was mix, by its own account: fasteners are its highest-margin line and shrank as a share of sales, and Onsite and vending customers are large and priced accordingly. Grainger's path went through [35.9% in 2020](https://www.sec.gov/Archives/edgar/data/277135/000027713521000007/gww-20201231.htm), "primarily driven by lower margins from COVID-19 pandemic-related product sales," then a [$118 million pandemic inventory adjustment](https://www.sec.gov/Archives/edgar/data/277135/000027713523000007/gww8kex991q42022.htm) in 2021, then the recovery. The High-Touch Solutions N.A. segment, which is the old U.S. and Canada business, has run 41.7% to 41.8% for three straight years and printed [42.6% in Q1 2026](https://www.sec.gov/Archives/edgar/data/277135/000027713526000053/gww-20260331.htm), two points above the whole company's 40.6% of 2016, the last full year before the reset. *[Data: price-realization-grainger-fastenal.csv](data/price-realization-grainger-fastenal.csv)* The inflation round is where the reset paid off. Grainger's 2021 10-K attributes [2.3 points of daily sales growth to price](https://www.sec.gov/Archives/edgar/data/277135/000027713522000012/gww-20211231.htm) and 10.1 to volume; the [2022 10-K](https://www.sec.gov/Archives/edgar/data/277135/000027713523000014/gww-20221231.htm) puts price at 10.6% and volume at 8.7%, and gross margin rose 2.2 points; [2023](https://www.sec.gov/Archives/edgar/data/277135/000027713524000011/gww-20231231.htm) was 4% price and 5% volume with another 100 basis points of margin. Fastenal disclosed product pricing worth [440 to 470 basis points in Q4 2021](https://www.sec.gov/Archives/edgar/data/815556/000081555622000003/ex_99112312021earningsrele.htm) and 350 to 380 in Q4 2022, then "[did not institute any broad pricing actions through 2023](https://www.sec.gov/Archives/edgar/data/815556/000081555624000009/fast-20231231.htm)" or 2024. A distributor with a believable web price could raise it 10% in 2022 and have the increase stick, because the customer could see the same increase on the competitor's website. A distributor that had never fixed its list would have spent 2022 renegotiating every contract discount instead. Tariffs restarted the clock. Fastenal "[implemented pricing actions to address the incremental tariffs beginning in the second quarter of 2025](https://www.sec.gov/Archives/edgar/data/815556/000081555626000009/fast-20251231.htm)," worth 170 to 200 basis points for the year, [310 to 340 in the fourth quarter](https://www.sec.gov/Archives/edgar/data/815556/000081555626000003/ex_99112312025earningsrele.htm) and about 350 in the [first quarter of 2026](https://www.sec.gov/Archives/edgar/data/815556/000081555626000019/ex_99103312026earningsrele.htm). Grainger's 2025 gross margin fell 30 basis points on "[tariff-related inflation which caused unfavorable price / cost timing and last-in, first-out (LIFO) inventory valuation headwinds](https://www.sec.gov/Archives/edgar/data/277135/000027713526000004/gww8kex991q42025.htm)," and then High-Touch sales in Q1 2026 grew 10% with "[equal contribution of 5% for both volume and price](https://www.sec.gov/Archives/edgar/data/277135/000027713526000053/gww-20260331.htm)." Five points of price in a quarter is the 2022 playbook rerun on a tariff bill, and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) is where to check whether the adder on a given SKU's country of origin comes anywhere near five points; the [copper panel](https://open-factory.com/articles/copper-tariff-panel/) and the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) do the same for the wire and cable category, where the metal moves faster than any list. *[Data: grainger-segment-sales.csv](data/grainger-segment-sales.csv)* The growth split tells you where the money is going. Grainger's Endless Assortment segment, Zoro in the U.S. and MonotaRO in Japan, sells [13 million and 29 million products](https://www.sec.gov/Archives/edgar/data/277135/000027713526000011/gww-20251231.htm) respectively at one web price each, at a 29.9% gross margin, and grew 16% in 2025 to $3.6 billion; High-Touch, the contract business at 41.7%, grew 2% to $14.0 billion. In the first quarter of 2026 Endless Assortment grew 20% and High-Touch 10%, half of it price. The single-price businesses are a fifth of Grainger and grew eight times as fast as the contract business in 2025, and Grainger's 2026 guide of a [39.2% to 39.5% gross margin](https://www.prnewswire.com/news-releases/grainger-reports-results-for-the-fourth-quarter-and-full-year-2025-302676819.html) on $18.7 to $19.1 billion of sales is what that mix shift costs each year. ## Ten Years of Moves *[Data: pricing-timeline.csv](data/pricing-timeline.csv)* Fifteen dated moves, nine of them Grainger's. The table runs from the November 2016 description to the Q1 2026 quarter in which Grainger took 5 points of price and Fastenal about 350 basis points, and the middle rows are the ones a buyer should keep: the March 2018 target date by which Macpherson said "[everything set on a web price that's reasonable](https://web.archive.org/web/20191113171336/https://www.inddist.com/home/blog/13773803/grainger-ceo-details-changes-to-salesforce-strategy-pricing)," the 2019 contract renegotiations, and the 2023 to 2024 pause in which Fastenal took no broad price and Grainger's High-Touch growth was "[primarily due to volume](https://www.sec.gov/Archives/edgar/data/277135/000027713525000010/gww-20241231.htm)." Those two years are the baseline. Any contract signed in 2023 or 2024 was priced off a list that had not moved much in five years, and every price increase since has been justified by a tariff on a country of origin that the invoice does not name. Amazon Business belongs on the timeline because it is the reason the reset happened when it did. Macpherson's March 2017 line was that "[you can't keep digital marketing if the price you feature is always higher than everyone else's](https://web.archive.org/web/20191113171336/https://www.inddist.com/home/blog/13773803/grainger-ceo-details-changes-to-salesforce-strategy-pricing)," and everyone else in 2017 meant Amazon Business, launched in 2015, and McMaster-Carr, which had never had a list-minus model to abandon. Amazon's August 2025 release sells "[business-only pricing, and quantity discounts](https://press.aboutamazon.com/2025/8/from-solopreneurs-to-fortune-500-amazon-business-now-powers-8-million-organizations-providing-even-more-value-selection-and-convenience-to-its-customers)" alongside spend analytics, which is the same structure in new clothes: a published price and a discount off it that depends on who you are. The difference is that the published price is the one an employee with a corporate card would pay, and that constraint is what Grainger's 2017 web price was built to match. ## What to Do Monday Pull the invoices for your top 300 MRO SKUs from the last twelve months and ask the distributor for the contract price file that produced them: category, discount basis (list, CRP or catalog), percentage, and the SKU-level market basket. Then price the same 300 SKUs on the distributor's own public website while logged out, on Zoro, on McMaster-Carr and on Amazon Business, the way the [50-SKU basket](https://open-factory.com/articles/mro-price-basket/) does at smaller scale. Where your contract price is above the logged-out web price, you are paying for the fiction Grainger admitted to in 2017. Where the category discount is under 10%, as it is on 24 of 35 lines in the University of California schedule, you are on the floor and should ask for the [Tucson](https://cdn.sanity.io/files/h427pqeg/production/2d08d01ad8dd1821c9f3c165c38590b683ef6f8e.pdf) numbers by category. Take the vending contract out of the drawer and check the renewal date against the tariff clauses added since Q2 2025, because the [IEEPA refund](https://open-factory.com/articles/ieepa-refund-fight/) question applies to a glove with a China origin in a vending machine as much as to a switchgear lineup. Put the country of origin per SKU into the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) before accepting any "tariff-related" line, and run any capital-equipment quote that rides on the same distributor relationship through [Quote Check](https://open-factory.com/tools/quote-check/), because the discount structure that inflates a bolt inflates a motor starter by the same mechanism. MRO commitments run about [48 days ahead](https://open-factory.com/articles/171-day-capex-lead-time/) on ISM's panel; capital equipment runs 171, which is why the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) tracks the second and this piece the first. If you are a public agency, school district, hospital or university, the Sourcewell and OMNIA schedules above are yours to use as written, and the Hot List file is the price you should already be paying on 811 SKUs. If you are a private plant, the same three documents are the floor to quote back at your distributor, because a cooperative with no volume commitment got 32% off fasteners and 40% off air filters. A cooperative price that beats your contract price on a category is the fastest evidence of overpayment a buyer can put in front of a sales rep, and it costs nothing to download. Behind the paywall: how to read the contract price file line by line, the **five categories** where the public contracts put contract price furthest from list, with the percentages from all 143 lines in the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), and the negotiation script that gets a distributor to re-base a category discount onto its own web price instead of its reference price. --- # Reading a Switchgear Quote: The Eleven Line Items That Hide the Schedule URL: https://open-factory.com/articles/switchgear-quote-anatomy/ Date: 2026-06-13 Procurement / How-to Base price, copper escalator, GOES surcharge, testing, freight, storage clause, cancellation terms, LD schedule, factory acceptance witness, spares, drawings Riverside Public Utilities paid $2,561,551.88 for one 15 kV metal-clad switchgear lineup in June 2024, and the unit was still in the factory eighteen months later. The [board memo](https://riversideca.legistar1.com/riversideca/attachments/e8afd5a4-06af-4cfa-a0d7-1d18104ff259.docx) that approved the purchase put delivery in the first quarter of 2026; the [next year's memo](https://riversideca.legistar1.com/riversideca/attachments/c543fa97-4254-4b6d-8413-8103f7cc5f04.docx) moved it to December 2025. The price was one line. The schedule was spread across ten others that nobody at the board meeting read aloud. **A medium-voltage switchgear quote has eleven line items, and four of them move the delivery date more than the price does.** This piece walks one real public quote line by line, prices each line from public bid tabs and one filed set of vendor terms, shows why liquidated damages at 0.5% a week cover a fifth of a plant's diesel bill, and ends with the four clauses to send back before the PO goes out. The [Open Factory Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds every bid Riverside received for a single 15 kV metal-clad lineup across four procurements since 2018: 13 bids, from $1.34M to $3.21M including California sales tax. The low bid went from [$1,343,878 in 2018](https://riversideca.legistar1.com/riversideca/attachments/28dbba7d-a43b-456d-99cc-3148790fb517.pdf) (4 bids) to $2,509,063 in 2023 (2 bids), $2,561,552 in 2024 (3 bids) and [$2,254,388 in 2025](https://riversideca.legistar1.com/riversideca/attachments/99a5fd23-cb4f-4c0e-b009-d7390ecf7818.pdf) (4 bids), a 68% to 91% rise on the same product at the same substation class, and the spread between low and high bid widened from 9% to 42%. ## The Quote We Walk Through Riverside's Bid SUB-871 is the cleanest public record of an MV switchgear purchase we have found: the city posts the [award recommendation](https://riversideca.legistar1.com/riversideca/attachments/37cb493a-5440-49a2-a51f-c84cbe1795ce.pdf), the board memo and the presentation on Legistar. Five vendors were invited on December 27, 2023. Three bid on February 13, 2024: Avail Switchgear Systems of Fulton, Missouri at $2,561,551.88, Crown Technical Systems of Fontana at $2,719,096.74, and Myers Power Products of Ontario, California at $3,176,380.88. Eaton and RESA Power were invited and did not submit. The purchasing manager signed on April 23. The board voted on June 24, 2024, nineteen weeks after the bids opened. The [award line](https://riversideca.legistar1.com/riversideca/attachments/37cb493a-5440-49a2-a51f-c84cbe1795ce.pdf) says "Includes Sales Tax" and nothing else. There is no copper adder, no freight line, no testing line, no spares line, because a public bid form collapses all of them into the lump sum. The eleven items are always there. The bid form decides whether you can see them. The board did see three numbers around the award: the $2,561,552 price, a 10% contingency of $256,155, and $82,293 of the utility's own project management and engineering, for a $2,900,000 work-order increase. The engineer's estimate was $1.45M against a $1.34M low bid in 2018, $2.10M against $2.51M in 2023, and $2.60M against $2.25M in 2025. The [2023 memo](https://riversideca.legistar1.com/riversideca/attachments/ae7986e2-8ee3-4675-8f9c-b8dbb6c3040f.docx) explains the miss in one honest paragraph: "The last time RPU procured a switchgear was in 2018. Due to the time gap, our most recent historical records for pricing this type of equipment are no longer relevant." That is the gap the [Quote Check](https://open-factory.com/tools/quote-check/) tool exists to close: a 15 kV metal-clad spec in, the Bid-Tab band and the [Lead-Time Monitor's](https://open-factory.com/models/lead-time-monitor/) published lead time out, with n and date on each. The switchgear is also a minority of the project it sits in. Riverside's [2018 Freeman Substation work order](https://riversideca.legistar1.com/riversideca/attachments/28dbba7d-a43b-456d-99cc-3148790fb517.pdf) shows where the rest of the money goes. Of the [$5,463,000 work order](https://riversideca.legistar1.com/riversideca/attachments/28dbba7d-a43b-456d-99cc-3148790fb517.pdf), the Crown switchgear was $1,343,878 (25%), the transformer $1,400,000 (26%), construction plus factory and field testing and commissioning $1,295,121 (24%), project management and engineering $600,813 (11%), contingency $529,612 (10%), rentals and miscellaneous material $240,000 (4%) and the relay panels $53,576 (1%). A week of switchgear delay holds up the other 75% of that budget. ## Line Items One to Three: Base Price, Copper, Steel The base price is the only line the buyer reads, and it comes with a clock. Riverside's 2025 award recommendation for [Bid SUB-876](https://riversideca.legistar1.com/riversideca/attachments/99a5fd23-cb4f-4c0e-b009-d7390ecf7818.pdf) prints "BID PRICE GOOD THROUGH: MAY 22, 2025" on a bid opened January 21: seventeen weeks of validity, and the board approved on May 12, ten days before expiry. Switchgear Power Systems of Winneconne, Wisconsin won at $2,254,387.50; Avail bid $2,403,320.63, Crown $2,567,340.62 and Myers $3,205,870.61 (n=4). A quote that expires before your capital committee meets is an option the vendor holds. Behind the base price sit two materials clauses. The public version of the seller's position is in the [Eaton (Cooper Power Systems) terms and conditions](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf) that Santa Clara's Silicon Valley Power filed with its council in January 2026 for a $1.5M switchgear authorization. Section 2(a): "Unless prices are bid or quoted as 'firm,' Seller reserves the right to invoice at prices in effect at the date of shipment, regardless of any prior bid and whether notice was received by Buyer." That single sentence is the copper escalator and the steel surcharge in one, with no index, no base date and no cap. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) gives the index and base date a clause needs. Our piece on [copper at $6.77 in a 200-amp panel](https://open-factory.com/articles/copper-tariff-panel/) shows what an uncapped version costs. The switchgear PPI stood at [394.3 in April 2026](https://fred.stlouisfed.org/series/PCU335313335313), up 14.0% from 345.8 a year earlier and 78% from 221.5 in January 2021; it rose 32% between Riverside's 2023 award and April 2026 alone. The IMF's global copper price was [$12,891 a tonne in April 2026](https://fred.stlouisfed.org/series/PCOPPUSDM), $5.85 a pound, up 40.5% year on year. The tariff layer is now three proclamations deep: copper derivatives at 50% on copper content since August 1, 2025 under [Proclamation 10962](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states); Section 232 duty on the full customs value since April 6, 2026 under [Proclamation 11021](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states), with a 15% transitional tier for "fixed industrial machinery and power equipment" through December 31, 2027; and [Proclamation 11032](https://www.federalregister.gov/documents/2026/06/04/2026-11314/further-adjusting-the-tariff-regimes-for-imports-of-aluminum-steel-and-copper-into-the-united-states) of June 1, 2026, which kept that tier and widened it. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) lists each instrument with its Federal Register citation. The decomposition of the PPI into copper, GOES, labor and margin is in [why your switchgear costs 50% more than in 2021](https://open-factory.com/articles/switchgear-50-percent/). The same repricing shows up on the smallest public switchgear tab we track. The City of Dover, Delaware buys S&C PME-9 pad-mounted switchgear for stock every year or two and [posts every unit bid](https://www.cityofdover.gov/bid-tabulations). Dover's winning unit price was [$20,703 in November 2019](https://www.cityofdover.gov/media/Purchasing/Tabulations/FY%2020%20Info/20-0007CS%20Bid%20Tab.pdf) (5 bids), $19,667 in February 2021 (2 bids), $27,689 in February 2022 (2 bids), $28,846 in February 2023 (4 bids) and [$31,771 in September 2024](https://www.cityofdover.gov/media/Purchasing/Tabulations/FY%2025%20Info/25-0004WH%20Bid%20Tab.pdf) (4 bids), up 53% in five years with the high bid at $39,723. Dover also shows what a thin market does to a bigger lineup: its Garrison Oak metal-clad switchgear bid of [September 2021 drew no award](https://www.cityofdover.gov/bid-tabulations) and was re-bid in June 2022 with one submission, from Siemens. ## Line Items Four and Five: Testing and Freight Testing is two lines pretending to be one. Routine production tests are in every base price. A witnessed factory acceptance test, where your engineer stands in the plant while they close the breakers, is priced as travel and a test slot, and the slot is the part that matters. The buyer-side practice is to make the signed FAT report the trigger for the middle payment, in the 30/40/30 pattern that [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/) lays out: [30% at order, 40% at drawings approved or FAT, 30% on delivery or SAT](https://www.elevexcapital.com/milestone-based-equipment-payments). Vendor paper does the opposite. Eaton's Section 4 makes acceptance automatic "when Buyer fails to reject within ten (10) days after delivery," and Section 3(b) lets the seller "ship up to forty (40) days in advance of shipping date," which is how a lineup lands on a site that has no pad yet. Freight looks like the smallest line and hides two costs. [Eaton's Section 3(a)](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf) quotes "F.O.B. destination" and then adds that "Charges for shipping may not reflect net transportation costs paid by Seller." A 15 kV metal-clad lineup ships in shipping splits with a crane at each end. The rigging is never in the switchgear quote and always in the contractor's. Riverside's 2018 work order carried $240,000 of "equipment rentals" and miscellaneous material next to a $1.34M switchgear, 18% of the gear's price, on a line the memo does not split. ## Line Items Six and Seven: Storage and Cancellation Storage is where a schedule problem turns into an invoice. [Eaton's Section 3(a)](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf): "Buyer shall reimburse Seller for all costs of storage and handling incurred by Seller after the date that Seller is prepared to make shipment." Read that with Section 3(b), "Delivery and shipping dates are approximate," and the asymmetry is complete: the seller may be late for free and the buyer pays to be late. On a 2024 order with delivery in December 2025, the site work Riverside had planned for fiscal 2025/26 has to be ready in the same month the gear is, or the meter starts. Cancellation is the other half of the same trade. [Section 7](https://santaclara.legistar1.com/santaclara/attachments/8a0f5d95-e164-43df-b632-cff5e2ac0672.pdf): "Orders shall not be subject to cancellation or modification either in whole or in part without Seller's written consent and then only with terms that will reimburse Seller for reasonable termination charges, including all progress billings and all incurred direct manufacturing costs." There is no schedule of charges, so the cancellation cost is whatever has been billed plus whatever the seller says it spent. The same section lets the seller cancel "if the manufacture or sale of the goods is or becomes technically or economically impractical," with a refund as the only remedy. Contracts like this are why the buyer's contingency line is growing. Riverside's board approved a 10% contingency of $256,155 on the 2024 award and a 20% contingency of $450,507 on the 2025 award, and 25% on the companion WEG transformer, which the [May 2025 memo](https://riversideca.legistar1.com/riversideca/attachments/c543fa97-4254-4b6d-8413-8103f7cc5f04.docx) explains as cover for "potential import duties or tariffs that may arise after the quoted price." A public buyer parked $450,507 of ratepayer money against a price clause it could have fixed with one sentence. ## Line Item Eight: The LD Schedule Against the Diesel Bill Liquidated damages are the one line most buyers ask for and most vendors strike. Eaton's Section 6 excludes "LIQUIDATED DAMAGES, OR LOSS OF USE" outright and caps liability at the purchase price; Section 3(b) adds "Time is not of the essence." Where LDs survive, the [market norm is 0.5% of contract value per week, capped at 2.5% to 5%](https://www.sciencedirect.com/topics/engineering/liquidated-damage), with [EPC-style caps at 5% to 10%](https://pathnovo.com/blog/liquidated-damages-epc-contracts) because that is the vendor's margin. On Riverside's $2,561,552 award that is $12,808 a week, and the cap arrives at week 5, 10 or 20. A 2 MW plant that opens on rented diesel because its switchgear is late burns 12,855 gallons a week at 50% load, using EIA's [10,512 Btu/kWh heat rate for petroleum internal-combustion units](https://www.eia.gov/electricity/annual/html/epa_08_02.html) and [137,381 Btu per gallon](https://www.eia.gov/energyexplained/units-and-calculators/); at the [$5.21 retail diesel price of June 8, 2026](https://fred.stlouisfed.org/series/GASDESW) that is $66,974 a week in fuel alone, 5.2 times the LD accrual, before the rental, the operator or one lost shift. The 10% cap, $256,155, is spent by week four of fuel. The full cost-of-delay arithmetic, including the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tool's temporary-power line, is in [Liquidated Damages at 0.5% a Week](https://open-factory.com/articles/liquidated-damages/). The point for the quote reader is simpler: an LD clause is a schedule signal, not an insurance policy, and its real value is that a vendor who accepts 0.5% a week from a firm date has told you the date is real. ## Line Items Nine to Eleven: FAT Witness, Spares, Drawings The FAT witness line is usually missing because the buyer is expected to supply it. The clause that matters is notice: how many days before the test the vendor must tell you, how many seats you get, and who pays for a retest. A [FAT sign-off starts the shipping clock](https://kneat.com/article/the-difference-between-a-fat-and-a-sat/), so a vendor who can schedule the FAT on seven days' notice controls when your 40% milestone falls due. Spares are the reverse problem: the "recommended spares list" arrives after award, priced against nothing. The fix is to get the spare breaker and the spare relay priced on the bid form at the unit price of the ones in the lineup; [the spares list is where the margin hides](https://open-factory.com/articles/recommended-spares-margin/) precisely because it is priced alone. Drawings are the schedule killer, and they are not a line item at all. Every MV switchgear lead time is quoted from approval of drawings, not from the PO. The vendor submits approval drawings some weeks after order; the buyer's engineer marks them up; the factory does not release material until they come back clean. Every week in that loop is a week the vendor does not own, and most quotes say nothing about its length. Riverside's own calendar shows how much of the total the factory then holds. For SUB-871 the invitation-to-opening period ran 7 weeks, opening-to-award 19 weeks, and award-to-scheduled-delivery 77 weeks, 103 weeks in all; for SUB-876 the segments were 9, 16 and about 131 weeks to a fourth-quarter 2027 delivery, roughly 156 weeks. Riverside's own [June 2024 presentation](https://riversideca.legistar1.com/riversideca/attachments/34f6d7f6-a219-4872-95f4-9220c7bb1edf.pdf) called switchgear "long lead-time (Average 70 weeks)"; the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) carries [Terrapin's June 2026 range of 60 to 80 weeks for 15/27 kV metal-clad](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) against about 24 weeks before 2020, and the [ISM's May 2026 report](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-may-2026-ism-manufacturing-pmi-report-302786165.html) has electrical components in short supply for an eleventh month, supplier deliveries at 60.6 and the average capex commitment lead time at 171 days. Nothing in those numbers is the drawing loop. A buyer who returns drawings in three weeks instead of eight has bought five weeks that no LD clause can buy back. The [171-day capex lead time](https://open-factory.com/articles/171-day-capex-lead-time/) piece shows how rarely that gets planned. ## The Eleven Items on One Page Seven of the eleven lines move price, and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands will tell you whether the base price is inside the range other buyers paid for a lineup of that class. Four move the schedule, and no price band can check them. The public record is thin on the seven: of the thirteen public MV switchgear procurements we pulled for this piece, from Riverside, Dover, Leavenworth and Chelan County PUD, not one itemized copper, freight, storage, testing or spares on the tab, which is why the clause language above comes from a filed set of vendor terms rather than from a bid form. That is also the reason [Quote Check](https://open-factory.com/tools/quote-check/) asks for the quote PDF and not just the total. ## What to Do Monday Pull the vendor's terms and conditions before the quote, not after; Santa Clara's Eaton filing shows what the default position looks like, and it is the same on most OEM paper. Write the RFQ so the bid form has eleven lines, with the copper index, the storage rate, the cancellation schedule and the drawing dates as blanks the vendor must fill, and run the total through [Quote Check](https://open-factory.com/tools/quote-check/) against the [Bid-Tab bands](https://open-factory.com/models/bid-tab-price-book/) before the capital committee meets, because the bid is only good for about seventeen weeks. Put the FAT report, not the calendar, on the 40% milestone. Return drawings inside fifteen business days and put that promise in the PO next to the vendor's delivery week, so the week is counted from the PO. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) will tell you which month the PO must leave for the plant to energize on time. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lists the same lines for a 20 MW hall. Behind the paywall: the order in which to negotiate the four schedule items, so the concession you need most is not the one you spend your goodwill on first; the exact clause language for drawings, FAT, LDs and storage, with the numbers filled in for a **$2.56M lineup**; and a line-by-line redline of the cancellation clause in the Eaton terms Santa Clara filed, from "Seller's written consent" to a five-stage charge schedule, calibrated to the [Public Bid-Tab Price Book's](https://open-factory.com/models/bid-tab-price-book/) 15 kV award bands. --- # 1,250 Hours: Reading BMW's Humanoid Pilot as a Buyer URL: https://open-factory.com/articles/humanoid-pilot-as-buyer/ Date: 2026-06-10 Machinery / Take Figure 02 at Spartanburg, Agility's $8,500 a month vs a $30.50 worker, the 3,500-hour breakeven, Interact Analysis under 100k units by 2030, humanoid disclosure scores, what to sign and refuse BMW's humanoid pilot, the most documented one on earth, delivered [about 1,250 operating hours in ten months](https://www.press.bmwgroup.com/global/article/detail/T0455864EN/bmw-group-to-deploy-humanoid-robots-in-production-in-germany-for-the-first-time?language=en), which is 0.6 of one worker's year. Hold that number against the next vendor video. Figure 02 moved more than 90,000 sheet-metal parts at Spartanburg on 10-hour shifts, Monday to Friday, while the plant built 30,000 X3s, and both BMW and Figure published the hours. Nobody else has. This piece reads every announced humanoid pilot the way a plant manager reads a bid: robot-hours delivered, task, dates, price, and what was disclosed against what was claimed; then prices a robot-hour against a fully loaded worker, scores the vendors on the five [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) criteria, and behind the wall lays out the pilot terms to sign, the terms to refuse, and the two use cases where the math already works. Nine pilots, one runtime figure (1,250 hours), one fleet-hours figure (Agility's 65,000 across nine committed facilities), one price ($8,500 a month, and only because Agility filed to go public), and six rows with "None" in both the hours and the price column. ## What BMW Actually Bought The Spartanburg deployment is the reference case because two parties published numbers that agree. BMW's 27 February 2026 release says Figure 02 "moved more than 90,000 components and covered approximately 1.2 million steps in around 1,250 operating hours" over ten months of ten-hour shifts, five days a week. Figure's own [19 November 2025 write-up](https://www.figure.ai/news/production-at-bmw) gives the task spec: a station requirement of 84 seconds total cycle and 37 seconds of load time, parts placed within a 5-millimetre tolerance in 2 seconds, a target of over 99% placement success per shift and a goal of zero interventions per shift. It also says the forearm was "our top hardware failure point" and that Figure 02 is being retired now that Figure 03 exists. **Ten months of ten-hour shifts, five days a week, is roughly 2,100 to 2,200 scheduled shift-hours; 1,250 hours of runtime is 57 to 60% of that, and at that pace a robot delivers about 1,500 productive hours a year against 2,000 for a one-shift worker.** That arithmetic is ours, in data/assumptions.csv, and it is generous to the robot because BMW's "operating hours" may include time the line was down. What neither party published is the achieved success rate against the 99% target, the interventions per shift against the zero goal, or what BMW paid, if anything. The pilot was a vendor demonstration on a live line, and it was a good one; it was not a purchase. The pace matters more than the parts count. 90,000 parts in 1,250 hours is 72 parts an hour, one every 50 seconds, on a station specified at 84 seconds. The robot kept up with the line when it ran. The question a buyer asks is how often it ran, and 57 to 60% of scheduled shift time is the only answer in the public record. A [Fanuc arm in a welding cell](https://open-factory.com/articles/robot-61k-cell-150k/) is expected to run above 95%, and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) prices a conventional cell at $100,000 to $250,000 installed for a task like this one, with a 15-year life. ## Agility Filed the Only Price Sheet Agility Robotics is the one humanoid vendor that has been forced to write numbers down, because it is merging into Churchill Capital Corp XI and filed an [S-4 on 4 September 2026](https://www.sec.gov/Archives/edgar/data/2074973/000121390026097764/ea0297114-04.htm). The illustrative unit economics for Digit v5 are: a RaaS fee of $8,500 a month per robot including software and maintenance, a one-time deployment fee of about $25,000, a five-year useful life; or an ownership model at about $200,000 per robot, $20,000 deployment, and $36,000 a year for Arc software and maintenance. Agility's cost to build is $150,000 at launch, targeted at $75,000 at 1,000 units a year and $30,000 at 10,000. The filing states plainly that these are "illustrative" and that actual contract terms "are subject to negotiation." The [June 2026 investor deck](https://www.sec.gov/Archives/edgar/data/2074973/000121390026071287/ea029548401ex99-2.htm) supplies the comparison the sales team will use. It sets a fully burdened human rate of $30.50 an hour, built from "$21.49 published hourly rate for material movers in 2025 per BLS data" with 5% wage inflation and a 35% benefits load, then runs the robot 120 hours a week, two ten-hour shifts, 52 weeks a year, for about 31,000 service-life hours over five years. On those assumptions one Digit replaces about $200,000 a year of labor for about $100,000 a year of RaaS. Every one of those assumptions is a negotiable input, and the deck says so in its footnotes. **The number the deck does not print is the cost per productive hour, and at BMW's disclosed pace it is $71.** RaaS at $8,500 a month plus the deployment fee spread over five years is $107,000 a year. Divide by Agility's 6,240 hours and a robot-hour costs $17.15. Divide by 4,000 hours, a real two-shift year, and it is $26.75. Divide by 1,500 hours, the only pace any humanoid pilot has disclosed, and it is $71.33, more than double Agility's own worker rate. The RaaS curve crosses Agility's $30.50 at 3,508 productive hours a year, crosses a fully loaded material mover at $27.65 at 3,870 hours, and crosses the loaded average manufacturing production wage of $43.39 at 2,466 hours. The ownership model at $80,000 a year crosses $30.50 at 2,623 hours. At 1,500 hours the RaaS robot costs $71.33 and the owned robot $53.33 an hour. Our worker rates come from public series, not the deck. BLS puts the [median material mover at $19.35 an hour and the median team assembler at $21.47](https://www.onetonline.org/link/summary/53-7062.00) in its 2025 wage data, and the [June 2026 Employer Costs for Employee Compensation](https://www.bls.gov/news.release/ecec.nr0.htm) release puts private-industry compensation at $46.89 an hour against $32.82 of wages, a 1.43 multiplier. Loaded, the material mover is $27.65 and the assembler $30.67, so Agility's $30.50 is fair for a warehouse and low for a body shop, where the [manufacturing production average of $30.37](https://fred.stlouisfed.org/graph/fredgraph.csv?id=CES3000000008) in August 2026 loads to $43.39. A Spartanburg or Savannah plant competing with datacenter construction for labor pays more than the national median, as we showed in [electricians' wages by metro](https://open-factory.com/articles/electricians-wages-by-metro/). ## Announced Capacity vs Delivered Units The gap between what humanoid vendors have announced and what they have delivered is the largest in any equipment class we cover, and it is the reason a plant manager should treat every capacity claim as marketing until it appears in a filing. Figure announced [12,000 humanoids a year at BotQ](https://www.figure.ai/news/botq) in March 2025 and reported [more than 350 Figure 03 units delivered](https://www.figure.ai/news/ramping-figure-03-production) by 29 April 2026, allocated to "internal research and development groups, data collection, housework efforts, and commercial use-case development." Agility's RoboFab is "designed to produce up to 10,000 Digits annually"; the S-4 does not count deployed robots, but it reports 2025 robot sales revenue of $1,550,360, which at the company's own illustrative $200,000 purchase price is fewer than eight robots, and total 2025 revenue of $1,781,967 against a net loss of $138.1 million. Tesla told investors in January 2025 that "several thousand Optimus robots will be doing useful things by the end of the year" and in January 2026 that Optimus is "not in usage in our factories in a material way," with start of production "before the end of 2026" and eventual capacity of one million a year, [as reported by Electrek](https://electrek.co/2026/01/28/musk-admits-no-optimus-robots-are-doing-useful-work-at-tesla-after-claiming-otherwise/) from the Q4 2025 call; Tesla's [10-K filed 29 January 2026](https://www.sec.gov/Archives/edgar/data/1318605/000162828026003952/tsla-20251231.htm) states "We have yet to commercialize Bots." Hyundai's [CES 2026 release](https://www.hyundai.com/worldwide/en/newsroom/detail/0000001100) plans Atlas on parts sequencing at Metaplant America "beginning in 2028" and a 30,000-unit-a-year robot facility; Boston Dynamics' [own account](https://bostondynamics.com/blog/atlas-evolution-from-research-robot-to-industrial-humanoid/) of the fall 2025 Metaplant test gives no hours or units. Apptronik has raised [$935 million in Series A money](https://apptronik.com/news-collection/apptronik-closes-over-935-million-series-a), named Mercedes-Benz, GXO and Jabil, and has published no unit, hour or price figure since its [March 2024 Mercedes agreement](https://apptronik.com/news-collection/apptronik-and-mercedes-benz-enter-commercial-agreement). **The units that exist are Chinese: Interact Analysis counts humanoid production rising from under 2,000 units in 2024 to more than 20,000 in 2025, over 90% of it in China, with about 10% of 2025 output in real-world applications.** That is [Interact Analysis's own June 2026 note](https://interactanalysis.com/insight/humanoid-robot-production-surges/), which puts Unitree and AgiBot above 5,000 units each and the commercial inflection "post-2032." The same firm's forecast of [fewer than 100,000 humanoids globally by 2030](https://www.manufacturingdive.com/news/us-robotics-rebounded-2025-on-track-more-growth-ifr-automate-2026/823874/), as reported by Manufacturing Dive, sits against 621,000 conventional industrial robots installed in 2025 alone, 38,000 of them in the US. Against that base, [North American buyers ordered 36,766 arms for $2.25 billion in 2025](https://www.therobotreport.com/north-american-robot-orders-rise-by-6-6-percent-2025/), about $61,000 each; the arm is the cheap part, and [the cell around it is what costs $150,000](https://open-factory.com/articles/robot-61k-cell-150k/). ## Scoring the Vendors on Disclosure The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) asks five observable questions of every equipment maker we cover: does the company publish lead times, list prices, backlog in dollars, capacity in units, and delivery performance. We applied the same five to six humanoid vendors, source beside every cell, with the same right of reply we gave the [sixteen electrical, power and cooling OEMs](https://open-factory.com/articles/oem-disclosure-scorecard/) in December. Agility scores 3.0: a dollar backlog ($300 million of Digit v5 orders for 1,000 robots on a three-year RaaS term, from one customer, with warrants issued to that purchaser), a unit capacity (10,000 a year), a partial price (illustrative, "subject to negotiation") and partial delivery performance (65,000 fleet hours and 98% tote accuracy, no uptime and no robot count). Figure scores 1.5 on capacity and the BMW hours. Unitree scores 1.0 because its [web store lists the G1 at $13,500 and the H2 at $29,900](https://shop.unitree.com/) while the H1 line says "Contact us for the real price." Tesla scores 0.5 for a planned capacity and nothing else; Boston Dynamics and Hyundai 0.5 for a planned 2028 facility; Apptronik 0.0. **No humanoid vendor publishes a lead time, and only one publishes a price a buyer could put in a purchase order.** For comparison, Schneider Electric publishes a monthly lead-time guide that feeds our [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/), Haas prints a [list price on its website](https://open-factory.com/articles/haas-vf2-true-cost/), and even the $2,850 market reports we criticised in [The $2,850 Report That Says Nothing](https://open-factory.com/articles/2850-report-says-nothing/) include a unit count. The humanoid sector in 2026 discloses less than the machine-tool trade did in 1990, which is a reason to write the disclosure into the pilot contract, because the vendor will not volunteer it. ## What to Do Monday Price the robot-hour, not the robot. Take data/cost-per-productive-hour.csv, replace our loaded wage with your own plant's for the specific position, and read off the productive hours a year the robot must deliver before it beats the person. For a two-shift 3PL floor at $27.65 that is 3,870 hours on Agility's published RaaS terms, 62% of the 6,240 hours Agility assumes and 2.6 times the pace BMW disclosed. A vendor whose proposal does not commit to logging productive hours has not agreed to be measured. Ask for the four numbers BMW and Figure published and the two they did not: runtime hours, parts moved, cycle time achieved, scheduled shift-hours, plus interventions per shift and first-pass success rate. Put them in the RFQ as reported metrics with a weekly cadence, the way [our capital-equipment RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) treats FAT and SAT, and run the price through [Quote Check](https://open-factory.com/tools/quote-check/) with the RaaS fee, deployment fee and software line separated. If you have signed a humanoid pilot, tell the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) the monthly rate. Nobody outside Agility's S-4 has published one, and the cell opens at five contributors. Behind the paywall: the six clauses to sign and the five to refuse, with model language, including the **30-day exit** that turns a pilot back into a pilot, and the two use cases where a Digit at $8,500 a month already beats the loaded wage on the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) numbers. --- # The $306k Machine: A Record Half-Year With Fewer Units URL: https://open-factory.com/articles/machine-tool-record-fewer-units/ Date: 2026-05-25 Machinery / Debrief USMTO $3.44B in H1 2026, units down 2.6%, aerospace and power generation over automotive, JMTBA's JPY 203B month, Germany at 73% utilization, Italy down 38.7%, what it means for 2027 delivery and price US manufacturers ordered [$3.44 billion of metalworking machinery in the first half of 2026](https://www.amtonline.org/article/2026-manufacturing-technology-orders-set-half-year-record), the largest half-year since AMT's USMTO series began in 1998, and they did it on 11,243 machines, 2.6% fewer than in the second half of 2025. Divide one by the other and the average US machine tool order is now about $306,000, up from about $277,000 six months earlier. That number, not the record, is the story for anyone buying a machine this year: the same count of machines is carrying more automation, more axes and more price. This debrief covers the USMTO record and the unit gap, who bought (aerospace and power generation) and who did not (job shops), the JPY 203 billion month in Japan, Germany at 73% utilization and Italy's index at 47.8, the Gardner index crossing 50, and, behind the wall, what the backlog arithmetic, the yen and the 15% Annex III rate mean for a machine you order this month. Seventy-eight months of AMT releases, transcribed one URL at a time into [data/usmto-monthly.csv](data/usmto-monthly.csv): the series bottomed at $321.7 million in July 2024, set a monthly record of [$814.3 million in December 2025](https://www.amtonline.org/article/manufacturing-technology-orders-set-record-in-december-2025), and has not printed a month below $441 million in 2026, with [June at $672.7 million](https://www.amtonline.org/article/2026-manufacturing-technology-orders-set-half-year-record), up 56.8% on June 2025. ## The Record Is in Dollars, Not Machines AMT has been warning about this divergence for a year. Its [December 2025 release](https://www.amtonline.org/article/manufacturing-technology-orders-set-record-in-december-2025) dates the "loosening correlation between the number of machines ordered and their value" to the second half of 2021, and its [March 2026 release](https://www.amtonline.org/article/mt-orders-predictably-decline-in-january-after-historic-december) says the units ordered in January 2026 were the lowest for any January since 2010 while the dollars were the highest since 2022. The [June release](https://www.amtonline.org/article/automation-fuels-strong-start-to-2026-manufacturing-technology-orders) closes the loop: average order values have grown faster than machine tool inflation since 2020, the gap widened in early 2026, and "a significant portion of the order value growth is due to additional automation being added to orders of increasingly sophisticated machinery." Read across the three panels: $2.53 billion in H1 2025, $3.20 billion in H2 2025 and $3.44 billion in H1 2026; about 11,543 machines in H2 2025 (derived from AMT's 2.6% decline) against 11,243 in H1 2026; and an average order of about $277,000 becoming about $306,000, a 10% rise in six months. **The average US machine tool order has risen about 10% in six months while the BLS machine tool PPI rose 1.4% over the same period, so most of the $306k is configuration, not inflation.** The PPI for machine tool manufacturing stood at [142.6 in July 2026](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCU333517333517), a 22.5% climb since January 2021, which we charted in [What a Haas VF-2 Really Costs](https://open-factory.com/articles/haas-vf2-true-cost/). A Haas VF-2 still lists at [$70,995](https://machinetoolindex.com/machines/haas-vf-2), so a $306k average means the mix has moved toward five-axis machines, pallet pools and robot-tended cells. DMG MORI's own average machine order price was [JPY 81.8 million in the first half of 2026](https://www.dmgmori.co.jp/corporate/en/ir/ir_library/pdf/fy2026_2shihanki_ex_e.pdf), about $515,000 at August's yen rate, and the company credits "robot integration and bundled peripheral equipment" for holding that price while it sells more entry-level BX machines. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) exists to split that average into machine, automation and options once contributed quotes clear its cell-size rules; until then, [Quote Check](https://open-factory.com/tools/quote-check/) flags a quote that folds the robot, the pallet changer and the install into one line. ## Aerospace and Power Generation Bought, Job Shops Waited The dollars are concentrated in a few customer industries. Aerospace ordered the most machinery on record in H1 2026 in both value and units, with value [nearly one-third higher than in H2 2025 and units up nearly a quarter](https://www.amtonline.org/article/2026-manufacturing-technology-orders-set-half-year-record), on commercial aircraft backlogs, space and munitions replenishment. Engine, turbine and power transmission manufacturers ordered so heavily in June that power generation and distribution finished H1 2026 [14% above the automotive sector](https://www.amtonline.org/article/2026-manufacturing-technology-orders-set-half-year-record) in machinery spend, the same buyers who sit behind the gas turbine slot market in [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/). | Segment | Change | Comparison | Source | |---|---|---|---| | Commercial and service machinery (incl. chip-fab inspection) | +121.5% | 2025 vs 2024 | [AMT, Feb 2026](https://www.amtonline.org/article/manufacturing-technology-orders-set-record-in-december-2025) | | Aerospace | +45.1% | 2025 vs 2024 | AMT, Feb 2026 | | All customers | +22.5% | 2025 vs 2024 | AMT, Feb 2026 | | Automotive | +22.2% | 2025 vs 2024 | AMT, Feb 2026 | | Contract machine shops | +19.1% | 2025 vs 2024 | AMT, Feb 2026 | | Aerospace, order value | ~+33% | H1 2026 vs H2 2025 | [AMT, Aug 2026](https://www.amtonline.org/article/2026-manufacturing-technology-orders-set-half-year-record) | | Aerospace, units | ~+25% | H1 2026 vs H2 2025 | AMT, Aug 2026 | | All customers, order value | +7.5% | H1 2026 vs H2 2025 | AMT releases, our arithmetic | | All customers, units | -2.6% | H1 2026 vs H2 2025 | AMT, Aug 2026 | | Contract machine shops, units | ~-8% | H1 2026 vs H2 2025 | AMT, Aug 2026 | *Source: AMT USMTO press releases, Feb and Aug 2026; CSV at [data/usmto-sector-changes.csv](data/usmto-sector-changes.csv)* **Job shops, the largest customer segment, ordered a record dollar value in H1 2026 and nearly 8% fewer machines than in H2 2025.** AMT's explanation is structural: OEMs "made the necessary investments to absorb increased production demands internally rather than by contracting with external job shops," so the machinery went to the aerospace primes and the turbine plants, not to the shops that used to machine their overflow. Regionally the West booked [$884.1 million in the first half, up 63.3%](https://www.americanmachinist.com/news/news/55397089/machine-tool-demand-is-booming-in-2026-us-manufacturing-technology-orders-report-june-2026), as reported by American Machinist. For a job shop the reading is uncomfortable: your customers are buying the machines you wanted to run for them, which is the same dynamic that put [the EV plants' machines on the secondary market](https://open-factory.com/articles/ev-plant-machines-second-hand/) a year earlier and the reason the [Used CNC Auction Index](https://open-factory.com/articles/used-cnc-auction-index/) is worth building. ## Japan Sells It: JMTBA's JPY 203 Billion Month The machines behind the US record are largely Japanese. The Japan Machine Tool Builders' Association reported [June 2026 orders of JPY 203.38 billion](https://www.jmtba.or.jp/english/wjmtbap/wp-content/uploads/2026/08/JMTBA-NOTES-2026-07.pdf), up 52.7% on June 2025, the first month ever above JPY 200 billion and the sixteenth straight month above JPY 120 billion. Foreign orders were JPY 145.36 billion (+55.8%), domestic JPY 58.02 billion (+45.5%), and the first-half total reached JPY 1,055 billion, up 35.7%. North America took JPY 40.65 billion in June, up 44.0% on the year, of which the United States alone was JPY 36.0 billion (+44.4%); China took JPY 56.5 billion (+75.6%), Asia as a whole JPY 81.4 billion (+74.4%), and Europe JPY 19.4 billion (+21.6%). Over the first half, US orders to Japanese builders were JPY 203.7 billion against JPY 149.6 billion a year earlier, a 36% rise that matches the 36% rise in USMTO almost to the point. One correction to a number that has circulated, including in our own earlier notes: the "+174%" figure attached to North America is JMTBA's index reading for Asia (174.4, meaning +74.4%). North America's June index was 144.0. The US is still the single largest destination for Japanese machine tools, and [Japan Is Eating the US Capex Boom](https://open-factory.com/articles/japan-eats-capex-boom/) stands, but the growth rate is 44%, not 174%. [DMG MORI](https://www.dmgmori.co.jp/corporate/en/ir/ir_library/pdf/fy2026_2shihanki_ex_e.pdf) booked a record JPY 335.2 billion of orders in the half (+34.8%), raised its full-year forecast to JPY 630 billion, and named "global key accounts in Europe and North America" placing "very large-lot orders for machines and automation" in aerospace, defense, power and semiconductors. Every one of those machines enters the US under the schedule in [Tariffs, September 2026](https://open-factory.com/articles/tariff-schedule-2026/), and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) puts a Japanese machining center at 15% all-in until December 31, 2027. ## Germany at 73%, Italy at 47.8 Europe's builders are not in the same cycle. The VDW reported [German order intake up 15% in Q1 2026](https://vdw.de/en/bottomed-out-first-quarter-sees-increase-in-orders-for-machine-tool-industry/) (domestic +18%, foreign +14%) after three years of decline, but production fell 11% to EUR 2.8 billion, capacity utilization slid to 73%, and employment dropped almost 9% to 60,600 in March. Exports to the US rose 8% while exports to China fell 32%. The association's economist called it "bottomed out" and "nowhere near to reversing the trend." For 2025 as a whole, [orders fell 3%](https://vdw.de/en/german-machine-tool-industry-leveling-off-china-further-dominating-production/) (domestic -16%, foreign +3%), and the VDW expects [2026 production of EUR 13.7 billion, up 1%](https://vdw.de/en/german-machine-tool-industry-focuses-on-diversifi-cation-and-technological-leadership/), a fifth below the 2018 peak in nominal terms. Italy is worse. UCIMU's order index printed [47.8 in Q2 2026 on a 2021 = 100 base, down 25.8%](https://www.ucimu.it/en/press/press-releases/read/2026/07/ucimu-order-index-still-falling-in-the-second-quarter-of-2026-258-foreign-orders-153-do/) on the year, with domestic orders at 33.1 (-38.7%) and foreign at 63.2 (-15.3%), after a [Q1 at 87.1](https://www.ucimu.it/en/press/press-releases/read/2026/04/ucimu-in-the-first-quarter-2026-machine-tool-orders-on-the-rise-31-foreign-orders-289-do/) that was carried entirely by exports. UCIMU's president blamed the Strait of Hormuz, the automotive transition and Italian buyers waiting for the Hyper-depreciation decree, which took effect on June 12 and had drawn 7,000 applications worth EUR 2.5 billion by July 9. **German shops running at 73% and Italian builders at half their 2021 order rate are the spare capacity in this market, and a buyer who can accept a DMG MORI from Pfronten or a Mazak from Worcester instead of Iga or Inabe should ask for the delivery difference in writing.** The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) tracks which builders publish anything at all about delivery; today the associations publish more than the companies. ## Gardner Says the Shops Caught Up in January The order data leads the shop-floor data by about a year. Gardner's Business Index sat below 50 for six straight months in the second half of 2025 while USMTO dollars climbed, then [jumped to 52.3 in January 2026](https://www.gardnerintelligence.com/news/gbi-surges-to-highest-reading-since-2022), its highest since June 2022, reached [56.2 in April](https://www.gardnerintelligence.com/news/gbi-rebounds-in-april-extending-expansion-streak) and eased to [54.4 in August](https://www.gardnerintelligence.com/news/gardner-business-index-dips-again-in-august-as-costs-ease-and-capital-plans-strengthen), still an eighth consecutive month of expansion. Gardner's metalworking sub-index followed the same path, at [53.0 in January](https://www.mmsonline.com/articles/metalworking-index-surges-into-expansion-), 56.3 in February, 56.2 in May and June and [55.7 in July](https://www.mmsonline.com/articles/metalworking-index-eases-in-july-but-extends-expansion-to-seven-months). Left panel: 46.2 in April 2025 after the tariff announcement, 48 to 49 through the autumn, 52.3 in January 2026, 56.2 in April, 54.4 in August. Right panel: USMTO from $357 million in January 2025 to $814 million in December and $673 million in June 2026; the two gaps in the Gardner line are months whose readings we could not find on Gardner's site. Two components matter more than the headline. Supplier deliveries read [62.0 in August and 63.2 in June](https://www.gardnerintelligence.com/news/expansion-streak-reaches-six-months-as-gbi-eases-slightly-in-june), which Gardner describes as "plants that remain busy enough to stretch lead times"; that is the machine-tool version of the ISM table we read in [The 171-Day Capex Lead Time](https://open-factory.com/articles/171-day-capex-lead-time/). And planned capital spending per plant rose to about $558,000 on a three-month average in August, from about $513,000, with 74% of plants planning a purchase. The buyers are not done, and the shops that answer Gardner's survey are the same job shops whose unit counts fell 8% in the USMTO data, so the two series now agree that the money is being spent by fewer, larger buyers. Hurco, the Indianapolis builder, reported [orders up 25% in its July quarter and 24% over nine months](https://www.sec.gov/Archives/edgar/data/315374/000110465926105339/hurc-20260904xex99d1.htm), "outpacing shipments," on 5-axis and larger machines; nine-month orders of $155.0 million against sales of $137.8 million is a backlog building at a small builder too. AMT's own forecast, from Oxford Economics at its Summer Economic Forum, calls for orders up 1.5% in the second half and [nearly $7 billion for 2026](https://www.amtonline.org/article/2026-manufacturing-technology-orders-set-half-year-record). ## What To Do Monday Three things follow from a record set on fewer machines. First, price the automation separately: ask the distributor to quote the machine, the pallet system or robot, the probing and tooling packages, and freight, rigging and install as separate lines, the way the [RFQ template for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) lays out, because the $306k average is built from bundled lines that nobody can benchmark, and [Quote Check](https://open-factory.com/tools/quote-check/) can only band a machine price once the robot and the rigging are stripped out of it. Second, put the HTS subheading, the country of origin and the tariff instrument on the quote, and tie delivery and entry dates to the price. A Japanese machining center is 15% all-in today and 29.2% from January 1, 2028, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) now includes the entry-date rate on every line. Third, ask where the machine will be built and when it will enter the country: a German or Italian plant at 73% utilization can ship what a Japanese plant with seven months of backlog cannot, and the builder that answers both questions in writing is the one the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) will score for it. Behind the wall: the backlog arithmetic that puts a Japanese five-axis ordered this month into **April 2027**, what the yen's 10% slide and the 15% Annex III rate leave on a landed $306k machine, and the three numbers to check on the second Monday of every month, all built on the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) machine-tool worked example. --- # From 36 Weeks to Stock: The ControlLogix Lead-Time Cycle URL: https://open-factory.com/articles/controllogix-36-weeks-to-stock/ Date: 2026-05-17 Machinery / Deep dive Rockwell bulletins 35 to 208 days, backlog $1.6B to $5.6B, a $25,920 surplus CPU, Siemens DI destocking, 250 bps of price, ISM at 60.6, the second-sourcing playbook On September 22, 2021, Rockwell Automation's own lead-time bulletin listed a ControlLogix 5580 controller at [35 calendar days](https://web.archive.org/web/20210923232228/https://www.rockwellautomation.com/content/dam/rockwell-automation/sites/downloads/pdf/Lead-Times-Rockwell-Automation.pdf) against a standard lead time of "Stock". On October 26, 2022, the same bulletin listed the same controller at [208 days](https://web.archive.org/web/20221103173047/https://www.rockwellautomation.com/content/dam/rockwell-automation/sites/downloads/pdf/Lead-Times-Rockwell-Automation.pdf), just under 30 weeks, and a 1756 analog input card at 233 days. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) has every edition of that bulletin we could recover, eleven of them between July 2021 and March 2023, and the PLC line in the Monitor today reads what Rockwell's standard column read before the shortage: stock. The chart reads 5 weeks for the ControlLogix 5580 CPU in September 2021, 6 in December, 8 in February 2022, 29 in July, 30 in October and January 2023, and 23 by March 29, 2023. The 1756 analog I/O line runs 6, 6, 15, 32, 33, 37, 37 over the same dates, and the CompactLogix 5380 CPU peaked at 31 weeks in January 2023 before falling to 17 by the end of March. Every row is in [plc-lead-times.csv](data/plc-lead-times.csv) with the archive URL it came from. This piece covers where those bulletins came from and what they said, the backlog that went from $1.6 billion to $5.6 billion, the chip supply that caused it, what a surplus controller cost when you could not get one, the destocking that followed at Siemens, Rockwell and Schneider, the price that never came back down, the component indices that are turning again in 2026, and, behind the wall, the second-sourcing playbook: which functions can leave the Rockwell rack without a rewrite, how to hold spares, and the frame-agreement clauses that guarantee allocation the next time. ## What Rockwell Published Every Two Weeks Rockwell does not publish product-level lead times today. Its supply-chain page says [some product lines have returned to pre-pandemic lead times, others remain constrained](https://web.archive.org/web/20260120011413/https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html), and sends the reader to a distributor. That page has read the same way since at least June 2023, and the number a buyer actually gets is the distributor's, which is what the [Quote Check](https://open-factory.com/tools/quote-check/) tool reads against the Monitor. But between August 2021 and March 2023 the same URL carried a PDF titled "Rockwell Automation Manufacturing Lead Times," refreshed every two weeks, listing every constrained product family, its standard lead time, its current average lead time in calendar days, and an arrow for the change. The August 2021 version of the page promised exactly that: [we will monitor and adjust our finished good lead times every two weeks (up and down)](https://web.archive.org/web/20210803035756/https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html). The Wayback Machine kept nine editions of the PDF; a distributor's mirror on pdf4pro kept the January 12, 2022 edition; two Scribd uploads kept February 23, 2022 and March 29, 2023. **The July 28, 2021 edition did not list the ControlLogix 5580 at all**, because it was still shipping at standard. It listed [CompactLogix L1 and L3 controllers at 45 days and Kinetix 5700 servo drives at 38](https://web.archive.org/web/20210802141640/https://www.rockwellautomation.com/content/dam/rockwell-automation/sites/downloads/pdf/Lead-Times-Rockwell-Automation.pdf). By [November 17, 2021](https://web.archive.org/web/20211123104045/https://www.rockwellautomation.com/content/dam/rockwell-automation/sites/downloads/pdf/Lead-Times-Rockwell-Automation.pdf) the 5580 was on the list at 41 days, PowerFlex 525 drives at 76, CompactLogix 5380 at 54. The [January 12, 2022 edition](https://pdf4pro.com/file/e959d/content_dam_rockwell_automation_sites_downloads_pdf_Lead_Times_Rockwell_Automation.pdf.pdf) had the 5580 at 42 days and the 1769 L2 CompactLogix at 123; six weeks later, on [February 23, 2022](https://www.scribd.com/document/566019038/Lead-Times-Rockwell-Automation), the 5580 was 57, the 5069 CompactLogix 132, Kinetix 5700 143, and the Micro800 line 136. Then the step: on [July 20, 2022](https://web.archive.org/web/20220721201005/https://www.rockwellautomation.com/content/dam/rockwell-automation/sites/downloads/pdf/Lead-Times-Rockwell-Automation.pdf) the 5580 read 200 days, 5069 CompactLogix 197, 1756 analog I/O 222, Kinetix 5700 225. The high-water marks by family, from the same bulletins: ControlLogix 5580 controllers 208 days (October 26, 2022); CompactLogix 5380 controllers 214 days and 1756 analog I/O 259 days and Kinetix 5700 264 days ([January 18, 2023](https://web.archive.org/web/20230124103535/https://www.rockwellautomation.com/content/dam/rockwell-automation/sites/downloads/pdf/Lead-Times-Rockwell-Automation.pdf)); 1732ES ArmorBlock Guard I/O 292 days and PanelView Plus 7 Standard 15-inch 256 days ([March 29, 2023](https://www.scribd.com/document/662418405/Lead-Times-Rockwell-Automation-2023-03-29)). Thirty-seven weeks for an analog card, forty-two for a safety I/O block. The "20 to 36 weeks on processors, 16 to 26 on I/O" that integrators [reported in the field](https://industrialmonitordirect.com/blogs/knowledgebase/plc-component-supply-chain-shortages-lead-times-alternatives) is consistent with Rockwell's own numbers, with the integrators' figures sitting slightly under the bulletin on CPUs and well under it on I/O. The distributor bulletins were downstream of this PDF: Electrical Supply & Equipment's Rockwell delivery-date page still describes [dates forecast and updated biweekly for ControlLogix controllers, I/O, CompactLogix, Kinetix and PowerFlex 755](https://ese-co.com/rockwell-delivery-date-updates), the same list of families. Note the one thing the bulletins never changed: the "standard lead time" column read "Stock" for every controller, I/O card, drive and HMI on the list. Rockwell's definition of normal is a distributor shelf. That is the baseline the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) uses for the PLC class, and it is why the PLC line is the only row in the Monitor whose recovered state is a word rather than a number of weeks. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) credits Rockwell for having published this table at all for twenty months; no other automation OEM did. It also records that the table stopped in 2023 and has not returned. ## The Backlog Went From $1.6 Billion to $5.6 Billion Rockwell's 10-K puts the order backlog at [$1,557.1 million on September 30, 2020 and $2,910.5 million a year later](https://www.sec.gov/Archives/edgar/data/1024478/000102447821000083/rok-20210930.htm). The FY2022 10-K reports [$5,197.0 million at September 30, 2022](https://www.sec.gov/Archives/edgar/data/1024478/000102447822000093/rok-20220930.htm): Intelligent Devices $2,086.1 million, Software & Control $1,456.8 million, Lifecycle Services $1,654.1 million. The peak came two quarters later. Manufacturing Dive, reporting the April 27, 2023 call, wrote that the [backlog sat at $5.6 billion at the end of Q2, up from $5.2 billion at the beginning of the year](https://www.manufacturingdive.com/news/rockwell-automation-order-backlog-lead-times-supply-chain-improve-earnings-call/649647/), with CEO Blake Moret guiding it to $5 billion by year-end. It fell faster than that: [over $4.1 billion at September 30, 2023](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2023/ROK%20Prepared%20Remarks%20Q4%20FY23_vfinal.pdf), then [$3,090.6 million in 2024](https://www.sec.gov/Archives/edgar/data/1024478/000102447824000107/rok-20240930.htm) and [$2,878 million in 2025](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm). The bars read $1.6 billion, over $2 billion, $2.9 billion, over $5 billion, $5.2 billion, $5.6 billion, $4.1 billion, $3.1 billion and $2.9 billion. The two product segments went from $549 million combined in 2020 to $3,543 million in 2022 and back to $1,358 million in 2025, while Lifecycle Services stayed between $1.0 billion and $1.75 billion the whole way. **Rockwell told investors in November 2022 that pre-pandemic it held about one month of the following year's product revenue in backlog, and that the backlog then represented over 50% of the FY2023 sales guide for both product segments**, in its own words: [this unprecedented backlog coverage adds to our confidence in our revenue outlook](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2022/Rockwell-Automation-Prepare-Remarks-2022-.Q4.pdf). Six months of product backlog against one month normal is the same ratio as 208 days against stock. The backlog was not demand in the ordinary sense. It was customers ordering forward because the lead time told them to. Rockwell said so at every step. In May 2022: [at some point we expect orders to moderate, as lead-times for our products return to more normal levels](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2022/Rockwell-Automation-Prepare-Remarks-2022-.Q2.pdf). In August 2023: [with improving lead times, machine builders do not need as many months of products on order and are no longer placing unusually large advance orders](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2023/ROK%20Prepared%20Remarks%20Q3%20FY23_vfinal.pdf). In November 2023: [orders decreased quicker than we expected as machine builders and distributors continued to work through inventory in response to our rapidly improving lead times](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2023/ROK%20Prepared%20Remarks%20Q4%20FY23_vfinal.pdf). The [Book-to-Bill League Table](https://open-factory.com/articles/book-to-bill-league/) tracks the same pattern across the electrical OEMs; Rockwell was simply two years ahead of the transformer makers on the way up and on the way down. ## Chips, Not Demand Rockwell's 10-Q for the quarter ending June 30, 2022 said the results by region, segment and industry [were primarily driven by component availability rather than the underlying demand](https://www.sec.gov/Archives/edgar/data/1024478/000102447822000063/rok-20220630.htm). The May 2022 prepared remarks named the fix list: [incremental capacity, improved allocation percentages, opportunistic broker buys](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2022/Rockwell-Automation-Prepare-Remarks-2022-.Q2.pdf), plus qualifying new semiconductor vendors for redundant sources with the benefit expected in fiscal Q4. By June 2023 the supply-chain page said Rockwell had [added a secondary source on over 1,000 components](https://web.archive.org/web/20230617105031/https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html) and duplicated core product lines. A PLC vendor spent eighteen months doing what its customers were told to do to it: second-sourcing. The macro series says the same thing with a lag. The ISM Manufacturing Supplier Deliveries index, where a reading above 50 means slower deliveries, peaked at [78.8 in May 2021](https://web.archive.org/web/20210610103853/https://www.ismworld.org/supply-management-news-and-reports/reports/ism-report-on-business/pmi/may/), its highest since 1974, and was back under 50 by November 2022. Rockwell's ControlLogix lead time peaked seventeen months after the ISM index did, in October 2022, and did not fall below 25 weeks until March 2023, four months after ISM's index said deliveries were speeding up. A PLC is at the end of a long chain: wafer, microcontroller, board, module, distributor shelf. The [171-day capex lead time](https://open-factory.com/articles/171-day-capex-lead-time/) that ISM's own table reported at its 2022 peak (186 days in June 2022) is the average across all capital goods. A controller ran at 200 to 208 days for nine months. Left panel: 57.5 in January 2019, 47.2 in May 2020, 78.8 in May 2021, 45.1 in December 2022, 43.5 in May 2023, and 60.6 in April 2026. Right panel: 5 weeks in September 2021, 29 in July 2022, 30 in October 2022 and January 2023, 23 in March 2023. The full monthly series is in [ism-supplier-deliveries.csv](data/ism-supplier-deliveries.csv), and the [Lead-Time Monitor launch note](https://open-factory.com/articles/lead-time-monitor-launch/) explains how we snapshot ISM's report each month. ISM's commodities list is the better leading indicator for automation specifically. "Electronic Components" entered the list of commodities in short supply in [December 2020 and stayed for 38 consecutive months](https://web.archive.org/web/20240223072318/https://www.ismworld.org/supply-management-news-and-reports/reports/ism-report-on-business/pmi/January/) through January 2024. Semiconductors were listed for 34 straight months to September 2023. When the electronic-components streak ended in early 2024, Rockwell's 10-Q for the quarter ending March 31, 2024 said the supply chain had [largely recovered to pre-pandemic lead times and service levels](https://www.sec.gov/Archives/edgar/data/1024478/000102447824000052/rok-20240331.htm). The two dates line up within a quarter. The line climbs from 2 months in January 2021 to 38 in January 2024, resets to 3 in June 2024, reaches 10 in January 2025, resets again to 4 in June 2025, and has climbed every month since to 14 in April 2026; semiconductors, which had made single-month appearances in January and September 2025, were listed in March 2026 and again in April. Read that with the Rockwell data: the last time the streak passed 14 months was February 2022, when the ControlLogix bulletin read 57 days and was about to read 200. ## What a Controller Cost When You Could Not Get One Rockwell does not publish list prices, and its distributors quote under agreements, so the only public price for a ControlLogix processor is the secondary market. We pulled archived pages for one part, the 1756-L83E, a 10 MB ControlLogix 5580 CPU that sits in the middle of the range. In February 2018 Radwell, a surplus dealer that states on every Rockwell page that it is [not an authorized distributor](https://web.archive.org/web/20180204170426/http://www.radwell.com:80/Buy/ALLEN%20BRADLEY/ALLEN%20BRADLEY/1756-L83E), asked $12,500 for a new unit through its independent supply chain, $9,375 for new surplus, $7,812.50 for used surplus, and $5,625 for a repair of yours. The same page, with a stated value date of May 20, 2022 and captured in March 2023, asked [$25,920 for a unit in original packaging, $23,040 in Radwell packaging, $18,720 for a Radwell-certified unit](https://web.archive.org/web/20230323144435/https://www.radwell.com/Buy/ALLEN%20BRADLEY/ALLEN%20BRADLEY/1756-L83E), and $5,600 for a repair, with five, one and six units in US stock respectively. The bars read $12,500 and $9,375 in 2018, $25,920 and $18,720 in May 2022, and $3,749.88 in March 2026, which is the median asking price across 54 new or sealed listings on [eBay's product page for the part](https://web.archive.org/web/20260304081411/https://www.ebay.com/p/7031093585) (range $2,600 to $4,398, two pre-owned listings at $3,385 and $3,499). **A sealed surplus 1756-L83E asked 2.1 times its 2018 price at the top of the shortage and asks 0.3 times that price now.** These are asking prices from single sellers and one marketplace snapshot, not transactions, and Rockwell's contract price to a distributor is not in any of them. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) records them as quotes with n and date, per its rules, and nothing more. The direction is not in doubt. When the lead time is stock, a surplus dealer is competing with a shelf. When the lead time is 208 days, a surplus dealer is competing with a shutdown. The repair line is the one to remember. Radwell's repair price for the same CPU was $5,625 in 2018 and $5,600 in 2022. It did not move, because the constraint was new silicon, not labor. In 2022 a plant with a failed 1756-L83E and no spare had three choices: wait 30 weeks at the distributor, pay $25,920 for a sealed surplus unit with no Rockwell warranty and no guarantee of firmware revision, or pay $5,600 and wait 5 to 20 days for a repair. The [recommended spares list](https://open-factory.com/articles/recommended-spares-margin/) that every integrator attaches to a panel quote is priced for the first world, not the second. ## The Bullwhip: Destocking at Siemens, Rockwell and Schneider Siemens Digital Industries is the cleanest public record of what happened on the way down, because Siemens describes its automation orders in plain language every quarter. Q1 FY2023 (quarter to December 2022): orders EUR 6,320 million, [book-to-bill ratio was at 1.24](https://assets.new.siemens.com/siemens/assets/api/uuid:164500b4-09db-41c2-91be-9fd44653e598/HQCOPR202302086645EN.pdf). Q3 FY2023: orders EUR 4,103 million, down 35% year on year, with an [accelerated order decline in the automation businesses, most particularly in the short-cycle factory automation business; orders in prior periods were elevated by proactive customer purchasing](https://assets.new.siemens.com/siemens/assets/api/uuid:ed13ce01-322e-4909-9701-4b0ed3766354/HQCOPR202308086759EN.pdf). Q4 FY2023: [customers continued destocking following former proactive purchasing](https://assets.new.siemens.com/siemens/assets/api/uuid:b17d5fb3-7830-49ae-b47b-0da136d68066/HQCOPR202311156817EN.pdf). Q4 FY2024: [customers were still reducing elevated stock levels due to weak global demand for manufactured goods](https://assets.new.siemens.com/siemens/assets/api/uuid:df33f4cd-3bb5-4cd1-9ff8-79314ee62e0f/HQCOPR202411047033EN.pdf). Q2 FY2025 (quarter to March 2025): [destocking of elevated stock levels at customers approached completion toward the end of Q2 FY 2025](https://assets.new.siemens.com/siemens/assets/api/uuid:d84dc895-95ef-40a2-864c-1a083f785026/HQCOPR202505127163EN.pdf), in China. Q3 FY2025: [revenue in the automation business increased year-over-year for the first time since Q4 FY 2023](https://assets.new.siemens.com/siemens/assets/api/uuid:d1a425b2-b59d-401b-aebf-9c622e812f4d/HQCOPR202508017210EN.pdf). The bars read EUR 6.3 billion, 5.3, 4.1, 4.9, 4.0, 4.3, 4.5, 4.3, 4.2, 4.3, 4.4, 5.5, 4.8 and 4.8 billion from Q1 FY2023 to Q2 FY2026, and the Digital Industries profit margin went from 23.5% in Q2 FY2023 to 14.5% in Q1 FY2025 and back to [18.5% in Q2 FY2026](https://assets.new.siemens.com/siemens/assets/api/uuid:d4072f41-f21d-4287-af75-0cb72c194262/HQCOPR202605117394EN.pdf), when Siemens raised its FY2026 Digital Industries guide to 7 to 10% revenue growth and a 17 to 19% margin. The quotes are in [siemens-di-automation.csv](data/siemens-di-automation.csv). Two full years of destocking, Q3 FY2023 to Q2 FY2025, against roughly fifteen months of stocking up. Rockwell's version ran on the same calendar with a US accent. In January 2024: [high levels of channel inventory and some lingering supply chain constraints continue to impact the timing of product shipments](https://www.sec.gov/Archives/edgar/data/1024478/000102447824000006/q1fy24ex99.htm). In May 2024: [the impact of high inventory levels at machine builders is larger than we expected](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2024/ROK%20Prepared%20Remarks%20Q2%20FY24%20Final.pdf). In February 2025: [new demand placed on our distributors is flowing through at close to 100% in terms of new orders on Rockwell, matching historical levels, giving further evidence that the destocking cycle is mostly behind us](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2025/ROK-Prepared-Remarks-Q1-FY25.pdf). Schneider Electric's Industrial Automation, the third of the big three in US panels, grew [3% organic for full-year 2025 and 8% in Q4 2025 "with continued recovery in Discrete automation"](https://www.emr-online.com/schneider-electric-schneider-electric-full-year-2025-results/), with strong growth in PLC and motion for packaging OEMs and an Industrial Automation backlog of EUR 4,022 million, up 8%; Q1 2026 Industrial Automation was [up 4.4%](https://finanzwire.com/article/schneider-electric-epa-su-schneider-electric-q1-2026-revenues-report-tpcOVcaN71H) as reported by finanzwire from the April release. Rockwell's product backlog at September 2025 ($1,358 million across the two product segments) is 2.5 times its September 2020 level on [sales about 30% higher](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000112/fy2025ex99.htm) ($8,342 million in FY2025 against $6,330 million in FY2020), so the channel is still carrying more than it did in 2019, and less than half of what it carried at the peak. The [Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) lists the quarter-by-quarter reading for all three OEMs. The buyer's lesson from the bullwhip is not that the OEMs behaved badly. It is that the order book lied in both directions. In 2022 the backlog overstated demand because every machine builder ordered twelve months of controllers to be sure of six; in 2024 the order intake understated demand because they were consuming those controllers off their own shelves. Anyone reading the [Book-to-Bill Table](https://open-factory.com/models/book-to-bill-table/) for a PLC vendor should discount both tails. The useful signal is the lead time itself, and that is what the Monitor tracks. ## Price Never Came Back Down Lead time round-tripped. Price did not. Rockwell announced [several price increases totaling 17%](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2022/Rockwell-Automation-Prepare-Remarks-2022-.Q2.pdf) between mid-2021 and May 2022, and told investors the realization would lag because "much of our pricing is set by pricing agreements we have with our customers" that renew annually. The realization then showed up in the growth bridge: about [7 points of organic growth from price in Q1 FY2023](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2023/ROK-Prepared-Remarks-Q1-FY23_vfinal.pdf), 6 in Q2, 3 in Q3, 6 in Q4; then 3, 1.5, 3.5 and 2 points through FY2024; 1, 3, 3 and 4 through FY2025; and 3 points in each of the first two quarters of FY2026, of which [about half came from underlying price realization and half from tariff-based pricing](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2026/rok-pepared-remarks-q2-fy26.pdf). The bars sum to about 46 points of price across fourteen quarters, and there is no negative bar: in no quarter since the shortage has Rockwell reported price as a drag. On May 5, 2026 the company raised its full-year price outlook to [250 basis points of total price for fiscal 2026, with 150 basis points coming from underlying price and 100 basis points from tariff-based price](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2026/rok-pepared-remarks-q2-fy26.pdf), up 50 basis points from the prior outlook, "all from underlying price," citing cost increases in areas affected by tariffs, demand for memory, and fuel. [Rockwell's 250 Basis Points](https://open-factory.com/articles/rockwell-250bps/) takes that number apart line by line and [Quote Check](https://open-factory.com/tools/quote-check/) applies it to a distributor quote. This piece only needs the shape: a shortage that let the vendor take 17% in a year and hold it, followed by a normal that adds 2 to 4 points a year on top. The tariff half has a procurement twist. Rockwell said in February 2025 it would [reprice our backlog to reflect our new price lists](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2025/ROK-Prepared-Remarks-Q1-FY25.pdf) when Canada and Mexico tariffs took effect, and its 10-Q for the quarter to March 31, 2026 says that after the Supreme Court's February 2026 IEEPA ruling the company [may be entitled to a refund of tariffs previously paid, and if tariff amounts are ultimately refunded, the Company expects to implement a refund process for qualified customers](https://www.sec.gov/Archives/edgar/data/1024478/000102447826000022/rok-20260331.htm). That sentence is the whole argument of [Who Gets the IEEPA Refund](https://open-factory.com/articles/ieepa-refund-fight/) applied to a PLC purchase order: the tariff surcharge you paid in 2025 is recoverable only if your PO documents it as a surcharge and your frame agreement says it flows back. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) prices the surcharge by HTS code. The [September 2026 tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) tracks which instruments still apply. ## The Semiconductor Side Is Tightening Again The component indices that led the 2021 to 2023 cycle by a year are moving again, and in the same direction. ECIA's Industry Pulse survey of component manufacturers, distributors and reps reported the share of participants seeing increasing lead times at [20% in August 2025](https://ecia.memberclicks.net/assets/workplace/Industry%20Pulse%20August%20and%20Q3%202025%20Outlook%20Article.pdf), 21% in November, [33% in December](https://growthzonecmsprodeastus.azureedge.net/sites/2395/2026/05/Industry-Pulse-December-2025-Outlook-Article.pdf), nearly 40% in January 2026, 34% in February, [61% in March](https://growthzonecmsprodeastus.azureedge.net/sites/2395/2026/05/Industry-Pulse-March-2026-Outlook-Article.pdf) and [69% in April, with 92% for semiconductors and "reports of declining lead time pressure essentially nonexistent"](https://growthzonecmsprodeastus.azureedge.net/sites/2395/2026/05/Industry-Pulse-April-2026-Outlook-Article.pdf). The pressure started in DRAM and NAND (80% reporting increases in December) and spread: 44 to 58% in other semiconductor categories in December, 78% in March. The bars read 23, 20, 21, 33, 40, 34, 61 and 69 percent from July 2025 to April 2026. ISM's Supplier Deliveries index printed [60.6 in April 2026](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-52-7-april-2026-ism-manufacturing-pmi-report-302759226.html), its highest since May 2022 (65.7), and its Backlog of Orders index has been above 50 since January 2026 after 38 months below it, November 2022 to December 2025. Rockwell itself named "demand for memory" as a cost driver on May 5, 2026. None of this is a PLC shortage. In the last cycle the ISM index crossed 60 in October 2020 and Rockwell's ControlLogix bulletin did not pass 50 days until February 2022, sixteen months later, and read 200 five months after that. That gap is the buyer's window, and it is the stress case to run in the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) now rather than then, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) PLC line has a watch flag for exactly this reason: standard lead time stock, component indices rising, no OEM bulletin yet. The row as it stands reads: standard lead time Stock (Rockwell's own column), last OEM bulletin March 29, 2023 at 162 days for the ControlLogix CPU, watch flag on since March 2026, the month ECIA's semiconductor share reached 78% and ISM put semiconductors back on its list. The structural difference from 2021 is that the OEMs did the second-sourcing. Rockwell's 1,000 components with a secondary source, its duplicated core product lines and added circuit-board capacity are real, and they are why the FY2025 10-K's backlog is $2.9 billion rather than $5 billion. The structural similarity is that a ControlLogix CPU is still built on a handful of microcontrollers and memory parts that Rockwell buys, not makes, and that every automation OEM buys from the same fabs. If memory allocation tightens the way ECIA's members say it is tightening, the first sign will be an OEM's standard lead time quietly changing from "Stock" to a number, and the second will be a distributor bulletin. We will print both the day we see them. ## What to Do Monday The last cycle is a dataset. The OEM bulletin was the best public signal and it lagged the component indices by twelve to sixteen months, so watch ISM's commodities list and ECIA's monthly share, not the distributor. Both are in the Monitor. The order book was noise in both directions, so size your spares to your own failure rate and lead time, not to what the distributor says is on order for you. And price does not round-trip: every increase taken during a shortage stays, and the tariff surcharge on top is recoverable only if the paperwork says so. For a plant or an OEM buying Rockwell today, the Monday list is short. Pull your installed base of controllers, safety controllers, analog I/O and servo drives by catalog number and build the D list, the single-source items that were over 26 weeks in 2022. That list is [plc-lead-times.csv](data/plc-lead-times.csv) sorted by peak. Put the D list into the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) with the 2022 peak as the stress case and see which project dates move. Check your open POs for a tariff surcharge line and your frame agreement for a refund clause before Rockwell's refund process opens. The 10-Q says qualified customers, and qualified is a paperwork test you can pass this week. And write the [RFQ](https://open-factory.com/articles/rfq-capital-equipment/) for the next panel so that the controller family is a requirement and the I/O, HMI and drive brands are not. One more thing the record says. Rockwell's bulletin listed the ControlLogix CPU at 41 days in November 2021, when the ISM index had already been above 60 for a year and electronic components had been in short supply for eleven months. A buyer who ordered in November 2021 got a controller in six weeks. One who waited for the bulletin to confirm the problem, in July 2022, waited twenty-nine. The bulletin is a lagging confirmation, not a warning. The warning was on ISM's commodities list, free, on the first business day of every month, and it is there again now, fourteen months into a new streak. Behind the paywall: the second-sourcing playbook. Which functions in a Rockwell architecture can move to Siemens, Beckhoff or a Codesys runtime without rewriting the controller program, which cannot under IEC 61508, how to hold spares when a repair costs **$5,600** and a sealed surplus CPU costs $25,920, and the five frame-agreement clauses that guarantee allocation, biweekly lead-time reporting and price hold the next time the bulletin reads 200 days. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists the PLC line and the watch flag for Pro subscribers. --- # Grainger vs McMaster vs Amazon Business: A 50-SKU Price Basket URL: https://open-factory.com/articles/mro-price-basket/ Date: 2026-05-14 Procurement / TCO Three of six storefronts blocked, Fastenal 126% above the cheapest, Amazon cheapest in 10 of 12 categories, Grainger's 811-line Hot List at 50% off list, 136,638 vending machines, the buying rule by category We priced the same 50 maintenance items at six distributors on one day, 10 September 2026, and the dearest storefront that would show us a price charged a median 126% more than the cheapest for the same part. A [3M 2097 P100 filter pair](https://www.mcmaster.com/5834T203/) was $8.80 at McMaster-Carr, $9.24 on Amazon and [$29.78 at Fastenal](https://www.fastenal.com/product/details/1001885). A [Master Lock 410RED](https://www.amazon.com/dp/B00192092C) lockout padlock was $12.99, $17.71 and $29.60. Three of the six sites, Grainger, its Zoro subsidiary and MSC, would not show a browser without a login any price at all. This piece covers what each storefront let us read, the spread by category, who is cheapest item by item, the size of the convenience premium, what Grainger's public cooperative contract says its list price is worth, why the vending machine on your floor is a pricing contract, and behind the wall the basket as a CSV, the buying rule by category, and the vending math. Across every SKU a storefront priced where at least one other storefront also priced it, Fastenal's median premium over the cheapest was 126.1% (n=43), McMaster-Carr's was 37.3% (n=38), and Amazon's was 0.0% (n=47) because Amazon was the cheapest on 34 of the 48 SKUs with a comparison. Only 16.3% of Fastenal's prices landed within 10% of the cheapest, against 34.2% of McMaster's and 76.6% of Amazon's. ## Three of Six Storefronts Would Not Show a Price The basket is 50 items a maintenance crib actually orders: Grade 5 hex bolts and nylon lock nuts, a Cubitron cut-off wheel and a Scotch-Brite 7447 pad, a 1/4 inch jobber bit and a 3/8-16 spiral point tap, a 3M 6200 half mask with 2097 filters and 8511 N95s, HyFlex 11-800 gloves and E-A-R Classic plugs, a Square D QO120 and an Eaton BR120, 500 feet of 12 AWG THHN, Ideal 74B connectors, a Leviton 5362-W, a SharkBite coupling, a PVC elbow, a brass ball valve, a 6205-2RS bearing, a Gates A38, a Climax collar, a UCP205-16 pillow block, WD-40, Brakleen, Loctite 243, Red N Tacky, WypAll X70, Simple Green, a Brute can and liners, Klein cutters and strippers, a Channellock 430, a FatMax tape, a Master Lock 410, a Justrite can, a wet floor sign, Pig mats, and three pleated filters. The full list with specifications is in [data/basket.csv](data/basket.csv), one row per SKU per storefront, 300 rows. We tried to read all six storefronts with the same headless Chrome, from a San Francisco IP, with no account and no cookies, on 10 September 2026. Three let us in. [McMaster-Carr](https://www.mcmaster.com/91251A540/) showed every price and pack size with no login ("$12.42 per pack of 50"). [Fastenal](https://www.fastenal.com/product/details/110120382) showed an "Online Price" and a unit price on every product page once the browser presented an ordinary Chrome user-agent string; with the default headless string it returned Akamai's "Access Denied". Amazon showed price, seller and item model number on every product page. The other three did not: grainger.com and zoro.com returned a DataDome captcha interstitial on every product and search URL, and mscdirect.com returned an Imperva challenge frame and, to a plain HTTP fetch, "MSCdirect.com is currently not available for your use." The table shows Amazon priced 49 of 50 SKUs (the fiftieth, a 3M cut-off wheel, showed a price but no count per box, so we did not compute a unit price), Fastenal 43 (3 not carried, 4 not found on the day), McMaster-Carr 38 (12 not found, mostly branded Klein, Ansell and Milwaukee items it does not list), and Grainger, Zoro and MSC 0 of 50 each. The rows for the blocked sites carry the URL and the error reference so anyone can repeat the attempt. We did not estimate one blocked price. Zoro is Grainger's own "endless assortment" business, the one its [2025 10-K](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000277135&type=10-K) says serves customers who "are focused on transparent pricing", and it is the one that would not let a robot read a price. A bot wall says nothing about the number behind it, so we do not treat the block as a verdict on Grainger's prices. It does mean the basket below compares three sellers, not six, and that the Grainger column comes from a different public document, which we get to below. The three storefronts we could not read are the three a plant buyer already has a login for. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) exists because that asymmetry is the norm in this industry, not the exception. ## Where the Spread Lives: 345% on Abrasives, 54% on Fasteners For each SKU we took the unit price at each storefront (per each, per pair, per foot, per ounce, per towel as the item demands, at the largest pack no bigger than the basket quantity) and computed the spread as the highest divided by the lowest, minus one. Forty-eight of the 50 SKUs had two or more prices; 32 had all three. The medians by category are abrasives 345%, PPE 238%, cutting tools 206%, janitorial 155%, electrical 130%, pipe and fittings 100%, lubricants 95%, hand tools 91%, bearings and power transmission 90%, safety 86%, HVAC filters 65%, and fasteners 54%. **The commodity categories with the most brand substitution, abrasives and PPE, carry the widest spreads. The categories where every seller sells the same catalog number, bearings and filters, carry the narrowest.** That pattern is the first thing to know about MRO pricing. A [6205-2RS bearing](https://www.mcmaster.com/5972K46/) was $12.26 at McMaster (unbranded), $12.30 on Amazon (SKF) and $12.67 at Fastenal (unbranded): a 3.3% spread, because the part number is the product and every buyer can check it in ten seconds. A 20x25x1 MERV 8 pleated filter was $5.62, $5.68 and $6.43: 14%. A [Channellock 430](https://www.mcmaster.com/5765A3/) was $18.95 on Amazon and $19.68 at McMaster: 4%. At the other end, a 5 inch 80-grit hook-and-loop sanding disc was [$0.14 on Amazon](https://www.amazon.com/dp/B0CQQ3BLJD) as a 50-pack from a brand nobody has heard of and [$2.02 at Fastenal](https://www.fastenal.com/product/details/0838765) as a five-pack of SIA Gold: 1,348%. Those are not the same disc, and the spread is partly a quality spread. It is also partly a tariff spread, since the generic disc is imported and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) puts Section 232 duty on the steel backing of a domestic one. We flag every such row in the CSV so the reader can decide which comparisons to trust. The spreads on identical parts are the ones that should bother a purchasing manager. [Loctite 243 in the 50 mL bottle](https://www.fastenal.com/product/details/0150201), same Henkel part, was $13.98 on Amazon, $48.57 at McMaster and $73.63 at Fastenal, a 427% spread on a bottle of glue. A [Stanley FatMax 33-725](https://www.mcmaster.com/6817A2/) was $20.48 (Amazon, two-pack), $28.31 (McMaster) and $47.37 (Fastenal). A [Klein D213-9NE](https://www.amazon.com/dp/B0000302W6) was $34.99 on Amazon and $71.42 at Fastenal. A Justrite 7150100 safety can was $64.67, $73.73 and $86.41. Five hundred feet of 12 AWG stranded THHN was [$135.00 on Amazon](https://www.amazon.com/dp/B002JFW00M) (Southwire 22964101), $173.16 at McMaster and [$353.05 at Fastenal](https://www.fastenal.com/product/details/0700360) (Southwire): 27, 35 and 71 cents a foot for the same copper, on a day when the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) puts the metal in that spool at well under a quarter of the highest price. ## Who Is Cheapest, Item by Item Counting SKU wins, Amazon was the cheapest storefront most often in ten categories, McMaster-Carr in two (fasteners, where it won 3 of 5, and pipe and fittings, a one-one-one tie), and Fastenal in none; Fastenal won three SKUs outright, the absorbent pads ($0.56 a pad against $0.64 on Amazon), a Hubbell duplex receptacle ($6.45 against a Leviton 5362-W at $12.99) and a Watts brass ball valve ($19.36 against an Apollo at $33.56). McMaster's premium over the cheapest is 0% on fasteners, 9% on lubricants and 10% on bearings, then 48% on abrasives, 64% on electrical and 137% on cutting tools. Fastenal's premium is 5% on filters and 34% on safety, then 81% to 251% everywhere else. | Category | SKUs | McMaster wins | Fastenal wins | Amazon wins | Cheapest most often | McMaster premium | Fastenal premium | Amazon premium | |---|---|---|---|---|---|---|---|---| | Fasteners | 5 | 3 | 0 | 2 | McMaster-Carr | 0% | 46% | 54% | | Abrasives | 4 | 1 | 0 | 3 | Amazon | 49% | 251% | 0% | | Cutting tools | 4 | 0 | 0 | 4 | Amazon | 137% | 206% | 0% | | PPE | 5 | 1 | 0 | 4 | Amazon | 83% | 160% | 0% | | Electrical | 4 | 1 | 1 | 2 | Amazon | 64% | 81% | 23% | | Pipe and fittings | 3 | 1 | 1 | 1 | tie | 50% | 116% | 50% | | Bearings and PT | 4 | 2 | 0 | 2 | Amazon | 10% | 90% | 0% | | Lubricants | 4 | 1 | 0 | 3 | Amazon | 9% | 95% | 0% | | Janitorial | 4 | 1 | 0 | 3 | Amazon | 87% | 155% | 0% | | Hand tools | 4 | 0 | 0 | 4 | Amazon | 21% | 104% | 0% | | Safety | 4 | 0 | 1 | 3 | Amazon | 40% | 34% | 0% | | HVAC filters | 3 | 0 | 0 | 3 | Amazon | 48% | 5% | 0% | *Source: Open Factory [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), MRO basket worked example. CSV: [data/cheapest-by-category.csv](data/cheapest-by-category.csv)* Two things the table hides. First, McMaster-Carr lists what it lists and nothing else: no Klein, no Ansell, no Milwaukee, no Eaton BR, no WypAll, and its 12 not-found rows are mostly those. Where it does list the item, it is often the second-cheapest by a small margin and the only one of the three that states the pack size, the spec and the delivery date on the same line. Second, Amazon's wins on commodity items are frequently a different product: the [1/2-13 x 2 Grade 5 bolt](https://www.amazon.com/dp/B00OHUIBMS) at $0.84 each is a Hard-to-Find Fastener 50-pack sold by Amazon itself, but the flap disc, the sanding disc, the pillow block and the end mill are marketplace brands with no datasheet. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) will eventually carry contributor grades on exactly this question, whether the cheap generic held up. The basket cannot, and neither can the [$2,850 category reports](https://open-factory.com/articles/2850-report-says-nothing/) that price nothing at the SKU level. The ten widest spreads run from 1,348% on the sanding disc to 245% on the HyFlex glove ($2.75 a pair on Amazon in a dozen, $9.48 at Fastenal in a dozen), and six of the ten are consumables a vending machine dispenses: discs, pads, drill bits, gloves, a threadlocker bottle and a filter. | SKU | McMaster | Fastenal | Amazon | Cheapest | Spread | |---|---|---|---|---|---| | Hook-and-loop sanding disc 5 in 80 grit | not listed | $2.02 | $0.14 | Amazon | 1,348% | | Zirconia flap disc 4-1/2 x 7/8 60 grit | $10.67 | $6.91 | $2.00 | Amazon | 434% | | Trash can liner 55 gal 1.5 mil black | $0.83 | $1.93 | $0.36 | Amazon | 432% | | Loctite 243 threadlocker 50 mL | $48.57 | $73.63 | $13.98 | Amazon | 427% | | Pillow block bearing UCP205-16 | $16.49 | $39.98 | $7.89 | Amazon | 406% | | Jobber drill bit 1/4 in HSS black oxide | $4.08 | $6.09 | $1.33 | Amazon | 358% | | Absorbent mat pad 15 x 20 lightweight | $2.12 | $0.56 | $0.64 | Fastenal | 282% | | 20 A industrial duplex receptacle | $23.14 | $6.45 | $12.99 | Fastenal | 259% | | 3M Cubitron II cut-off wheel 4-1/2 x .045 | $3.92 | $13.95 | not comparable | McMaster-Carr | 256% | | Ansell HyFlex 11-800 glove, per pair | not listed | $9.48 | $2.75 | Amazon | 245% | *Source: Open Factory [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), MRO basket worked example. CSV: [data/sku-spreads.csv](data/sku-spreads.csv)* ## The Convenience Premium Has a Name The word for what Fastenal charges above Amazon is not markup, it is service, and Fastenal is explicit about it. Its 2025 10-K reports [45.0% gross margin](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000815556&type=10-K) against Grainger's 39.1%, and it explains the model as a branch, a bin and a machine "right to the point of consumption within customers' facilities", with 65% of sales to national accounts on contract. The web price we captured is the price for the customer who has none of that: no contract, no bin, no rep, and no plant on the delivery route. It is the ceiling from which the contract discount is measured, and the size of the ceiling is the finding. The premium is largest where a plant is least likely to check. Among the 43 SKUs Fastenal priced, the premium was under 10% on 7 (the filters, the bearing, the receptacle, the pads, the valve, and two fasteners) and over 100% on 24. McMaster-Carr's premium over the cheapest was under 10% on 13 of 38 SKUs and over 100% on 6, four of them cutting tools and abrasives where its unbranded catalog item met a marketplace generic. The buyer who orders bearings and filters from a rep is paying a few points. The buyer who orders gloves, discs and threadlocker from the same rep is paying double. Amazon's floor has its own cost. Of Amazon's 34 wins, 11 were marketplace or no-name brands, and 9 of the 49 priced items were sold by third parties whose stock read "only 3 left" or "only 4 left" on the day. The Square D QO120 was cheapest on Amazon at $11.70 apiece, but only as a [ten-pack from a marketplace seller](https://www.amazon.com/dp/B07RN8ZCS3); the single QO120CP from Amazon itself showed no price. The [Rubbermaid Brute](https://www.amazon.com/dp/B01N6SRFOT) at $63.82 was a third-party listing with 20 in stock, and Rubbermaid's own listing at $53.27 was out of stock. Amazon Business, which we could not log into, sells [Savings Insights and Spend Anomaly Monitoring](https://press.aboutamazon.com/2026/7/amazon-business-reaches-60-billion-in-annualized-gross-sales-serving-over-11-million-organizations-worldwide) to 11 million organizations doing $60 billion a year, which is Amazon telling its buyers that the price they see is not the price they are paying somewhere else. ## The Basket Total: $1,782, $4,254, $5,812 At the basket quantities in the CSV (100 nuts, 500 feet of wire, 200 pairs of earplugs, 12 filters and so on) the 32 SKUs priced at all three storefronts cost $1,782.09 on Amazon, $4,253.52 at McMaster-Carr (+139%) and $5,812.21 at Fastenal (+226%); strip out the 9 SKUs where Amazon's item is a marketplace or no-name brand and the 23-SKU brand-name basket is $1,463.08, $3,514.29 and $5,244.17. **A crib that fills this order at Fastenal's web price pays $4,030 more than the same order on Amazon, and $1,559 more than at McMaster-Carr, before anyone negotiates.** The gap is not driven by one item. Removing the wire, the earplugs and the threadlocker, the three largest dollar spreads, still leaves Fastenal at roughly double Amazon. Nobody with a Fastenal contract pays the web price, which is the point of the next section: the web price is the reference from which a discount is quoted, and a discount is only as good as the number it is taken from. Grainger made this argument itself in 2017, when it admitted its list prices had become fiction and cut them. We told that story in [Grainger's 2017 Confession](https://open-factory.com/articles/grainger-confession/). The 2026 basket says the list is still fiction at the storefronts that publish one, and that the storefronts with the most to lose from comparison are the ones that block it. ## What Grainger Filed With Sourcewell Grainger will not show a robot its web price, but it filed 811 of its prices with a public purchasing cooperative. Sourcewell, the Minnesota cooperative whose contracts any US public agency can use, awarded Grainger contract [091422-WWG](https://www.sourcewell-mn.gov/cooperative-purchasing/091422-wwg) for facility MRO in November 2022 (maturing 8 November 2028), and posts a 14-page ["Hot List"](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/00004572/Additional%20Documents/Grainger%20091422-WWG%20Price%20Information.pdf) that gives, for each Grainger SKU, a "US Contract Reference Price" and a "US Contract Price". The PDF was generated on 19 December 2025. We parsed all 811 rows into [data/sourcewell-grainger-hotlist.csv](data/sourcewell-grainger-hotlist.csv). The distribution reads: 5% of items at under 20% off, 11% at 20 to 30%, 12% at 30 to 40%, 21% at 40 to 50%, 25% at 50 to 60%, 14% at 60 to 70%, and 11% at 70% or more; the median is 50.4%, the interquartile range 36.9% to 60.1%. **The median public-contract price on Grainger's own Hot List is 50.4% below Grainger's reference price, and one item in four sells at 60% off or more (n=811, Dec 2025).** A Dayton A38 V-belt carries a reference price of $11.04 and a contract price of $2.69. A box of 100 Ideal 600 V twist-on connectors is $19.64 and $7.05. A 3/4 inch schedule 40 PVC elbow is $0.64 and $0.18. Sorted into our twelve categories by keyword, the median discount is 75% on bearings and power transmission (n=39, nearly all Dayton belts), 65% on cutting tools (n=16), 59% on PPE (n=44), 51% on janitorial (n=167) and fasteners (n=32), 49% on safety, 46% on lubricants and electrical, 42% on filters, 40% on pipe and fittings, 36% on hand tools and 28% on abrasives (n=11). The order is close to the inverse of our web spread chart: where the web spread between storefronts is widest, abrasives, Grainger's contract discount is smallest, and where the Grainger discount is deepest, belts, the web spread is narrow. The reference price is doing different work in different aisles. The Hot List is, in the [contract's](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/00004572/Contract%20Documents/Grainger%20Contract%20091422.pdf) own words, a "specific selection" offered "at discounts greater than those listed in the Contract", so these are Grainger's deepest public prices, not its average contract price. The category schedule that governs the rest of the catalog was uploaded as "Pricing-091422.zip" and is not posted. Fastenal's schedule for the same solicitation is posted, and it is one page. [Contract 091422-FAS](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/00004574/Additional%20Documents/091422-FAS%20Pricing%20Information.pdf) gives a flat percentage off catalog by category: 57% on fasteners, 40% on wipers and rags, 35% on filters, trash liners and hand protection, 30% on electrical, plumbing, janitorial and safety, 28% on power transmission and motors, 25% on abrasives, cutting tools, lubricants and hand tools, 23% on power tools. Sourcewell's evaluators scored twenty bidders on price out of 400 points: Grainger took 313 and ranked first overall on everything else, Fastenal 321 and third, Motion Industries 339 and second, MSC's Sid Tool Co. 311 and fifth ([proposal evaluation](https://files.sourcewell.org/public/Shared%20Documents/Solicitations/10590/Solicitation%20Documents/Prop.%20Eval.-Facility%20MRO%20091422.pdf)). None of the three storefronts that blocked us won on price, and the one that ranked first scored twelfth of twenty on it. ## Grainger's Reference Price Against the Web Ten items in the basket have a near-twin on the Hot List, which gives a Grainger column of a kind: not a web price captured on the day, but Grainger's own December 2025 reference price and its public contract price for the same or equivalent item, against the three web prices we did capture. | Item | Grainger reference (Dec 2025) | Sourcewell contract | McMaster web | Fastenal web | Amazon web | Cheapest web vs reference | |---|---|---|---|---|---|---| | WD-40 16 oz can | $14.68 | $8.61 | $9.92 | $20.71 | $11.78 | -32% | | Simple Green 1 gal | $25.60 | $13.53 | $18.64 | $31.85 | $9.98 | -61% | | Pleated filter 20x25x1 MERV 8 | $9.77 | $5.41 | $6.43 | $5.68 | $5.62 | -42% | | Pleated filter 16x25x1 MERV 8 | $8.75 | $5.15 | $9.64 | $6.12 | $5.83 | -33% | | Yellow twist-on connectors, box of 100 | $19.64 | $7.05 | $13.72 | not found | $20.13 | -30% | | 3M half mask respirator | $25.77 | $16.46 | $32.96 | $37.81 | $17.99 | -30% | | A38 V-belt (Dayton vs Gates) | $11.04 | $2.69 | $12.24 | $14.72 | $10.13 | -8% | | CRC brake parts cleaner, 20 oz basis | $12.06 | $5.70 | $12.26 | $16.59 | $11.25 | -7% | | Klein wire stripper 18 to 10 AWG | $32.73 | $19.56 | not listed | $37.38 | $20.99 | -36% | | PVC 3/4 in sch 40 90 elbow | $0.64 | $0.18 | $0.69 | $2.29 | $1.04 | +8% | *Source: Open Factory [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), MRO basket worked example; Sourcewell 091422-WWG. CSV: [data/grainger-reference-vs-web.csv](data/grainger-reference-vs-web.csv)* Read across: on eight of the ten the cheapest web price is 7% to 61% under Grainger's reference price and within a few dollars of the Sourcewell contract price; on the filters, Amazon and Fastenal's web prices are within 10% of what a public agency pays Grainger on contract; on the elbow and the belt, Grainger's contract price is far below anything a walk-up web buyer can get anywhere, and the reference price is roughly what the web charges. **Grainger's reference price is not a market price; on this evidence it is a ceiling set 30% to 60% above what the same item costs to anyone who looks, and the contract discount returns the buyer to the market.** That is the mechanism Grainger described in its own [April 2017 release](https://www.sec.gov/Archives/edgar/data/0000277135/000027713517000013/gww8kex991q12017.htm): a "pricing model that applied contractual discounts to high list prices", which made it "difficult to win new business", followed by new web prices on 450,000 SKUs and a 1.7-point drop in US gross margin as "customers with access to lower pricing bought more than company expectations". Nine years on, the web prices exist and a robot cannot read them. ## The Vending Machine Is a Pricing Contract The basket above is what a buyer sees with a browser. Most industrial MRO is not bought that way. Fastenal ended 2025 with 136,638 weighted FASTBin and FASTVend devices installed on customer floors, signed 25,892 more during the year (102 a day), and booked $2,675.0 million of sales through them, 32.2% of company revenue; with FASTStock bins the managed-inventory total is $3,712.7 million, 44.7% of sales ([10-K FY2025](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000815556&type=10-K)). MSC had [29,611 vending machines](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001003078&type=10-K) in service and 411 in-plant programs at fiscal year end in August 2025, up from 27,003 and 342. Grainger's KeepStock is the same product without a device count in the 10-K. Divide Fastenal's device sales by its device count and each machine-equivalent moved about $19,600 of product in 2025. The 10-K lists the customer benefits as "reduced consumption, reduced purchase orders, reduced product handling, and 24-hour product availability" and does not list price. A vending machine is a supply agreement in a steel box: the SKUs inside are chosen once, the prices are set by the contract behind the machine, and replenishment is automatic. Nobody at the machine sees a price, and nobody in purchasing re-quotes a glove the machine has dispensed 4,000 times. Six of the ten widest spreads in our basket are vending SKUs. Fastenal's gross margin in 2025 was 45.0%; Grainger's High-Touch segment ran 41.7% and its Endless Assortment segment, the transparent-price web business, ran 29.9%. The twelve-point gap between the two Grainger segments is the price of the rep and the bin, and the plant pays it on every glove. None of this makes the machine a bad deal. ISM's August 2026 report puts the [average lead time for MRO supplies at 48 days](https://www.textileworld.com/textile-world/2026/09/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-manufacturing-expanded-for-the-eighth-consecutive-month/), and a stock-out on a $2 glove can idle a $200-an-hour line; we made the same argument about the spares list in [The Recommended Spares List Is Where the Margin Hides](https://open-factory.com/articles/recommended-spares-margin/) and about the 171-day capex queue in [The 171-Day Capex Lead Time](https://open-factory.com/articles/171-day-capex-lead-time/). The machine buys availability. What the basket shows is what availability costs when the price inside the machine is never checked against the price outside it, and that the check takes an afternoon. ## What to Do Monday Pull twelve months of MRO spend from the ERP, sort by extended dollars, and take the top 50 lines. Price them on McMaster-Carr and Amazon with a browser. Both show prices without a login, and the [basket CSV](data/basket.csv) is a template with the columns you need (part number, pack, unit, URL, date). Then put your contract price beside them. If the contract price is within 10% of the cheapest web price, as it was on the filters and the bearing here, leave the line alone. The rep is earning the margin. If it is 50% above, as it was on the threadlocker, the gloves, the earplugs and the wire, you have found the lines the machine is paid to hide. Send the list to the rep with the URLs and ask for the category discount off the web price, not off list, in writing; Fastenal's Sourcewell schedule is public and 30 to 57% off catalog is what a public agency gets with no negotiation. Ask the vending rep for the price file behind the machine, dated, and put a 90-day re-quote clause on it, the same clause the [capital equipment RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) puts on escalation. And when the capital side of the same plant is buying switchgear or a transformer, run the quote through [Quote Check](https://open-factory.com/tools/quote-check/), because the same list-versus-net structure sits under a $400,000 lineup as under a $14 can of WD-40, and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is the public record for that. Behind the paywall: the full 50-SKU basket with all 300 rows as a CSV, the category-by-category buying rule that names the storefront and the caveat for each of the twelve aisles, and the vending math, including what the 32-SKU basket costs at Fastenal's public **57% and 25 to 35%** contract discounts and what that implies for the roughly **$19,600** a year that passes through each of its 136,638 machines, built on the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) worked example and the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) contribution form. --- # A Factory Owner's Guide to the PJM Capacity Charge URL: https://open-factory.com/articles/pjm-capacity-factory-bill/ Date: 2026-05-08 Power / How-to $28.92 to $333.44 per MW-day, the $7.27B datacenter attribution, five hours that set a Pennsylvania plant's bill, ELRP at $112k per MW-year, the $12B modeling fight, the PLC playbook A 5 MW plant in Met-Ed territory will pay about $568,000 for capacity in the PJM delivery year that starts June 1, 2026. In the year that started June 1, 2024 the same plant, with the same load, paid about $50,000. Nothing on the plant floor changed. The [RTO clearing price went from $28.92 to $329.17 per MW-day](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2026-2027/2026-2027-bra-report.pdf) in two auctions, and it is billed against five hours of load from the summer before. At $28.92 the plant's 4.73 MW obligation cost $49,924 a year; at $269.92 it cost $465,957; at $329.17 it costs $568,239, and at the 2027/28 cap of $333.44 it will cost $577,187 because that delivery year has 366 days. This piece covers where the price came from, what the market monitor blames, exactly how the charge reaches a Pennsylvania and an Ohio bill, what PJM's emergency load program pays you to give back, the fight over whether the number should have been half, and the five-hour playbook that cuts it. ## Three Auctions, One Price Cap PJM buys capacity three years ahead in the Base Residual Auction, in $/MW-day of unforced capacity, against a demand curve built from its own load forecast. The 2024/25 auction cleared at $28.92. The [2025/26 auction cleared at $269.92](https://www.pjm.com/-/media/DotCom/about-pjm/newsroom/2024-releases/20240730-pjm-capacity-auction-procures-sufficient-resources-to-meet-rto-reliability-requirement.pdf) for most of the footprint, $466.35 in BGE and $444.26 in Dominion, with 135,684 MW procured and a total bill of $14.7 billion. PJM's own drivers list: about 6,600 MW of retirements, a peak forecast up from 150,640 MW to 153,883 MW, and new risk modeling that cut the accredited value of gas plants. Eleven auctions, and the last three sit within $8 of each other because a cap now holds them there: $329.17 for 2026/27, $333.44 for 2027/28, $325.00 for 2028/29. The collar came out of Pennsylvania's complaint at FERC (docket ER25-1357) and was set at $256.75/MW-day in installed-capacity terms, which converts to a different UCAP number each year as the reference unit's accreditation moves. [FERC extended it on April 28, 2026](https://insidelines.pjm.com/ferc-approves-capacity-auction-price-collar-to-next-two-capacity-auctions/) to the 2028/29 and 2029/30 auctions at roughly $325 cap and $175 floor, with all 13 governors and the White House National Energy Dominance Council in support. The cap is not a market price. In the 2026/27 auction [every LDA cleared at exactly $329.17](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2026-2027/2026-2027-bra-report.pdf) and the RTO cleared 134,311 MW against a 146,105 MW reliability requirement, 139 MW over the mark by PJM's count once fixed-resource-requirement zones are added. **The 2027/28 auction was the first in PJM history to fall short of the reliability requirement across the whole RTO, by 6,517 MW.** It [procured 134,479 MW at the $333.44 cap](https://www.pjm.com/-/media/DotCom/about-pjm/newsroom/2025-releases/20251217-pjm-auction-procures-134479-mw-of-generation-resources.pdf), a 14.8% reserve margin against a 20% target, and PJM said the peak forecast had risen 5,250 MW of which nearly 5,100 MW was data centers. Then the [2028/29 auction on July 14, 2026 procured 138,318 MW at the $325 cap](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2028-2029/2028-2029-bra-results-report.pdf) and came in 6,831 MW short, a 14.7% reserve margin, with only 525 MW of new generation clearing. PJM told reporters that [without the collar the RTO would have cleared near $555/MW-day and ComEd at $777](https://www.utilitydive.com/news/pjm-capacity-auction-price-cap-reserve-shortfall/825282/), a $29.7 billion year instead of $16.4 billion. The collar saves your plant about $390,000 a year against that counterfactual, and it also stops the price from telling anyone to build. ## The $23 Billion Data Center Attribution Monitoring Analytics, the independent market monitor, reruns each auction with the data center load taken out of the forecast. For 2026/27 it found that [including 11,993 MW of existing and forecast data center load raised auction revenue by $7,271,197,971, or 82.1%](https://www.monitoringanalytics.com/reports/Reports/2025/IMM_Analysis_of_the_20262027_RPM_Base_Residual_Auction_Part_A_20251001.pdf), from $8.85 billion to $16.12 billion. For 2027/28 the forecast carried [17,071 MW of data center load and the attribution was $6,497,653,512, or 65.5%](https://www.monitoringanalytics.com/reports/Reports/2026/IMM_Analysis_of_the_20272028_RPM_Base_Residual_Auction_Part_A_20260105.pdf). Existing data centers accounted for 4,053 MW of that; 13,018 MW was forecast growth. The three attributions total $23.1 billion on $47.2 billion of auction revenue: $9.33 billion of $14.7 billion in 2025/26, $7.27 billion of $16.12 billion in 2026/27, $6.50 billion of $16.41 billion in 2027/28. Joseph Bowring, the monitor's president, told Utility Dive the 2028/29 number is [$6.3 billion, or 38%, and $29.4 billion across the last four auctions](https://www.utilitydive.com/news/pjm-data-centers-capacity-auction-imm-bowring/825626/). His standing recommendation since 2025 is that new data center load "bring their own new generation" so that the forecast uncertainty stops landing on everyone else. PJM's board adopted a version of that on July 27, 2026: [incremental new large loads above the 2028/29 forecast will be excluded from the demand used in auctions from 2029/30 on](https://insidelines.pjm.com/pjm-board-directs-action-on-resource-adequacy-affordability-and-large-loads/), and a Large Load Registry will track every site of 50 MW or more. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) lists the PJM zones next to ERCOT, Georgia and Arizona for a 50 MW connection, and in every PJM zone the capacity line is now the second-largest number on the sheet after energy. The [EIA's June 2026 industrial rates](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a) show Pennsylvania at 10.10 cents/kWh against 9.31 a year earlier and Ohio at 9.89 against 8.66, while Texas moved from 6.36 to 6.58. The [1.6% funnel in ERCOT](https://open-factory.com/articles/ercot-funnel/) is a queue problem; PJM's is a bill problem. ## Five Hours Set Your Bill Your capacity charge is not billed on your peak demand, your contract capacity or your kWh. It is billed on your Peak Load Contribution, and PJM Manual 19 defines how the PLC is built: for each summer, from June 1 through September 30, PJM adds estimated load drops back to metered load, then picks [the five highest non-holiday weekday RTO daily peaks, the 5CP](https://www.pjm.com/-/media/DotCom/documents/manuals/m19.pdf), and publishes them in mid-October. Your utility averages your metered load in those five hours, grosses it up for line losses, and that is your PLC from the following June 1 to May 31. A plant that ran flat out at 6 pm on five days in 2025 pays for it every day from June 2026 to May 2027. The [five 2025 hours](https://www.pjm.com/-/media/DotCom/planning/res-adeq/load-forecast/summer-2025-peaks-and-5cps.pdf) were June 23 at 160,649 MW, June 24 at 160,629 MW, July 29 at 156,035 MW, June 25 at 152,903 MW and July 28 at 151,525 MW, four of them hour ending 18:00 and one at 15:00, and the fifth-place day was only 1,378 MW above the sixth. PJM's own review of that heat wave records [preliminary peaks of 161,770 MW on June 23 and 162,401 MW on June 24, the third- and fourth-highest in its history, with more than 4,000 MW of demand response deployed](https://insidelines.pjm.com/maintaining-grid-reliability-through-highest-peaks-in-a-decade/). The add-back rule matters: load that a PJM event removed at 6 pm on June 24 is put back into the PLC, so being paid to curtail in a PJM event does not lower next year's tag. Curtailing on your own does. Two utilities, two documents. AEP Ohio's [Settlement Policies](https://www.aepohio.com/lib/docs/company/about/choice/OH/2018/AEPOhioSettlementPolicies-Rev-5-2018.pdf) spell it out: for interval-metered accounts "the actual hourly usage at those five hours is averaged," losses are applied to reach the generation level (1.0932 secondary, 1.0552 primary, 1.0341 sub-transmission in the 2018 revision), add-backs apply only to PJM program curtailments and not to a customer that "intentionally curtails on their own for operational reasons," and the new PLC tags go out to suppliers each December by EDI 814C, [effective June 1 through May 31](https://www.aepohio.com/company/about/choice/cres/calculations). FirstEnergy Pennsylvania's retail tariff defines PLC as "a Customer's contribution to the Company's transmission zone normalized summer peak load," and its [supplier tariff](https://www.firstenergycorp.com/content/dam/customer/Customer%20Choice/Files/PA/tariffs/FEPA-supplier-Tariff.pdf) posts each customer's PLC to suppliers within 10 days of calculating it. **From the tag to the dollar, PJM Manual 18 section 7.5 is one line: daily UCAP obligation equals PLC times the zonal RPM scaling factor times the Forecast Pool Requirement.** The [scaling factor and FPR are published with each auction](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2027-2028/2027-2028-bra-results.xlsx): for 2026/27 the FPR is 0.917, Met-Ed's base zonal factor is 1.01418 and AEP's is 0.96289; for 2027/28 the FPR is 0.926 with Met-Ed at 0.97785 and AEP at 0.91830; for 2028/29 the FPR is 0.9401 with Met-Ed at 0.94516 and AEP at 0.92242. Multiply the obligation by the zonal capacity price and by 365 days (366 in 2027/28) and you have the year's charge. The [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) does this for any zone and PLC. The arithmetic below is what it runs. ## The Same Plant, Four Bills Take a plant with a 5,000 kW PLC at the meter (its average draw in those five hours), primary-voltage service, and 30,000 MWh a year. In the Met-Ed zone the tariff's published primary-service loss multiplier is 1.0171, so the PLC at generation is 5,086 kW; times 1.01418 times 0.917 gives a 4.7295 MW obligation; times $329.17 times 365 gives $568,239 for 2026/27. In AEP Ohio the primary loss factor is 1.0552, so 5,276 kW; times 0.96289 times 0.917 gives 4.6585 MW; times $329.17 times 365 gives $559,711. That is $113.65 and $111.94 per kW of metered PLC per year, and it is what a competitive supplier in either state passes through, plus margin. The two supplier routes land within $8,500 of each other at $568,239 and $559,711. The two utility default-service routes land at $482,100 and $1,848,172. AEP Ohio's standard service offer bills capacity through a [Generation Capacity Rider of 1.607 cents/kWh for demand-metered primary GS customers from June 1, 2026](https://www.aepohio.com/lib/docs/ratesandtariffs/Ohio/September_2026_AEP_Ohio_Tariff_Book.pdf), which on 30,000 MWh is $482,100, or 15% under the PLC route because the rider is priced for the class's load factor rather than yours. Met-Ed's [Hourly Pricing Default Service Rider](https://www.firstenergycorp.com/content/dam/customer/Customer%20Choice/Files/PA/tariffs/fe-pa-retail-tariff.pdf), mandatory for Rate GP and TP customers who do not shop, bills a "Cap-AEPS-Other" component of $0.06057 per kWh from the April 2026 auction times a 1.0171 loss multiplier: $1,848,172 on 30,000 MWh, which is $370 per kW of this plant's PLC against $113.65 on the PJM formula. The rider is a class-average price for capacity, alternative energy credits and the supplier's other costs, and for a three-shift plant it is 3.3 times the auction cost of the same capacity by the tariff's own numbers. | Step | Met-Ed (PA) | AEP Ohio | Source | |---|---|---|---| | Average metered kW at the five 2025 CP hours | 5,000 | 5,000 | PJM 5CP file; AEP Settlement Policies | | Loss factor to generation | 1.0171 | 1.0552 | FE PA tariff p. 208; AEP p. 8 | | PLC at generation (kW) | 5,086 | 5,276 | computed | | Base zonal RPM scaling factor, 2026/27 | 1.01418 | 0.96289 | PJM 2026/27 BRA workbook | | Forecast Pool Requirement, 2026/27 | 0.917 | 0.917 | PJM 2026/27 BRA report | | Daily UCAP obligation (MW) | 4.7295 | 4.6585 | Manual 18 s7.5 | | Zonal capacity price ($/MW-day) | 329.17 | 329.17 | PJM 2026/27 BRA report | | Annual charge, 365 days ($) | 568,239 | 559,711 | computed | | Same plant, 2027/28 ($) | 561,976 | 547,521 | scaling 0.97785 / 0.91830, FPR 0.926, 366 days | | Same plant, 2028/29 ($) | 536,029 | 542,731 | scaling 0.94516 / 0.92242, FPR 0.9401 | [CSV: plant-arithmetic.csv](data/plant-arithmetic.csv) The charge drifts down $32,000 over three years for the Met-Ed plant even though the price barely moves, because Met-Ed's scaling factor falls from 1.014 to 0.945 as PJM's forecast for the zone flattens. Peter Cavan of Unison Energy put the same arithmetic to Utility Dive for a bigger site: [a 10 MW industrial customer's monthly capacity charge goes from about $6,000 in 2024 to about $70,000 in 2028](https://www.utilitydive.com/news/pjm-capacity-auction-price-cap-reserve-shortfall/825282/). A [2 MW service upgrade](https://open-factory.com/articles/2mw-plant-service-upgrade/) now carries about $230,000 a year of capacity exposure before the first kWh. The [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool adds it to the study fees and security for any PJM zone. ## ELRP: The Same Megawatt Sold Back PJM pays load to be curtailable. Manual 18 section 4.3 says a load management program offered as a Demand Resource and cleared in the auction [is paid the resource clearing price](https://www.pjm.com/-/media/DotCom/documents/manuals/m18.pdf) for the delivery year, and the resource must be registered in the Pre-Emergency or Emergency Load Response Program through a Curtailment Service Provider. Demand response cleared [7,299 MW in the 2027/28 auction, up from 5,531 MW, because its accredited value rose from 69% to 92%](https://www.pjm.com/-/media/DotCom/about-pjm/newsroom/2025-releases/20251217-pjm-auction-procures-134479-mw-of-generation-resources.pdf) once PJM required availability in all hours of the year. In 2028/29 it [fell 277 MW to 7,365 MW of unforced capacity](https://www.utilitydive.com/news/pjm-capacity-auction-price-cap-reserve-shortfall/825282/) despite the price. A megawatt nominated into ELRP for 2027/28 earns 0.92 times $333.44 times 366 days, which is $112,276 before the CSP's share, and that is the [$112k per MW-year Voltus quotes to customers](https://www.voltus.co/blog/pjm-capacity-auction-results-2027-2028); a megawatt cut from the PLC saves the Met-Ed plant $112,395. In 2026/27 the ELRP number was $82,901 because the accreditation was 69%. Voltus's published cases are a manufacturer that earned $75,000 and offset 55% of its 2025/26 capacity cost, and a data center that nominated 139 MW for $11 million, an 87% offset; enrollment for 2026/27 closed March 31, 2026. For the worked plant, 2 MW of curtailable compressors, chillers and one line is $224,552 a year gross in 2027/28, about 40% of the capacity charge. The obligation is real. A Capacity Performance demand resource must be [available for an unlimited number of interruptions in all hours of the delivery year from 2027/28](https://www.pjm.com/-/media/DotCom/documents/manuals/m18.pdf), and the non-performance charge is the shortfall in MW times a rate equal to the RTO Net CONE times the days in the year divided by 30, spread over the settlement intervals: at the [2027/28 RTO Net CONE of $68,347 per MW-year in installed-capacity terms](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2027-2028/2027-2028-planning-period-parameters-for-base-residual-auction-pdf.pdf) that is about $2,280 per MWh you promised and did not drop, capped for the year at 1.5 times the clearing price times the days times your committed MW, or roughly $183,000 per MW. PJM deployed demand response on June 23, 24 and 25, 2025, so a plant that cannot shed three afternoons running should not enroll. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool prices the alternative many plants use, running standby diesel through the event, with its Tier 4 permitting hours. ## The $12 Billion Fight Over the Reserve Study SemiAnalysis spent six months rebuilding PJM's Reserve Requirement Study from Manual 20A and the posted 2028/29 workbooks and published the result on August 16, 2026: [PJM understates its existing fleet by about 3.8 GW](https://newsletter.semianalysis.com/p/12b-of-us-ratepayers-money-wasted) because it derates thermal units for summer heat but never uprates them for winter cold, and because its outage statistics still carry Winter Storm Elliott, when 46 GW and 24% of the fleet failed, even though 400 of roughly 700 gas units had reported winterization upgrades by January 2024 and the fleet lost 9% in Storm Gerri, 9% in the January 2025 MLK storm and 10% in Storm Fern. With the two corrections, 2025/26 clears at $135 instead of $270 ($6.7 billion) and 2026/27 at $230 instead of $329 ($4.9 billion), $11.6 billion in a range of $8.0 to $14.5 billion. The bars are 6,831 MW and about 3,000 MW: the backstop procurement PJM plans for October is sized to the first number, and the second would have fit inside the voluntary subscription design that PJM's members endorsed on June 30, 2026 and the board rejected on July 27. PJM has answered the argument before. The market monitor found in September 2024 that crediting winter capability would have cut the 2025/26 bill by [$2.7 billion with the reserve margin held fixed and up to $8.0 billion without](https://www.monitoringanalytics.com/reports/Reports/2024/IMM_Analysis_of_the_20252026_RPM_Base_Residual_Auction_Part_A_20240920.pdf). PJM's [October 11, 2024 response](https://pjm.com/-/media/DotCom/library/reports-notices/reliability-pricing-model/20241011-response-to-imm-25-26-bra-report.ashx) agreed "there potentially exists additional thermal capacity available during winter months" but said the IMM ignored winter deliverability limits, and that if extra winter capability shifted risk into summer "this would increase, not decrease, the reliability requirement." PJM's own May 2025 sensitivity measured the winter uplift at 8,561 MW; its task force passed seasonal ratings 178 to 54 in July 2025; the senior committee killed the package on August 20, 2025 at 30.7% of the weighted vote; E3's December 2025 review said PJM "asymmetrically applies ambient derates." No vote since. **For the worked plant the difference between the two models is about $233,000 a year: at $135 the 2025/26 charge would have been $233,000 instead of $465,957.** We think the cold-air half of the argument is settled and the winterization half is a judgment call PJM should be making in the open rather than by default. Deliverability is the one objection that has not been tested, because PJM never ran the studies. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) lists the docket dates: comments on the backstop filing closed August 21, and the 2029/30 auction runs in December. ## The Backstop and What to Do Monday PJM [filed the Reliability Backstop Procurement at FERC on July 31, 2026 as docket ER26-3380](https://www.utilitydive.com/news/pjm-backstop-capacity-auction-ferc-data-centers/826792/): a one-time window from September 30 to October 21 for new resources against the 6,831 MW shortfall, contracts of up to 15 years, a MW-weighted average cap of $555/MW-day, results by December 2, resources online no later than June 1, 2032, and the target reduced by any bilateral contracts signed first. Costs go to zones pro rata to the procurement target and then to load-serving entities, and [PJM's board wrote that "because PJM does not have jurisdiction to allocate retail costs directly to individual data centers, state action will be essential"](https://insidelines.pjm.com/pjm-reliability-backstop-proposal-outlines-steps-to-secure-new-supply-and-maintain-reliability/). At the cap, 6.8 GW is $1.38 billion a year, or $24.19/MW-day spread across the 156,013 MW requirement. Whether a 5 MW plant sees it depends on whether Harrisburg and Columbus put it on loads of 50 MW and above, and neither has decided. Monday, in this order. Ask your supplier or utility for your current PLC tag and the five hours behind it; AEP Ohio sends it every December by EDI 814C and FirstEnergy posts it to the supplier lists. If you are on Met-Ed, Penelec or Penn Power hourly-priced default service with a load factor over 50%, get two supplier quotes with the capacity line broken out as PLC times the pass-through, because the rider is charging your plant $370 per PLC-kW for a $114 product. Put PJM's Hot Weather Alerts on the plant manager's phone and pre-plan a 2 MW shed for 3 to 7 pm on any June-to-September weekday PJM forecasts above ~148 GW. Five of those days in 2026 set your bill from June 2027 to May 2028. Call a CSP before next March 31 about 2027/28 enrollment, with June 23 to 25, 2025 as the test of whether you can drop load three afternoons running. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) has the PJM zones' capacity line beside Georgia, Texas and Arizona for anyone deciding where the next plant goes, and [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/) walks the rest of the sheet. Behind the paywall: the five-hour playbook that takes the worked plant's capacity line from $561,976 to **$337,186** before ELRP and to about $113,000 net with 2 MW enrolled, the battery that pays back in 2 to 6 years on the [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/)'s numbers, and the four December 2 outcomes priced for this plant. --- # A Robot Costs $61k. The Cell Costs $150k URL: https://open-factory.com/articles/robot-61k-cell-150k/ Date: 2026-05-08 Machinery / TCO IFR 621k installs, A3 $61,198 per arm, Fanuc CRX up 12% to $51k, UR20 at $54k to $73k in federal awards, three cells line by line, payback by shift, welder wages by state, the integration multiple and three contract clauses North American buyers ordered [36,766 robots for $2.25 billion in 2025](https://www.therobotreport.com/north-american-robot-orders-rise-by-6-6-percent-2025/), which is $61,198 per arm, and not one of those arms did anything on its own. A working cell, the arm plus the gripper, the guarding, the vision, the integrator's hours and the training, costs 1.9 to 2.4 times that. We priced three of the most common cells line by line from published distributor prices, federal award records and Cloos's published welding-cell ranges, then ran each against a fully loaded operator at the BLS median wage. This piece covers the market (IFR's 621,000 installs, the US at 38,000), what an arm costs by payload and brand, the three cells, where the other $87,000 goes, payback by shift and by state, and behind the wall, the integration multiple by seller type from 23 federal awards and the three contract terms a first-time buyer needs. Left to right: a cobot palletizer at $130,515 with a $62,615 arm, a cobot MIG welding cell at $113,336 with a $44,636 arm, and a cobot machine-tending cell at $144,432 with a $56,932 arm. Integrator labor at $150 an hour is $23,400 to $32,400 of each, and the tooling, safety, vision, spares, training and freight between the arm and the labor is $44,000 to $55,000. ## 621,000 Robots, 38,000 of Them Here IFR's preliminary count for 2025 is [621,000 industrial robots installed worldwide, up 15%](https://ifr.org/downloads/press_docs/2026_06_24_IFR_Executive_Roundtable_market_presentation.pdf), presented at Automate in June 2026 ahead of the World Robotics 2026 report due 24 September. The US installed [38,000, up 11%](https://ifr.org/ifr-press-releases/news/us-robot-industry-returns-to-double-digit-growth), its third-best year behind 2018 and 2022 after two years of decline, and IFR says China's 2025 installs are "about ten times" the US figure. Asia took 79% of the world's installs, Europe 13%, the Americas 9%, per [Manufacturing Dive's report from the floor](https://www.manufacturingdive.com/news/us-robotics-rebounded-2025-on-track-more-growth-ifr-automate-2026/823874/). The US installed 33,300 arms in 2019, 39,576 in 2022, 34,200 in 2024 and 38,000 in 2025; China went from 140,500 to 295,000 over the same window and the world from 373,000 to 621,000. Every row is in [data/ifr-installs-by-country.csv](data/ifr-installs-by-country.csv) with the release it came from, because IFR revises prior years in each edition. **Robot density in the US stands at 307 per 10,000 manufacturing employees, eighth in the world, against 1,220 in Korea, 449 in Germany and 446 in Japan.** IFR's [April 2026 density release](https://ifr.org/ifr-press-releases/news/robot-density-surges-in-europe-asia-and-americas) puts the world average at 132 and China at 166 (22nd) on revised labor data. The US mix in 2025: automotive over a third of units, food up 30% to about 3,000 installs, the unspecified non-manufacturing group up 41%, metal and machinery down 15%. Cobots were [19.6% of North American units and 10.7% of dollars in 2025](https://www.therobotreport.com/north-american-robot-orders-rise-by-6-6-percent-2025/), 7,212 arms for $241 million, and 28.6% of units in the fourth quarter alone. Material handling took 60% of first-quarter 2026 orders. The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) lists these series forward as each release lands. ## What an Arm Costs, by Payload and Brand A3 does not publish a price index, but it publishes units and dollars, and dividing one by the other gives the only transaction-weighted arm price that exists for this market. $2.25 billion over 36,766 units is $61,198; $241 million over 7,212 cobots is $33,417; the remaining 29,554 conventional arms averaged $67,977. In the [first half of 2026](https://www.therobotreport.com/q2-2026-robotics-demand-increased-across-industries-reports-a3/) the average rose to $64,796 on 17,995 units and $1.166 billion, with cobots at $41,096 and conventional arms at $69,113. Dollars grew 6.6% on 2.0% more units, the same mix shift toward bigger, more configured machines that the [$306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/) traced in AMT's machine tool data. The A3 number is an average of everything from a $12,800 SCARA to a $400,000 press-tending giant, so we built the model-level view from the two places prices are actually printed: Vention's parts library, which lists Fanuc CRX and Universal Robots arms at unit prices, and USAspending.gov, where every federal purchase order above the micro-purchase threshold carries a description and a price. Reading up the chart: a UR3e (3 kg) at $28,990 to $38,171 across two federal awards, a UR5e (5 kg) at $28,828 to $48,000 across six awards and one published price, a Fanuc CRX-10iA (10 kg) at [$51,433.73 on Vention in September 2026](https://vention.io/parts/fanuc-crx-10ia-collaborative-robot-arm-1923), a UR20 (20 kg) at $54,149 to $73,439 across three Army and NASA awards, and a [CRX-30iA at $67,988.64](https://vention.io/parts/fanuc-crx-30ia-collaborative-robot-arm-2155). [Standard Bots publishes $37,000 for its 18 kg Core and $49,500 for the 30 kg Thor](https://standardbots.com/blog/universal-robot-price), the only vendor-published list prices we found. Universal Robots, Doosan and Techman publish none. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) records the absence. Every observation, with award number and offer count, is in [data/arm-prices-observed.csv](data/arm-prices-observed.csv). **The same UR20 arm sold to the Army for $62,058 in August 2024 and, with a controller box, for $73,438.58 in June 2025, while NASA Langley paid $54,149 in June 2023: a 36% spread on one SKU in two years.** The three records are [W911QX24P0194](https://www.usaspending.gov/award/CONT_AWD_W911QX24P0194_9700_-NONE-_-NONE-), [W911QX25PA038](https://www.usaspending.gov/award/CONT_AWD_W911QX25PA038_9700_-NONE-_-NONE-) and [80NSSC23PB655](https://www.usaspending.gov/award/CONT_AWD_80NSSC23PB655_8000_-NONE-_-NONE-); the first drew four offers, the second one, the third eight. That is the pattern the [Quote Check](https://open-factory.com/tools/quote-check/) tool is built to catch: a single-offer buy lands at the top of the band. The Fanuc side moved for a different reason. Vention's [October 2025 guide](https://vention.com/blogs/industrial-automation-design/fanuc-robot-costs-guide-1004) listed the CRX-10iA at $46,026; the same part's page reads $51,433.73 today. CRX-5iA $43,207 to $47,528.64 (+10.0%), CRX-10iA $46,026 to $51,433.73 (+11.7%), CRX-10iA/L $50,723 to $56,931.82 (+12.2%), CRX-20iA/L $55,889 to $62,615.30 (+12.0%), CRX-30iA $60,868 to $67,988.64 (+11.7%). Fanuc builds the CRX in Japan, and Japanese machinery has entered the US at the 15% Annex III tier since [Proclamation 11021](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states), which [Tariffs, September 2026](https://open-factory.com/articles/tariff-schedule-2026/) walks through by HTS heading; the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) puts a 15% duty on a $46,026 customs value at $6,904, which is $1,500 more than the observed step. Vention has not said what moved its price and we have not asked it to. The point for a buyer is that a distributor's list moved 10 to 12% in eleven months while the BLS machine-tool PPI moved [1.4% over a similar period](https://open-factory.com/articles/haas-vf2-true-cost/). ## Three Cells, Line by Line The arm is the easy line. Everything below it in a quote is where a first-time buyer gets lost, so we priced three cells the way an integrator would, one line at a time, with a source or an (A) on every input in [data/assumptions.csv](data/assumptions.csv). The hardware lines come from Vention's published component ranges (vacuum grippers $1,500 to $5,500, custom grippers $2,500 to $10,500, pedestals $900 to $5,800, safety modules and area scanners $2,000 to $3,800, enclosures $6,000 to $13,000, all from the [October 2025 guide](https://vention.com/blogs/industrial-automation-design/fanuc-robot-costs-guide-1004)) and from [Cloos North America's welding-cell breakdown](https://www.cloosna.com/how-much-does-a-robotic-welding-cell-cost-in-2025/) (fixturing $5,000 to $40,000, training $3,000 to $10,000, shipping and installation $5,000 to $15,000, service contracts $5,000 to $25,000 a year). Integrator labor is $150 an hour, the bottom of the [$150 to $225 OEM distributor service rate](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/) we used for machine tools; a robotics technician's [BLS median wage of $35.53](https://www.onetonline.org/link/localwages/17-3024.01?st=OH) loaded at 1.43 is $50.80, so $150 is a 2.95x billing multiple, which is ordinary for an engineering shop. | Line | Palletizing (CRX-20iA/L) | MIG welding (UR10e) | Machine tending (CRX-10iA/L) | |---|---|---|---| | Arm and controller | $62,615 | $44,636 | $56,932 | | End of arm tooling | $4,500 vacuum gripper | $7,500 power source, feeder, torch (A) | $9,000 dual gripper | | Motion and mounting | $9,000 lift column (A), $5,800 pedestal | $5,800 cart | $5,800 cart, $6,000 CNC interface (A) | | Fixtures and presentation | $6,000 conveyor, pallet stands (A) | $8,000 fixture table, $6,000 fume extraction (A) | $8,000 part table (A) | | Vision | none | none | $8,000 2D pick camera (A) | | Safety | $7,600 two area scanners | $5,000 screens, light curtain (A) | $6,300 scanner, door light curtain (A) | | Integration engineering | 80 h, $12,000 | 60 h, $9,000 | 100 h, $15,000 | | Programming | 24 h, $3,600 | 40 h, $6,000 | 60 h, $9,000 | | Commissioning, risk assessment | 56 h, $8,400 | 56 h, $8,400 | 56 h, $8,400 | | Spares, training, freight | $3,500, $5,000, $2,500 | $3,000, $5,000, $5,000 | $4,000, $5,000, $3,000 | | Working cell | $130,515 | $113,336 | $144,432 | | Arm as share of cell | 48% | 39% | 39% | *Source: Open Factory [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), robot worked example. CSV: [data/cell-cost-stack.csv](data/cell-cost-stack.csv)* **The arm is 39 to 48% of the cell in all three cases, inside the 30 to 50% that [EVS's 2026 cost guide](https://www.evsrobot.com/industrial-robot-price-2026-cost-breakdown-roi-quote-guide.html) and every integrator's rule of thumb give, and the number that varies is the labor: 160 hours for a palletizer that ships with pattern software, 216 hours for a tending cell that has to talk to a CNC.** The welding cell is the cheapest of the three because it uses the cheapest arm and needs no vision, and it is also the one where our bottom-up number can be checked against published turnkey prices. Safety is the line that decides whether a cell is $6,300 or $20,000 before the arm moves. All three cells here are cobots running under ISO/TS 15066 speed-and-separation monitoring, so the safety line is two area scanners or a scanner and a light curtain, $5,000 to $7,600 from Vention's [$2,000 to $3,800 per device range](https://vention.com/blogs/industrial-automation-design/fanuc-robot-costs-guide-1004). An industrial arm, a Fanuc M-20iD/35 at $70,000 to $75,000, needs a fenced enclosure at $6,000 to $13,000 plus interlocked gates and a hard-guarded risk assessment, and Vention puts the fencing a CRX avoids at $10,000 to $30,000 per cell. The trade is speed: the M-20iD runs the same palletizing pattern faster than a CRX-20iA/L and needs the fence to do it, which is why Cloos's fenced welding cells start where our cobot cell ends. ## Where the Other $87,000 Goes Take the tending cell, the one a job shop with a [Haas VF-2](https://open-factory.com/articles/haas-vf2-true-cost/) is most likely to buy first. From a $56,932 arm: $25,000 of gripper, vision and part table, $11,800 of CNC interface and cart, $6,300 of safety, $15,000 of integration engineering, $9,000 of programming, $8,400 of commissioning and risk assessment, $12,000 of spares, training and freight, for $144,432. Labor at 216 hours is $32,400, or 22% of the cell, and it is the line that grows on the change order: every extra part family is another 8 to 16 hours of programming, and every CNC that lacks an automatic door and a robot I/O package adds the $6,000 interface kit we carry as an assumption. The welding cell cross-checks. Our bottom-up $113,336 sits between [Vectis Automation's published $95,000 to $140,000 all-in](https://vectisautomation.com/pricing/) and [Hirebotics' published $105,000 starting price](https://www.hirebotics.com/pricing) for a Cobot Welder that includes a UR8 Long, a Miller Deltaweld 350, a Tregaskiss torch, a 48 by 32 inch cart and its phone-based programming software, with a 15-month warranty on the UR arm and three years on the welder. Hirebotics is the only cell builder we found that prints its prices. Its page says so in its first line. Cloos's [pre-engineered fenced cells at $130,000 to $250,000](https://www.cloosna.com/how-much-does-a-robotic-welding-cell-cost-in-2025/), with an industrial arm, a positioner and a two-station table, are the next class up; custom systems run $250,000 to $600,000 and multi-robot lines $600,000 to $1.2 million. The full table, with our three cells alongside, is [data/published-cell-prices.csv](data/published-cell-prices.csv). | System | Application | Published price | Basis | |---|---|---|---| | Hirebotics Cobot Welder | MIG welding, cobot | $105,000 | published starting price, Sep 2026 | | Vectis Automation cobot welding tools | MIG welding, cobot | $95,000 to $140,000 | published all-in range; "Super DIY" from $75,000 | | Hirebotics Cobot Cutter | Plasma cutting, cobot | $100,000 | published starting price | | Hirebotics Cobot Painter | Painting, CRX-10iA/L Paint | $140,000 | published starting price | | Cloos NA pre-engineered welding cell | MIG welding, fenced industrial | $130,000 to $250,000 | published typical range, 2025 | | Cloos NA custom system | MIG welding, industrial | $250,000 to $600,000+ | published typical range, 2025 | | Cloos NA multi-robot system | MIG welding, multi-robot | $600,000 to $1.2M+ | published typical range, 2025 | | Open Factory bottom-up MIG cobot cell | MIG welding, cobot | $113,336 | this article | | Open Factory bottom-up palletizer | Palletizing, cobot | $130,515 | this article | | Open Factory bottom-up tending cell | Machine tending, cobot | $144,432 | this article | *Source: Hirebotics, Vectis Automation, Cloos North America; Open Factory arithmetic. CSV: [data/published-cell-prices.csv](data/published-cell-prices.csv)* The (A) lines are where the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) will replace our guesses with contributed quotes: the $7,500 power source and torch (Miller's and the welding retailers' pages blocked our reads), the $8,000 vision package, the $6,000 CNC interface kit and the $9,000 lift column. None of them is a large share of a cell; together they are $30,500 of the $388,283 across three cells, or 8%. ## Payback Against a Real Wage Every robot ROI page uses a round number for the operator. We used the BLS one. Welders, cutters, solderers and brazers (SOC 51-4121) had a [median wage of $25.84 an hour in May 2025](https://www.onetonline.org/link/localwages/51-4121.00?st=TX), hand packers (53-7064) [$17.44](https://www.onetonline.org/link/localwages/53-7064.00?st=TX), and CNC tool operators (51-9161) [$24.37](https://www.onetonline.org/link/localwages/51-9161.00?st=TX), all from the OES tables as republished by O*NET. The burden comes from the BLS Employer Costs release of 9 September 2026: for private industry, [wages and salaries are 70.0% of total compensation and benefits 30.0%](https://www.bls.gov/news.release/ecec.nr0.htm), so a fully loaded hour is 1.43 times the wage. A welder is $36.91 an hour loaded, a packer $24.91, a CNC operator $34.81, on 2,000 hours per shift-year. One shift: welding 2.26 years, palletizing 2.85, tending 3.03. Two shifts: 1.08, 1.37 and 1.45. Three shifts: 0.71, 0.90 and 0.95. The palletizer displaces one packer per shift. The welding and tending cells displace three quarters of an operator per shift because someone still loads fixtures and handles setups, an (A) in the CSV. Operating cost is small: 1 kW average draw at the [EIA industrial rate of 9.17 cents per kWh](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a) is $183 a shift-year, plus 3% of the cell a year for consumables and safety re-validation, or Cloos's $5,000 low-end service contract for the welding cell. **A one-shift shop buying a $144,432 tending cell against a $34.81 loaded CNC operator waits three years for payback; the same cell on two shifts pays back in 17 months, and the second shift costs the buyer nothing but a second program.** That is why the vendors' 18-month claims ([Hirebotics: "most systems pay for themselves within 18 months"](https://www.hirebotics.com/pricing); [Vectis: "payback in 2 years or less"](https://vectisautomation.com/pricing/)) are true for a two-shift fabricator and false for a one-shift one, and why the first question an integrator asks is how many shifts you run. Section 179 expensing of the full cell in year one, at the [2026 limit of $2,560,000](https://www.irs.gov/publications/p946), moves cash but not payback. The wage also moves by state, and a welding cell that pays back in 2.1 years in Wisconsin takes 2.6 in Tennessee. Wisconsin $58,410, California $58,060, Texas $53,340, the US $53,750, Ohio $50,340, Michigan $49,990, Tennessee $48,040. The 10th to 90th percentile band runs $39,240 to $77,530 nationally, and a fabricator paying the 75th percentile ($63,010, or $43.28 loaded) gets the welding cell back in 1.9 years on one shift. Per-state loaded hourly rates are in [data/welder-wages-by-state.csv](data/welder-wages-by-state.csv). The electricians' version of this map is in [Datacenter Construction Is Eating Your Electricians](https://open-factory.com/articles/electricians-wages-by-metro/). ## The Market Underneath the Price Two things about the supply side matter to a buyer this year. First, the arm makers are consolidating: ABB signed to sell its Robotics division, $2.3 billion of 2024 revenue at a 12.1% operational EBITA margin, to SoftBank for an [enterprise value of $5.375 billion](https://new.abb.com/news/detail/129685/abb-to-divest-robotics-division-to-softbank-group), closing mid-to-late 2026, which is 2.3 times sales for a business growing in the single digits. A buyer with ABB arms should ask in writing who honors the spare-parts price list after close. Second, cobot share is rising and cobot prices are falling relative to the market: 19.6% of units in 2025 at $33,417 average against $67,977 for everything else. Fanuc's CRX carries an [eight-year zero-maintenance interval](https://vention.com/blogs/industrial-automation-design/fanuc-robot-costs-guide-1004) on motors, reducers and cables, which is why our maintenance line is 3% of the cell rather than the 2 to 3.5% of machine price that a [VMC](https://open-factory.com/articles/haas-vf2-true-cost/) includes on top of a spindle rebuild. Who builds the cell matters as much as whose arm is in it. A3 has run a [certified robot integrator program since 2012](https://www.automate.org/robotics/robotic-integrator-certification-program/become-a-certified-robot-integrator) that audits project management and safety practice, and Fanuc, UR and Yaskawa each keep their own authorized integrator lists; a cell from a company outside all of them is a cell whose risk assessment nobody but the seller has read. The federal records below make the point in dollars: the most competed cell award in the set, 16 offers, came from a Fanuc integrator at roughly twice the arm price, and the least competed welding system, two offers from a welding distributor, came in at five and a half times. Service after the sale follows the same line. Cloos puts annual service contracts at [$5,000 to $25,000](https://www.cloosna.com/how-much-does-a-robotic-welding-cell-cost-in-2025/) and notes that a downtime shift can cost a fabricator more than $10,000; Hirebotics prints a 15-month warranty on the UR arm, 12 months on a Fanuc arm and three years on the welder; a cell whose quote has none of those numbers has left them for the change order. The humanoid question comes up in every automation meeting now, so, for comparison: BMW's Figure 02 pilot logged about 1,250 robot-hours in ten months, under one FTE, while 621,000 conventional arms went in worldwide, and [1,250 Hours](https://open-factory.com/articles/humanoid-pilot-as-buyer/) reads the pilot as a buyer would. For a 2026 purchase order the arm is a six-axis or a cobot, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) has no robot class because arm makers ship from stock or in weeks; Hirebotics quotes [about two weeks from purchase order](https://www.hirebotics.com/pricing) and Cloos 8 to 14 weeks for a pre-engineered cell against 16 to 28 for a custom one. The long pole is the integrator's calendar and your CNC's door, not the arm, which the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) handles as a labor line. ## What to Do Monday Price the arm yourself before the integrator does. Vention prints Fanuc CRX and UR unit prices; USAspending prints what the Army and NASA paid for the same SKUs; the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) worked example carries both with dates, and a quote that puts a CRX-10iA/L above $57,000 or a UR20 above $62,000 in September 2026 is carrying margin in the wrong line. Then make the integrator price the other 55 to 60% the way our table does: hours by task at a stated rate, hardware by part number, spares as a list with unit prices, per [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/). Ask how many shifts the payback assumes and make them show the wage; at one shift and the BLS median, none of the three cells beats 2.2 years, and a vendor page that says 18 months has assumed a second shift. Run the quote through [Quote Check](https://open-factory.com/tools/quote-check/) against the cell table above and send us the redacted result for the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/). Behind the wall: the integration multiple from 23 federal awards, which runs **2.0x the arm** for a Fanuc integrator's CRX-20 cell and **5.5x** for a welding distributor's, the 32 to 67% premium government resellers charged for the same UR arm a distributor sold, and the three clauses (acceptance cycle time, spare-arm pricing, programming ownership) that keep a first-time buyer from paying for the cell twice, all from the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). --- # The $60 Million Deposit: How Equipment Slots Became Datacenter Currency URL: https://open-factory.com/articles/slot-economy/ Date: 2026-05-08 Power / Deep dive GE Vernova 116 GW and 63 GW of paid reservations, Siemens 95 GW and Mitsubishi 35 GW, $25.8B of turbine deposits, Vertiv deferred revenue at $3.6B, Modine's $165M, deposit terms by equipment class, who holds the slots One GE Vernova 7HA.02 is [384 MW](https://www.gevernova.com/gas-power/products/gas-turbines/7ha). At the [~$790/kW that BNP Paribas estimates](https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-climbs-to-116-gw/826039/) GE Vernova is now getting for a heavy-duty turbine (July 2026), that one machine is ~$300 million of equipment, and the 20% reservation payment that holds a production slot for it is ~$60 million, wired about three years before the turbine ships. GE Vernova ended June 2026 with [116 GW of gas equipment under contract, 63 GW of it as slot reservation agreements](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm) that are not yet orders, and customers had paid the Power segment [$25.8 billion in advance](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000148/gev-20260630.htm). That cash is the currency this piece is about: what a slot costs, who has paid, where it shows up on the other OEMs' books, and what the [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) does with it. We cover the reservation mechanism as GE Vernova describes it in its filings, the three-OEM count of firm backlog versus reservations, the deposit pile in contract liabilities, Vertiv and Modine as the cooling and power-gear shadow of the same behavior, deposit terms by equipment class, what a reservation costs per kilowatt, the delivery years that are already gone, and the Monday actions for a buyer without a slot. Firm backlog went from 29 GW to 53 GW in five quarters; reservations went from 21 GW to 63 GW, so 54% of everything GE Vernova calls "under contract" is a deposit against a future order, not an order, and the reservation line has grown faster than backlog in every one of the five quarters. ## The Reservation Is Not an Order GE Vernova's 10-K spells out what a slot reservation agreement is by describing what can go wrong with one: "some counterparties to slot reservation agreements may not place orders equal to the value of their reservation amount or at all, and the volume of orders we expect under such agreements may fail to materialize" ([GE Vernova 10-K, January 2026](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000015/gev-20251231.htm)). The company says it makes "capacity expansion decisions and supply commitments based on demand forecasts, orders, slot reservation agreements, and deposits." In the cash flow statement the same instrument appears every quarter as "down payments on orders and slot reservation agreements at Power." A reservation is a dated claim on a production position, paid for with cash, that converts to a purchase order later. Backlog, in GE Vernova's usage, is only the converted part. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists both numbers per quarter because a buyer waiting on a turbine competes with both. **Between July 2025 and June 2026 GE Vernova signed 86 GW of gas equipment contracts and shipped 19 GW.** The quarterly releases give the flows: 9 GW signed in Q2 2025 (7 GW of reservations, 2 GW of orders), then [12 GW](https://www.sec.gov/Archives/edgar/data/1996810/000199681025000159/gevpressrelease3q25.htm), [24 GW](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000012/gevpressrelease4q25.htm), [21 GW](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000063/gevpressrelease1q26.htm) and [20 GW](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm). Conversions from reservation to order ran 3, 7, 8, 6 and 10 GW. Shipments ran 5, 4, 3, 4 and 3 GW. The company says it reached a 20 GW annual output rate in the third quarter of 2026, rising to 24 GW in 2028 and 30 GW in 2030. In Q2 2026 GE Vernova signed 20 GW, converted 10 GW of old reservations into orders and shipped 3 GW, so the book grew by 16 GW in one quarter, the same as the whole of what it shipped in the prior five. Pricing rides on the queue. In April 2026 CEO Scott Strazik said new bidding and winning activity was running ["10% to 20% growth in price"](https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-hits-100-gw-as-prices-rise/818332/) relative to the fourth-quarter 2025 backlog, and in December 2025 the investor update said [new reservations were pricing above current orders](https://www.utilitydive.com/news/ge-vernova-gas-turbine-investor/807662/). A reservation signed in early 2025 therefore holds a price the OEM will not offer again, which is the whole reason a slot has a secondary value at all. Our [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lists the turbine line at the disclosed price path, not at the price the holder locked. ## Three OEMs, 246 GW of Commitments Siemens Energy uses the same instrument and the same acronym. Its Q3 FY2026 analyst deck (August 5, 2026) lists Gas Services at ["69 GW backlog / 26 GW SRAs"](https://assets.siemens-energy.com/dam/3e846440-66dd-4a56-be75-b49d004e6741/2026-08-05_Q3_Analyst_presentation-pdf_Original%20file.pdf), with 15 GW booked in the quarter, 12 GW of it converted from slot reservation agreements, and 6 GW shipped. That is 95 GW of commitments, up from [80 GW at the end of December 2025 (22 GW for data centers)](https://www.powermag.com/full-throttle-five-trends-reshaping-the-gas-power-boom/) and [87 GW at the end of March 2026 (24 GW for data centers)](https://www.naturalgasintel.com/news/natural-gas-turbines-arent-gating-data-center-buildouts-ge-vernova-says/). Siemens booked [102 gas turbines in its first fiscal quarter](https://www.siemens-energy.com/global/en/home/press-releases/earnings-release-q1-fy-2026.html) alone, 40% of them for the US, and its CFO said in August that lead times run [three years or more](https://www.utilitydive.com/news/siemens-gas-turbine-backlog-nears-70-gw-as-company-expands-manufacturing/827390/). Mitsubishi Power reports differently. Its large-frame backlog was [35 GW in August 2026, up from 23 GW a year earlier](https://www.utilitydive.com/news/mitsubishi-gas-turbine-backlog-earnings/827761/), after selling 35 turbines (16 GW) in the fiscal year to March 2026. MHI does not split out reservations. Its CFO told analysts the [ten large-frame orders booked in the June quarter deliver between 2028 and 2030](https://www.mhi.com/finance/library/result/pdf/fy20261q/qa_20261q.pdf), that the company is "being selective in the projects we contract" with US utilities, and that customer cash arrives as "advance payments that will be used over time as we execute the project backlog." The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) tracks the three books side by side each quarter. Add them up and the three OEMs hold 157 GW of firm backlog and 89 GW of disclosed reservations, against combined heavy-duty output that GE Vernova puts at 20 GW a year for itself and Siemens at [15 to 16 GW deliverable in 2026](https://www.utilitydive.com/news/siemens-gas-turbine-backlog-nears-70-gw-as-company-expands-manufacturing/827390/). Wood Mackenzie counted [110 GW of global gas turbine orders in 2025 against 60 to 70 GW of manufacturing capacity](https://www.woodmac.com/press-releases/gas-turbine-prices-soar-195-as-market-faces-supply-demand-crisis/); EPRI counted [846 turbines and 100.3 GW ordered worldwide in 2025, 427 units and 43.1 GW of them in the US](https://www.utilitydive.com/news/5-year-waits-and-rising-costs-how-demand-is-redefining-the-gas-turbine-mar/813385/), more than double 2024. The [Recips vs Turbines vs Fuel Cells TCO](https://open-factory.com/articles/turbines-recips-fuel-cells/) piece prices the alternatives for a buyer who cannot get into any of these books. ## The Deposit Pile: $25.8 Billion at GE Vernova Power The cash side of the reservations sits in one balance-sheet line. GE Vernova's contract liabilities note splits "equipment and other service agreement liabilities" by segment. **At Power, that line went from $7.9 billion at December 2024 to $25.8 billion at June 2026, a 3.3x increase in eighteen months** ([10-K](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000015/gev-20251231.htm); [Q2 2026 10-Q](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000148/gev-20260630.htm)). The 10-K defines the line plainly: "Contract liabilities primarily represent cash received from customers under ordinary commercial payment terms in advance of delivery of equipment orders." Company-wide, contract liabilities and deferred income reached $40.0 billion, up $8.2 billion in calendar 2025 and another $14.1 billion in the first half of 2026. Power equipment cash grew $1.7 billion across the three quarters of 2025 through September, then $3.9 billion, $3.9 billion and $7.3 billion in the three quarters since, as the company added 21, 17 and 16 GW of commitments. The cash per new gigawatt roughly doubled between the two periods, which is what rising prices and larger conversion payments look like on a balance sheet. The ratio is the useful part. From March 2025 to June 2026 GE Vernova added 66 GW of gas commitments and $17.0 billion of Power equipment cash, ~$258 per kilowatt of new commitment (our arithmetic from the two filings; the Power line also carries nuclear, steam and hydro orders, so treat it as an upper bound). A cleaner cut: at March 2026 the Power segment held [$42.4 billion of backlog](https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-hits-100-gw-as-prices-rise/818332/) and 56 GW of reservations. Valuing the reservations at BNP's $790/kW gives ~$44 billion, so ~$86 billion of committed equipment against $18.5 billion of equipment cash: **customers have paid in ~21% of what they have committed to buy.** That is the number behind the $60 million in the title, and it is the default the [MEP Cost Table](https://open-factory.com/models/mep-cost-table/) uses for the turbine deposit line until an OEM discloses better. The same pattern is visible in Europe. Siemens Energy raised its fiscal 2026 free cash flow guidance from a €4 to €5 billion range to around €8 billion in May 2026 and attributed it to Gas Services and Grid Technologies, "which both experience [strong cash inflows driven by customer payments related to strong order momentum](https://www.siemens-energy.com/global/en/home/press-releases/earnings-release-q2-fy-2026.html)." Neither company discloses a deposit percentage; both disclose the pile. Siemens does disclose the conversion mechanics: of the 15 GW it booked in the quarter to June 2026, [12 GW came out of existing slot reservation agreements](https://assets.siemens-energy.com/dam/3e846440-66dd-4a56-be75-b49d004e6741/2026-08-05_Q3_Analyst_presentation-pdf_Original%20file.pdf), so four fifths of new firm backlog was already reserved and already partly paid for. A buyer reading a 69 GW backlog as the queue in front of them is short by the 26 GW of reservations that will convert ahead of any new order. ## Vertiv's $3.6 Billion Shadow The reservation habit did not stay with turbines. Vertiv's deferred revenue, the balance-sheet line where customer prepayments sit before shipment, went from [$1.06 billion at December 2024 to $1.81 billion at December 2025 to $3.63 billion at June 2026](https://data.sec.gov/api/xbrl/companyconcept/CIK0001674101/us-gaap/ContractWithCustomerLiabilityCurrent.json), per the 10-Q and 10-K balance sheets. In the same span Vertiv's quarterly sales went from $2.35 billion to $3.27 billion. **Customers have prepaid Vertiv 1.11 quarters of sales, up from 0.45 quarters at the end of 2024.** The Q2 2026 10-Q notes that [$604 million of the opening deferred balance](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050609/vrt-20260630.htm) was recognized as revenue in the quarter, so roughly one dollar in six of the pile turned into shipments while five stayed put. Deferred revenue sat between $500 million and $1.3 billion for eleven quarters, then added $683 million, $647 million and $1.17 billion in the last three. The timing matters because Vertiv's [Q4 2025 release](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm) was the last to publish backlog ($15 billion), organic orders (+252%) and book-to-bill (~2.9x); the [Q1 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026026379/q12026exhibit991vrt04222026.htm) and [Q2 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050323/q22026exhibit991vrt07292026.htm) releases have none of the three. Deferred revenue is now the only public series that tracks Vertiv's order book, which is why our [Vertiv backlog estimate](https://open-factory.com/articles/vertiv-backlog-estimate/) is built on it, and why the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) flags Vertiv as the one OEM in the electrical group that dropped a metric mid-cycle. The ratio of deferred revenue to that quarter's sales ran 0.30x to 0.54x from March 2023 to September 2025, then 0.63x, 0.93x and 1.11x. Modine put an actual number on a cooling reservation. Its FY2026 10-K discloses that in March 2026 it "entered into a long-term capacity agreement with one of its strategic data center customers" and "received a [$165.0 million up-front deposit](https://www.sec.gov/Archives/edgar/data/67347/000110465926066795/mod-20260331x10k.htm) from the customer to support the Company's investments necessary to meet the planned sales volume," against expected sales of "more than $4 billion of data center cooling products to this customer during calendar years 2027 through 2029." That is a 4.1% deposit for a three-year claim on a chiller factory, paid a year before the first delivery year. The 10-K also says the deposit is why operating cash flow rose despite working capital, which is the Vertiv and GE Vernova story at a smaller scale. Our [cooling M&A price list](https://open-factory.com/articles/cooling-ma-price-list/) covers what the majors paid to own the capacity being reserved. ## Deposit Terms by Equipment Class Nobody publishes a rate card, so we compiled what is disclosed. The table is the deposit line of the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/); each row names its source. | Equipment class | Instrument | Deposit or security | Refund terms as disclosed | Source (date) | |---|---|---|---|---| | Heavy-duty gas turbine, GE Vernova | Slot reservation agreement | ~20% implied: $18.5B Power equipment cash vs ~$86B committed (our estimate) | Reservation amount at risk if no order placed | [GEV 10-Q Mar 2026](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000064/gev-20260331.htm); [10-K Jan 2026](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000015/gev-20251231.htm) | | Heavy-duty gas turbine, Siemens Energy | Slot reservation agreement (26 GW open) | Not disclosed; "strong cash inflows driven by customer payments" | Not disclosed | [Q3 FY26 deck Aug 2026](https://assets.siemens-energy.com/dam/3e846440-66dd-4a56-be75-b49d004e6741/2026-08-05_Q3_Analyst_presentation-pdf_Original%20file.pdf) | | Large-frame turbine, Mitsubishi Power | Order with advance payments | Not disclosed; "advance payments that will be used over time" | Not disclosed | [MHI Q&A Aug 2026](https://www.mhi.com/finance/library/result/pdf/fy20261q/qa_20261q.pdf) | | Data center chillers, Modine | Long-term capacity agreement | $165M up-front on >$4B of 2027 to 2029 sales (4.1%) | Not stated | [Modine 10-K May 2026](https://www.sec.gov/Archives/edgar/data/67347/000110465926066795/mod-20260331x10k.htm) | | Thermal and power gear, Vertiv | Orders with customer deposits | Deferred revenue $3.63B, 111% of Q2 2026 sales | Not disclosed | [Vertiv 10-Q Jul 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050609/vrt-20260630.htm) | | Transformers, switchgear, gensets | Factory slot hold at land close | 10 to 30% of equipment cost | Not disclosed | [Global Data Center Hub Aug 2026](https://www.globaldatacenterhub.com/p/procurement-strategy-equipment-epc); [Terrapin Jun 2026](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) | | Recip engines, INNIO for Rehlko | Capacity reservation, then framework | 700 MW reserved, widened to 1.25 GW; terms not disclosed | Not disclosed | [Power Engineering Apr 2026](https://www.power-eng.com/gas/data-center-power-crunch-lifts-engines-aeroderivatives-into-larger-role/) | | Grid connection, ERCOT Batch Zero | Financial security | $50,000/MW | 20% refundable; converts to non-refundable interconnection fee | [ERCOT Apr 2026](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf) | | Grid connection, PUCT proposed 25.194 | Security plus fee | $50,000/MW at intermediate agreement, $50,000/MW non-refundable at final | 80% forfeited on withdrawal; full refund after 5 years at contracted peak | [Foley Mar 2026](https://www.foley.com/p/102mokd/public-utility-commission-of-texas-issues-proposed-rules-for-large-load-interconn/) | | SMR pressure vessels, X-energy and Doosan | Component reservation, 16 units | "Upfront financial commitment", amount not disclosed | Not disclosed | [POWER Dec 2025](https://www.powermag.com/x-energy-doosan-lock-in-16-unit-xe-100-component-reservation-as-doosan-commits-to-new-smr-factory/) | [Data: deposit-terms.csv](data/deposit-terms.csv) Ten instruments, three with a disclosed percentage (4.1%, 10 to 30%, ~20% implied), two with a disclosed refund rule, and the only party that tells you what happens to your money if you walk is the Texas grid operator. Two things stand out. First, the electrical room and the turbine hall run on the same convention: Terrapin's June 2026 note says "[most manufacturers will hold a production slot for 10-30% of equipment cost](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026)," and Global Data Center Hub describes the same 10 to 30% "at land close" with the order date "[preceding groundbreaking by six to twelve months](https://www.globaldatacenterhub.com/p/procurement-strategy-equipment-epc)." The transformer version of this queue, with 128 to 160 week lead times, is in [Transformer Lead Times Hit 160 Weeks](https://open-factory.com/articles/transformer-160-weeks/), and the [Transformer Sizing and Cost tool](https://open-factory.com/tools/transformer-sizing-cost/) returns the published lead time next to the price band. Second, the grid side has moved to the same structure with harsher terms. ERCOT's Batch Zero requires [$50,000/MW of financial security, 20% refundable](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf), and the PUCT's proposed rule adds a second [$50,000/MW non-refundable fee at the final agreement](https://www.foley.com/p/102mokd/public-utility-commission-of-texas-issues-proposed-rules-for-large-load-interconn/) with 80% of the security forfeited on withdrawal. For a 100 MW load that is $10 million posted before one transformer is ordered, which the [Large-Load Interconnection Cost tool](https://open-factory.com/tools/interconnection-cost/) and the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) include market by market. ## What a Reservation Costs per Kilowatt The turbine deposit only has meaning against a price, and turbine prices are quoted three ways. Wood Mackenzie's April 2026 release puts the average gas turbine at [$600/kW by the end of 2027, a 195% increase since 2019](https://www.woodmac.com/press-releases/gas-turbine-prices-soar-195-as-market-faces-supply-demand-crisis/), which backs out to ~$203/kW in 2019, and says the turbine is 20 to 30% of a combined-cycle project's cost. BNP Paribas's read on GE Vernova's second-quarter 2026 book, [as reported by Utility Dive](https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-climbs-to-116-gw/826039/), is ~$790/kW for heavy-duty, ~$950/kW for HA-class combined cycle and ~$1,800/kW for aeroderivatives. EPRI's plant-level number, the whole combined-cycle station, [rose from ~$2,000/kW to ~$3,000/kW in the six months to March 2026](https://www.utilitydive.com/news/5-year-waits-and-rising-costs-how-demand-is-redefining-the-gas-turbine-mar/813385/), and NRG's 1.2 GW Texas plant for a hyperscaler implies [~$2,670/kW](https://www.utilitydive.com/news/nrg-nears-12-gw-hyperscaler-deal-amid-texas-data-center-pause/827212/). William Blair's December 2025 figure of [$2,500/kW for GE Vernova turbine orders](https://www.utilitydive.com/news/ge-vernova-gas-turbine-orders-offshore-wind-earnings/810974/), as reported, sits 3x above BNP's; we read it as a plant-scope number and print the conflict rather than pick. The equipment range runs $600 to $1,800/kW depending on class and forecast date. The installed plant runs $2,000 to $3,000/kW, so a turbine reservation is a deposit on roughly a quarter to a third of the eventual plant. [Data: implied-deposit-per-kw.csv](data/implied-deposit-per-kw.csv) At 20% the reservation is $158/kW on a heavy-duty frame, $190/kW on an HA combined-cycle block and $360/kW on an aeroderivative, and GE Vernova's realized ~$258/kW of new customer cash per new GW lands between the 20% and 30% marks for its heavy-duty book. Scale that to a campus. A 1 GW behind-the-meter block built on HA machines is ~$950 million of turbines at BNP's estimate and ~$190 million of reservation cash, paid in 2026 for machines that ship in 2029 or 2030. A 100 MW factory bridge on aeroderivatives is ~$180 million of equipment and ~$36 million down. The [Recips vs Turbines vs Fuel Cells](https://open-factory.com/articles/turbines-recips-fuel-cells/) TCO shows why buyers at the 100 MW scale have been walking to reciprocating engines instead: NGI reported in May 2026 that [new large recip orders land in late 2027 at the earliest, most in 2028](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/), which is still a year ahead of any heavy-duty slot, and [Cummins and Caterpillar are sold out through 2028](https://open-factory.com/articles/gensets-sold-out-2028/) on the genset side. The [Genset TCO tool](https://open-factory.com/tools/genset-tco/) runs the ten-year comparison with your run hours, and the reservation cash sits in it as a year-zero outflow, not as a footnote. Stack the deposits and the pre-construction cash for a 100 MW Texas campus is a real number before one foundation is poured. The aeroderivative reservation is ~$36 million at 20% of BNP's $1,800/kW. The grid side adds [$5 million of ERCOT security at $50,000/MW](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf), of which only 20% is refundable, and the PUCT's proposed rule would add [another $5 million non-refundable at the final agreement](https://www.foley.com/p/102mokd/public-utility-commission-of-texas-issues-proposed-rules-for-large-load-interconn/). The transformer and switchgear slot holds run 10 to 30% of their equipment cost at land close. Before the EPC is even selected, the developer has committed something on the order of $50 million of cash that is either non-refundable or refundable only after five years of operation at contracted peak, and Wood Mackenzie's forecast says the turbine part of that bill rises another ~25% between now and the end of 2027. This is why the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) reports the security and study fees next to the equipment deposits: they are the same cash, spent in the same twelve months, and the [$50 MW in 15 markets](https://open-factory.com/articles/what-50mw-costs-15-markets/) comparison shows how much of it the market choice controls. ## Delivery Years: Sold Out Through 2028 The deposit buys a year, and the years are going. GE Vernova has "sold out heavy-duty turbine deliveries through 2028, with only about [10 GW of capacity remaining across 2029 and 2030](https://www.naturalgasintel.com/news/natural-gas-turbines-arent-gating-data-center-buildouts-ge-vernova-says/)" as of May 2026, down from 10 GW available in 2029 alone in January. In July the company said it is [taking reservations for 2031 and expects to be more than half contracted for that year by the end of 2026](https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-climbs-to-116-gw/826039/), while on 2032, "we need more time before we can articulate the timing of contracting." Jefferies, [as reported by Latitude Media](https://www.latitudemedia.com/news/is-the-gas-turbine-bottleneck-solving-itself/), puts large turbines over 90% booked for 2028 and 70% for 2029, against 65% and 33% for the smaller behind-the-meter systems. | Delivery year | GE Vernova | Siemens Energy | Mitsubishi Power | |---|---|---|---| | 2026 | 20 GW/yr output rate reached Q3 2026; sold out | 15 to 16 GW deliverable in 2026; 6 GW shipped Apr to Jun | 16 GW (35 units) sold in FY2025; capacity doubling vs 2024 | | 2027 | Not disclosed; lead time "about three years" | Large-frame capacity 35 units/yr plus 15 in 2027 | Not disclosed | | 2028 | Sold out; capacity 24 GW/yr | Medium turbines 80 to ~100 units/yr by 2028 | Q1 FY2026 orders deliver 2028 to 2030 | | 2029 | ~10 GW open across 2029 and 2030 (Apr 2026) | Lead times "three years or more" (Aug 2026) | Q1 FY2026 orders deliver 2028 to 2030 | | 2030 | Capacity 30 GW/yr planned; expected sold out by end-2026 | Discussions for projects in early 2030s | Some projects 2030 to 2031 | | 2031 | Taking reservations; more than half contracted by year-end 2026 (target) | Not disclosed | Not disclosed | | 2032+ | "Need more time" before contracting (Jul 2026) | Not disclosed | Not disclosed | [Data: delivery-year-disclosures.csv](data/delivery-year-disclosures.csv) Across the three OEMs, the earliest delivery year with disclosed open capacity is 2029, the OEM quoting the longest lead time is EPRI's composite at [more than five years for a large turbine](https://www.utilitydive.com/news/5-year-waits-and-rising-costs-how-demand-is-redefining-the-gas-turbine-mar/813385/), and the only delivery-year detail Mitsubishi gives is the 2028 to 2030 window for orders booked in mid-2026. The small end of the market has its own queue, and it is the one most factory buyers actually stand in. EPRI's count for 2025 puts [70% of all turbines sold under 100 MW](https://www.utilitydive.com/news/5-year-waits-and-rising-costs-how-demand-is-redefining-the-gas-turbine-mar/813385/), with the fastest growth in the 3 to 20 MW range, while H- and J-class machines were 14% of units and more than half of megawatts. Lead times for the small frames run 18 to 36 months against more than five years for a large one, and EPRI's explanation is the forging chokepoint: "[there's a very limited number of places in the world where you can have extremely large components forged](https://www.utilitydive.com/news/gas-turbine-supply-crunch-set-to-raise-prices-195-by-2027-woodmac/816904/)," which bites the 400 MW rotor and not the 20 MW one. Siemens Energy is adding 20 units of medium-turbine capacity on top of 80 to reach about [100 units a year by 2028](https://www.utilitydive.com/news/siemens-gas-turbine-backlog-nears-70-gw-as-company-expands-manufacturing/827390/), against 35 large-frame units plus 15 more in 2027. A 20 to 40 MW plant that reserves a medium frame in 2026 is a 2028 plant. The same plant waiting on a heavy-duty block shared with a datacenter campus is a 2029 or 2030 plant. The deposit convention is the same at both ends, so the buyer's decision is which queue to pay into, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists the two classes as separate rows for that reason. Strazik's own framing is that the turbine is often not the gating item: "[when you're talking about a three-year cycle from when a project starts](https://www.naturalgasintel.com/news/natural-gas-turbines-arent-gating-data-center-buildouts-ge-vernova-says/), the EPC build out, the permitting, the fuel availability," a 2029 turbine slot fits a plant that starts today. That is true for a utility with a site, a permit and a gas lateral. It is not true for the developer who has a reserved 2029 slot, a site under option and no air permit, and Siemens Energy's CEO said the observable trend in 2026 was projects "[being shuffled between sites based on which could secure air permits faster](https://www.naturalgasintel.com/news/natural-gas-turbines-arent-gating-data-center-buildouts-ge-vernova-says/)." The slot is portable across sites in a way the permit is not, which is the property that makes it currency. Every behind-the-meter megawatt also needs a generator step-up transformer at [144 weeks](https://open-factory.com/articles/behind-meter-90gw/), so the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) puts the GSU, not the turbine, on the critical path for most 2028 energizations. ## What to Do Monday For a buyer with a 50 to 500 MW load and no slot, the disclosed facts point to five actions. Price the deposit as a line item, not a contingency. At ~20% of turbine cost, or 10 to 30% of transformer and switchgear cost at land close, the reservation cash for a 100 MW campus runs into the tens of millions and lands 24 to 36 months before the equipment. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) now lists a deposit-timing column per component so the carrying cost of that cash sits next to the carrying cost of the delay it prevents. Ask the OEM for the reservation amount, the conversion deadline and the forfeiture rule in writing before wiring anything. GE Vernova's 10-K refers to a "reservation amount" and to counterparties not ordering "equal to the value" of it, which means the agreement has a face value and a conversion expectation. None of the three OEMs publishes the refund terms; ERCOT and the PUCT do, and their terms (20% refundable, 80% forfeited) are the worst case to plan against. The [Liquidated Damages](https://open-factory.com/articles/liquidated-damages/) piece covers the delivery-side clauses that should sit opposite your deposit. Buy the slot, not the machine. A reservation written against a frame class and a delivery quarter, rather than a serial number and a site, is the version that can move to another project or another owner. Chevron and Engine No. 1 reserved seven 7HA turbines in January 2025 with no named site, and by June 2026 had assigned 2.67 GW of that capacity to a [20-year contract with Microsoft in West Texas](https://www.power-eng.com/gas/new-co-located-gas-plant-to-power-microsoft-data-center-in-west-texas/) with first power in 2028. The instrument came before the customer. Put the GSU and the switchgear on the same reservation calendar as the turbine. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) publishes 144 weeks for a GSU and 60 to 80 weeks for 15 kV metal-clad switchgear (Terrapin, June 2026), so a 2029 turbine slot with a 2027 GSU order is a 2029 plant, and a 2029 turbine slot with a 2028 GSU order is a 2030 plant. The [Quote Check tool](https://open-factory.com/tools/quote-check/) returns the published lead time for the class you are quoting. Decide by year-end 2026 whether you are a 2029 buyer or a 2031 buyer. GE Vernova expects 2030 sold out by the end of this year and 2031 half contracted; Wood Mackenzie expects [orders to peak in 2026](https://www.woodmac.com/press-releases/gas-turbine-prices-soar-195-as-market-faces-supply-demand-crisis/). A buyer who cannot commit a deposit in 2026 should plan on recips or fuel cells for 2028 and a grid connection priced off the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) for 2030, and stop carrying a turbine in the base case. Behind the paywall: the twelve named holders of disclosed turbine reservations, allocations and framework agreements, led by **American Electric Power's ~13 GW secured plus options on 10 GW more**, what the record shows and does not show about the resale value of a 2028 slot, and the six contract terms that make a reservation transferable, all of it in the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # Who Gets the IEEPA Refund URL: https://open-factory.com/articles/ieepa-refund-fight/ Date: 2026-04-29 Procurement / Take CIT refund order March 4, importers of record, tariff surcharges passed to buyers, the contract clause nobody wrote, Skadden's allocation language, bonded warehouses as the 2026 hedge Ford told analysts this morning that it booked a $1.3 billion one-time benefit for IEEPA tariffs it paid between March 2025 and February 2026 ([Ford Q1 2026 call, April 29, 2026](https://www.fool.com/earnings/call-transcripts/2026/04/29/ford-f-q1-2026-earnings-call-transcript/)). A plant that bought a German switchgear lineup last November and paid a 15% "tariff surcharge" line will book nothing, because the refund goes to whoever signed the entry summary, and on most equipment POs that is the vendor's broker. **CBP's own declaration to the Court of International Trade puts the pool at about $166 billion across 53 million entries and 330,000 importers, and the regulation that pays it out, 19 CFR 24.36, does not have a line for the customer who funded the duty** ([Skadden, March 24, 2026](https://www.skadden.com/insights/publications/2026/04/insights-april-2026/tariff-refund-mechanism-takes-shape); [19 CFR 24.36](https://www.law.cornell.edu/cfr/text/19/24.36)). The [Open Factory Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) flags every IEEPA chapter 99 line on a 2025 entry as refundable to the importer of record, with the CAPE cutoff dates. This piece covers the pool, the payment chain, the earnings calls, a $2M switchgear order line by line, the liquidation clocks and, behind the wall, the allocation clause, the protest deadlines and the bonded warehouse and FTZ math for a 2026 to 2027 order. ## What CBP Collected and What It Owes Gross customs duties ran $6.2 billion to $7.9 billion a month through fiscal 2024, crossed $16 billion in April 2025, peaked at $33.1 billion in October 2025 and were still $24.0 billion in March 2026, the first full month after the ruling ([Treasury MTS Table 4](https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v1/accounting/mts/mts_table_4?filter=classification_desc:eq:Customs%20Duties&sort=-record_date); CSV in [data/customs-duties-monthly.csv](data/customs-duties-monthly.csv)). March has not fallen back because the 10% Section 122 surcharge took over on February 24 and Section 232 never stopped. The stack by HTS line lives in [Tariffs: The Actual Schedule](https://open-factory.com/articles/tariff-schedule-2026/) and is not repeated here. The refundable slice is the IEEPA share. CBP's Executive Director for Trade Programs told Judge Eaton on March 4 that 71,647,732 entries were filed from February 2025 to February 2026 against 39,134,522 the year before, that CBP was still liquidating with IEEPA duties, and that it had issued no refunds ([Declaration of Brandon Lord, Atmus Filtration v. United States, ECF 19](https://storage.courtlistener.com/recap/gov.uscourts.cit.19346/gov.uscourts.cit.19346.19.0_4.pdf)). The same day the court ordered CBP "to liquidate those entries without regard to the IEEPA duties" and to reliquidate "any liquidated entries for which liquidation is not final," for every importer, not only the 2,500 who had sued ([Order, ECF 21, March 4, 2026](https://storage.courtlistener.com/recap/gov.uscourts.cit.19346/gov.uscourts.cit.19346.21.0_6.pdf)). CAPE, the refund module in ACE, went live on April 20 ([CSMS 68396594](https://content.govdelivery.com/bulletins/gd/USDHSCBP-413a632)). ## The Buyer Is Not in the Payment Chain **CBP pays the importer of record or the party it names on Form 4811, and a plant that bought DDP was neither.** The regulation says a refund "shall be prepared in the name of the person to whom the refund is due," and CBP's April 13 message narrows the filer to "the IOR associated with the entry summaries or ... the broker that filed the entry summaries" ([CSMS 68340863](https://content.govdelivery.com/bulletins/gd/USDHSCBP-412cc7f)). Six ways a plant bought imported gear in 2025, and who holds the money in each. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) records the Incoterm on every award for this reason. [CSV: data/refund-entitlement.csv](data/refund-entitlement.csv) Six structures, and in four the refund lands with someone other than the party that funded the duty: the OEM on a DDP quote, the distributor on a stocked panel, the carrier on a consignee's shipment, the OEM again on US stock sold at a raised list price. Only the buyer who entered through its own broker files its own CAPE Declaration. Skadden's February 20 note said it plainly: "the applicable regulations (19 C.F.R. § 24.36) provide that refunds are paid to the importer of record," and "more likely, however, contracts will allocate responsibility for tariffs without clarifying which party is entitled to the benefit of any refunds or reductions" ([Skadden, February 20, 2026](https://www.skadden.com/insights/publications/2026/02/the-supreme-court-ends-ieepa-tariffs)). The firm's advice, which is the allocation language in our subtitle, was to "proactively discuss responsibility for seeking refunds and the ultimate allocation of those refunds with their counterparties, before a dispute arises." The carriers moved first. FedEx Corporation and FedEx Logistics sued at the CIT on February 23, three days after the ruling, pleading that they "transport imports on behalf of their customers" and "are responsible for paying these tariffs," and asking for "a full refund from Defendants of all IEEPA duties Plaintiffs have paid" ([FedEx v. United States, CIT 26-01150, complaint](https://storage.courtlistener.com/recap/gov.uscourts.cit.19237/gov.uscourts.cit.19237.2.0_5.pdf)). Costco filed in November ([CIT 25-00316](https://www.courtlistener.com/docket/71973233/costco-wholesale-corporation-v-united-states/)). In January the CIT revised its filing forms to make plaintiffs disclose third-party financing, anticipating suits where "the ultimate economic benefit of the remedy would flow to a party other than the plaintiff." The switchgear buyer is that other party, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) now records whether an OEM has said publicly what it will do with refunds on surcharged orders. ## What the Earnings Calls Admit **The OEMs that surcharged you have told investors the refund is theirs to recognize, and none of the four we read has said it will pass it back.** Ford's CFO: "in Q1, we recognized a $1.3 billion benefit related to [IEEPA] tariffs ... related to [IEEPA] tariffs paid between March 2025 and February 2026," booked $700 million to Ford Blue and $500 million to Ford Pro ([Ford, April 29, 2026](https://www.fool.com/earnings/call-transcripts/2026/04/29/ford-f-q1-2026-earnings-call-transcript/)). Generac, an hour earlier, said its raised margin guidance "excludes the future impact of any potential tariff recovery as a result of the recent Supreme Court ruling," and its CFO added that the company assumes the removal of IEEPA duties "will get fully offset by a new tariff framework made up of incremental section 122, 232 and 301 tariffs" ([Generac, April 29, 2026](https://www.fool.com/earnings/call-transcripts/2026/04/29/generac-gnrc-q1-2026-earnings-call-transcript/)). Caterpillar put its net incremental tariff cost at $1.7 billion for 2025 and guided $2.6 billion for 2026 before the ruling ([Caterpillar, January 29, 2026](https://www.fool.com/earnings/call-transcripts/2026/01/29/caterpillar-cat-q4-2025-earnings-call-transcript/)); Eaton's CEO said on February 3 that "tariffs in 2025 was not part of the plan. We mastered that pretty well," the polite form of priced through ([Eaton, February 3, 2026](https://www.fool.com/earnings/call-transcripts/2026/02/03/eaton-etn-q4-2025-earnings-call-transcript/)). The buyer's side of the same transactions is a surcharge line, and surcharges were rarely the duty. Japanese machine-tool builders added a flat 10% surcharge to US quotes in April 2025, before any entry had been made at the new rate ([Seisanzai Japan, April 2025](https://seisanzai-japan.com/article/p5371/)); [Rockwell's 250 Basis Points](https://open-factory.com/articles/rockwell-250bps/) walked through how that pass-through lands in a PLC quote in December, and [Copper at $6.77](https://open-factory.com/articles/copper-tariff-panel/) did it for a 200-amp panel in February. A surcharge is a price term; a duty is a deposit with CBP; only the deposit comes back, and to the depositor. [Quote Check](https://open-factory.com/tools/quote-check/) separates the two on any 2025 quote, and [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/) shows where the line hides. ## A $2M Switchgear Order, Line by Line **On a $2 million German MV lineup entered in November 2025, the refund pot is $308,856 and the PO most buyers signed gives them $0 of it.** The arithmetic: HTS 8537.20.00 carries a 2.7% column 1 rate, and the EU order of July 31, 2025 set the IEEPA rate so that column 1 plus IEEPA equals 15%, so the IEEPA duty was 12.3% of customs value ([EO 14326, 90 FR 37963, sec. 2(c)](https://www.federalregister.gov/documents/2025/08/06/2025-15010/further-modifying-the-reciprocal-tariff-rates); [HTS 8537.20.00](https://hts.usitc.gov/reststop/search?keyword=8537.20.00)). With $1.8 million of the $2 million PO as entered value (freight and commissioning invoiced separately), the vendor's broker deposited $221,400 of IEEPA duty. The vendor billed a flat 15% surcharge on the PO, $300,000. CBP will refund the $221,400 with interest under 19 CFR 24.36 at the corporate overpayment rate, 6% for the first quarter of 2026 ([FR Doc 2026-01175](https://www.federalregister.gov/documents/2026/01/22/2026-01175/quarterly-irs-interest-rates-used-in-calculating-interest-on-overdue-accounts-and-refunds-of-customs)), which over eight months from deposit to liquidation is $8,856. Full computation in [data/switchgear-2m-example.csv](data/switchgear-2m-example.csv). Four contracts, four splits: with no refund clause the vendor keeps $308,856, the $78,600 surcharge excess plus the $230,256 CBP sends back; with a surcharge equal to the duty and no clause, $230,256; with a FAR 52.229-3 style clause the price drops by the $221,400 "after-relieved" duty and the vendor keeps $87,456; with the clause behind the wall the buyer recovers all $308,856. The federal clause is the one public model: FAR 52.229-3(d) says "the contract price shall be decreased by the amount of any after-relieved Federal tax," a term that covers a duty the contractor "obtains a refund or drawback" of through "judicial ... action taking effect after the contract date" ([FAR 52.229-3](https://www.acquisition.gov/far/52.229-3)). The private forms have nothing like it: AIA A201-2017 §3.6 makes the contractor pay taxes "legally enacted" when bids are received and is silent on refunds, and ConsensusDocs 200 allocates the tax, not the recovery ([AIA A201](https://www.aiacontracts.com/contract-documents/25131-general-conditions-of-the-contract-for-construction); [ConsensusDocs 200](https://www.consensusdocs.org/contract/200/)). A PO on either form, or on the vendor's paper, has no refund provision unless someone typed one in. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) records which form each public award used. ## The Clocks Started at Liquidation, Not at the Ruling **Judge Eaton's order reaches only entries that are "not final," and finality is counted from liquidation, which for a March 2025 entry has already happened.** Entries liquidate about 314 days after entry and are deemed liquidated at one year ([19 USC 1504](https://www.law.cornell.edu/uscode/text/19/1504); [Skadden, March 24, 2026](https://www.skadden.com/insights/publications/2026/04/insights-april-2026/tariff-refund-mechanism-takes-shape)). CBP may reliquidate on its own within 90 days of liquidation ([19 USC 1501](https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title19-section1501&num=0&edition=prelim)); the importer may protest within 180 days ([19 USC 1514](https://www.law.cornell.edu/uscode/text/19/1514)); and CAPE Phase 1 accepts a liquidated entry only up to 80 days after liquidation, then sets unliquidated entries to liquidate 45 days after the declaration is accepted, with the money "generally issued within 60 - 90 days" ([CSMS 68340863](https://content.govdelivery.com/bulletins/gd/USDHSCBP-412cc7f)). The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) includes those three marks on every 2025 PO line. [CSV: data/litigation-timeline.csv](data/litigation-timeline.csv) Read the table against an order book: a lineup entered in March 2025 liquidated around January 2026, left the CAPE window in early April and can be protested only until about July 2026; a November 2025 entry stays in CAPE to December. Two more dates sit on top. The government's deadline to appeal the March 4 order runs to early May, and an appeal "could lead to a stay of Judge Eaton's order" ([Skadden](https://www.skadden.com/insights/publications/2026/04/insights-april-2026/tariff-refund-mechanism-takes-shape)). And the 10% Section 122 surcharge that replaced IEEPA on February 24 is statutorily limited to 150 days, so it lapses at 12:01 a.m. on July 24, 2026 unless Congress extends it, with Section 232 goods and USMCA goods already exempt ([Proclamation 11012, 91 FR 9339](https://www.federalregister.gov/documents/2026/02/25/2026-03824/imposing-a-temporary-import-surcharge-to-address-fundamental-international-payments-problems)). USTR opened 60 Section 301 investigations on March 12 into economies without a forced-labor import ban, which is the candidate replacement ([91 FR 12884](https://www.federalregister.gov/documents/2026/03/17/2026-05151/initiation-of-section-301-investigations-of-acts-policies-and-practices-of-various-economies-related-to-the)); the [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) lists the docket dates for the machinery lines. ## What to Do Monday **Send every vendor that billed you a tariff surcharge in 2025 a letter asking for three things: the entry summary numbers, confirmation that it has filed or will file a CAPE Declaration, and its position on passing the refund through.** The letter costs nothing and starts the record for the "quasi-contractual claims" over unjust enrichment that follow if the vendor keeps the money. Ask for the Incoterm and importer of record on every open PO the same day. If you were the importer, confirm your ACE Portal account has refund bank details on file, because CBP will not transmit without them ([CSMS 68315804](https://content.govdelivery.com/bulletins/gd/USDHSCBP-4126a9c)). Run each 2025 entry through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) to split the IEEPA line from the Section 232 and 301 duties that stay, and write the HTS line and origin into every new RFQ as [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/) lays out. Second, do not sign a 2026 PO with the 2025 language: vendors are quoting the Section 122 surcharge into prices today, and if it ends early that is a second refund with the same problem. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) has MV switchgear at 44 to 54 weeks and a 2,500 kVA pad-mount at 40 to 65, so everything ordered this quarter enters after July 24 under whatever replaces the surcharge. The entry date sets the duty, which makes the [slot deposits](https://open-factory.com/articles/slot-economy/) buyers are paying tariff bets too. Behind the paywall: the refund allocation clause we would not sign a capital-equipment PO without, in full, the protest and reconciliation steps with the day counts for a March 2025 and a November 2025 entry, and the bonded warehouse and FTZ arithmetic for a **$180,000** Section 122 bill on a June 2026 delivery, mapped to the [Tariff Exposure Calculator's](https://open-factory.com/tools/tariff-exposure-calculator/) citation fields. --- # Behind the Meter, In Front of the Queue: 90 GW Announced, 2 GW Running URL: https://open-factory.com/articles/behind-meter-90gw/ Date: 2026-04-28 Power / Forecast xAI Colossus at 1,498 MW of turbines, the GSU transformer nobody prices, 143-week GSU lead times, DOE 202(c) orders, what a developer can actually energize by 2028 Datacenter developers have announced [about 90 GW of behind-the-meter generation across 59 US projects, and about 2 GW of it runs](https://cleanview.co/reports/behind-the-meter-data-centers), which is 2.2%. Another 1.2% is under construction, 36% holds an air permit, and 60% exists as a press release. The same month EIA published its July 2026 planned-generator inventory: every gas turbine, combined-cycle block and engine plant any owner in the country has told EIA it intends to build comes to [68.1 GW, and 11.0 GW of that is under construction](https://www.eia.gov/electricity/data/eia860m/). The announced private fleet is bigger than the visible national one. This piece rebuilds the 90 GW from permits, filings and EIA's unit list, prices the two gates that decide who energizes, and forecasts what runs by the end of 2028 using the [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). We cover the source of the 90 GW and what EIA sees of it, the nineteen campuses we could trace to a document, xAI's 1,498 MW as the benchmark, the GSU on every frame-turbine critical path, the permits that stopped two Stargate sites, DOE 202(c) and FERC, and our forecast through 2028, with winners and payers behind the wall. For 2028, EIA lists 2.9 GW of gas under construction, 6.9 GW approved or awaiting permits and 6.6 GW planned with no approvals, while the named behind-the-meter campuses we could document for that year and which are absent from EIA's list add 10.1 GW on top; for 2027 the split is 3.7, 1.3, 6.4 and 8.3 GW. ## Where 90 GW Comes From The number is Cleanview's. Its May 2026 report, built from air permits, SEC filings, utility dockets and satellite imagery, identified [59 datacenters with about 90 GW of planned on-site generation, 92% of it announced since the start of 2025](https://cleanview.co/reports/behind-the-meter-data-centers). Its February 2026 edition had counted [46 projects and 56 GW, with equipment deals identified at about two thirds of them and 75% of the identified equipment (23 GW) gas-fired](https://newsletter.cleanview.co/p/bypassing-the-grid-how-data-centers). The full dataset is a $4,900 report. We cite the free summary and newsletter posts. Global Energy Monitor arrives at the same order of magnitude from a different direction: the US had [252 GW of gas capacity in development at the start of 2026, more than a third of it slated to power datacenters on site, with Texas alone at 80.6 GW of which 40 GW is for datacenters](https://globalenergymonitor.org/research/betting-big-data-centers-us-now-leads-world-new-gas-power-development). Two trackers, two methods, 84 to 90 GW. Cleanview's own split of the 90 GW is 54 GW announced or early-stage, 32.4 GW permitted, 1.1 GW under construction and 2.0 GW operating, and it expects [2.8 to 3.2 GW operating by the end of 2026](https://cleanview.co/reports/behind-the-meter-data-centers). **EIA can see 16 GW of the 90.** We went through all 487 planned natural-gas generator rows in the July 2026 EIA-860M workbook (turbines, combined-cycle parts and engines, 68.1 GW) and matched owners and plant names against the datacenter campuses in the press and the permit record. Two campuses are there: Fermi America's Project Matador in Carson County, Texas, at [11,679 MW across 137 generator rows, 420 MW of it with approvals received and the rest at status P, planned with approvals not initiated](https://www.eia.gov/electricity/data/eia860m/xls/july_generator2026.xlsx), and Homer City in Pennsylvania at 4,396 MW across 14 rows, all status P, with online dates from July 2028 to November 2029. Nothing else in the file is a datacenter campus by name. Colossus, Stargate Abilene, Stargate Frontier, Project Kilby, GW Ranch, Monarch, the Meta plants in New Albany: none has a row. EIA-860M is a survey of generators [1 MW and larger](https://www.eia.gov/electricity/data/eia860m/) and its planned sheet depends on the owner filing; mobile turbines that arrive on trucks, engine packs owned by a power-as-a-service provider and plants that have not passed a board's final investment decision tend not to be filed. The gap is the measurement problem of this whole market, and it is why the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes a behind-the-meter pipeline table from permits rather than waiting for EIA. The mix in EIA's file also says something about who is actually building. Of the 68.1 GW, MISO holds 20.6 GW and ERCOT 16.7 GW, and the largest single owner after Fermi is Entergy Louisiana at 9.5 GW. Entergy's Q2 2026 10-Q describes [seven new combined-cycle units totaling 5,278 MW at about $12.9 billion, four of them near a Meta site in Richland Parish, with in-service dates in 2030 and 2031](https://www.sec.gov/Archives/edgar/data/65984/000006598426000283/etr-20260630.htm), on top of the earlier Franklin Farms and Waterford 5 units for the first Meta campus. That is dedicated datacenter generation built by a utility and rate-based, the front-of-meter cousin of everything in this piece, and it lands two to three years after the private plants promise to. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) treats utility-built dedicated generation as its own category for exactly that reason. ## The Nineteen We Could Document We built the pipeline from documents, not announcements: an air permit, an SEC filing, a utility 10-Q, a state agency letter or a company release with equipment counts. Nineteen campuses met the bar. They add to 45 GW, half of Cleanview's 90, and the other half is either in Cleanview's paid rows or in announcements we could not tie to a filing. The full table with URLs is [data/btm-pipeline.csv](data/btm-pipeline.csv). Of the nineteen, three are operating (Colossus 1 and 2, Vantage's Virginia site, and the southern half of Williams Socrates), two more are under construction for 2026 (Socrates North and the EdgeConneX New Albany plant), and the remaining thirteen are permits, orders and PPAs targeting 2027 to 2029, with 7.65 GW of the total in a single Amazon permit. The operating benchmark is xAI. Cleanview puts [Colossus 1 and 2 outside Memphis at 1,498 MW of operating gas-turbine capacity, with nearly 60 turbines visible on satellite and foundations laid for more](https://newsletter.cleanview.co/p/data-center-developers-want-to-build). The Southaven, Mississippi half runs [27 portable turbines totaling 495 MW under the Mississippi Department of Environmental Quality's position that portable units are mobile sources exempt from a stationary air permit; the NAACP sued under the Clean Air Act in April 2026 and the Department of Justice moved to intervene on xAI's side in June](https://www.utilitydive.com/news/doj-intervenes-xai-data-center-gas-turbine-lawsuit/823267/). On the Tennessee side the Shelby County Health Department [approved a permit for 15 turbines on July 2, 2025, which the NAACP and the Southern Environmental Law Center appealed](https://tennesseelookout.com/2025/07/17/naacp-others-appeal-xai-turbine-permits-for-memphis-data-center/). NGI reports that VoltaGrid's engine packs and grid supply took [Colossus from start to finish in 122 days](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/). That is the speed everyone else is selling against: 1.5 GW in under two years, built from leased and mobile units, with the permits litigated afterwards. The largest documented projects are frame-turbine plants that have not broken ground on generation. Chevron's Project Kilby will supply Microsoft's Pecos, Texas campus with [about 2.67 GW under a 20-year PPA signed June 22, 2026, a majority from large GE Vernova turbines plus Solar Turbines units, with FID expected by the end of 2026 and first power in 2028](https://www.chevron.com/newsroom/2026/q2/chevron-signs-20-year-power-agreement-with-microsoft-for-west-texas-data-center); Chevron's New Energies president told analysts the machines are [seven GE Vernova 7HA.02s and twelve Caterpillar Titan 350s, that the PPA is take-or-pay, and that "only one" announced behind-the-meter project "is at multi-gigawatt scale" with a customer commitment](https://www.naturalgasintel.com/news/chevron-touts-natural-gas-fired-kilby-project-as-ai-power-demand-outruns-grid/). Thirty miles east, Amazon confirmed it acquired the GW Ranch site in Pecos County, where Pacifico Energy's [TCEQ prevention-of-significant-deterioration permit, issued January 21, 2026, covers 35 turbines at 7.65 GW nameplate and about 5 GW nominal delivered output](https://newsletter.cleanview.co/p/scoop-amazon-is-behind-one-of-the), and where land clearing began in August 2026. Amazon also confirmed talks with Homer City's developers the same week. Two Pecos County campuses and one Pennsylvania campus hold 14.7 GW of frame-turbine plans and 42 large turbines that must come out of the queues we measured in [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/). Fermi America is the only one of the giants that has filed with EIA, and its own filings are the best public record of what a behind-the-meter plant actually requires. The 10-K lists [seven leased GE TM2500 mobile units rated about 132 MW, six Siemens SGT-800s, three Siemens SGT6-5000F heavy-duty turbines at about 240 MW each in simple cycle, three GE 6B frame units, "and certain substations, switchgear, transformers, and related balance-of-plant components"](https://www.sec.gov/Archives/edgar/data/2071778/000207177826000010/frmi-20251231.htm), with TCEQ's approval of a roughly 6 GW air permit on February 25, 2026 and an application for 5 GW more filed March 27. By August 2026 the company had [1.5 GW of turbines landed, a 222 MW lease with TensorWave worth about $6.5 billion over 15 years, 86 MW of grid supply from Xcel's SPS expected in the second half of 2026 and 114 MW more contracted by October 1, 2027, and a target of "about 200 megawatts of initial commercial power over the next six months, and about 1.5 GW over the next 18 to 24 months"](https://www.sec.gov/Archives/edgar/data/2071778/000207177826000046/fermiq22026erex_991.htm). Hillcore will build, own and operate [about 2.6 GW more under a ground sublease, with first power of about 350 MW targeted within 24 months of notice to proceed](https://www.sec.gov/Archives/edgar/data/2071778/000207177826000051/frmi-20260630.htm). Cleanview's satellite series shows the site [at least a year behind the schedule in the IPO filing, with no vertical construction between February and April 2026](https://cleanview.co/reports/fermi). One detail for anyone reading EIA's file: 860M codes Matador to ERCOT, while Fermi's 10-Q says the campus "resides in the Southwest Power Pool," outside the reach of Governor Abbott's August 2026 ERCOT audit; the [Interconnection Cost tool](https://open-factory.com/tools/interconnection-cost/) treats the two as different regimes. ## The GSU Nobody Prices Every press release in the pipeline names the prime mover and none names the transformer. A frame turbine or an aeroderivative generates at 11.5 to 15 kV. To move hundreds of megawatts across a campus, or to a 138 kV or 345 kV collector bus where the grid will eventually be tied in, that output goes through a generator step-up transformer, one per unit or one per pair, sized at 100 to 500 MVA. Wood Mackenzie's survey put the [average GSU lead time at 143 weeks in Q2 2025, with GSU demand up 274% since 2019 and a projected shortfall of more than 700 units in 2025](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/); Terrapin's June 2026 buyer note has [50 to 150 MVA GSUs at 100 to 150-plus weeks and 15 kV metal-clad switchgear at 60 to 80](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026); SemiAnalysis's grid piece says [gas turbine and GSU lead times "have each extended to three to four years, versus a historical norm of roughly 18 months"](https://newsletter.semianalysis.com/p/us-grid-constraints-towards-40gw). We covered the transformer side of this in [Transformer Lead Times Hit 143 Weeks](https://open-factory.com/articles/transformer-160-weeks/); here the point is where the GSU sits relative to the machine it serves. A new frame turbine is 130 to 208 weeks, a GSU 100 to 156 with 143 typical, a new large reciprocating engine 78 to 130, MV switchgear 44 to 104, an aeroderivative or mobile fleet 17 to 78, and a fuel-cell block 8 to 78 weeks from order to delivery. **A recip plant does not need a GSU to energize, and that is why engines are winning the first two years.** A Jenbacher J624 or a Caterpillar G3520 generates at 4.16 to 13.8 kV. The 25 MW packs VoltaGrid ships put that output straight onto medium-voltage switchgear and distribute across the campus at 13.8 or 34.5 kV through pad-mount unit transformers, the class the [Transformer Sizing and Cost tool](https://open-factory.com/tools/transformer-sizing-cost/) prices at 52 weeks or less. The step-up to transmission voltage happens only when the site chooses to tie into the grid, which most say they will do later. A 7HA.02 at 384 MW cannot do that. It needs a 400-plus MVA GSU on the day it fires, and that unit was ordered, if it was ordered at all, in the same quarter as the turbine. Fermi is the only developer in the pipeline whose filings say the substations and transformers were bought alongside the turbines, and it raised a [dedicated equipment facility "intended to support the procurement of long-lead electrical infrastructure required for Project Matador, including substations, transformers, breakers, and related high-voltage switchgear"](https://www.sec.gov/Archives/edgar/data/2071778/000207177826000010/frmi-20251231.htm). Chevron's Kilby release mentions "associated electrical infrastructure" and no date. For every frame-turbine campus targeting 2028, ask when the GSU purchase order was placed. If the answer is after mid-2025, 143 weeks puts delivery in 2028 at the earliest, before installation, oil fill and commissioning. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) does that arithmetic for a bill of gear. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes the GSU as its own line for on-site generation because no vendor quotes it inside the turbine package. The turbine side of the same gate is the one the market talks about. GE Vernova ended June 2026 with [116 GW of gas equipment under contract, 63 GW of it slot reservations, and booked 52 heavy-duty and 61 aeroderivative units in the quarter while shipping 3 GW](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm); Siemens Energy's CFO put lead times at ["three years or more"](https://www.utilitydive.com/news/siemens-gas-turbine-backlog-nears-70-gw-as-company-expands-manufacturing/827390/). Jefferies' February 2026 note, as reported by Latitude Media, has [large turbines more than 90% booked for 2028 and 70% for 2029, while behind-the-meter systems are about 65% and 33% booked for the same years, with at least 19 GW of available equipment capacity by 2028](https://www.latitudemedia.com/news/is-the-gas-turbine-bottleneck-solving-itself/). The same note flagged that "high-voltage equipment, such as transformers and switch gears, could see a supply crunch." That is the GSU. ## The Permit Is the Second Gate The equipment queue is one gate. The air permit is the other, and in 2026 it stopped more megawatts than any turbine shortage. Cleanview's report attributes the biggest delay to Project Jupiter in Doña Ana County, New Mexico, where [the state blocked a planned gas pipeline that would have fueled the project's 2.45 GW on-site plant, so the project "is likely to miss its target of coming online in 2027"](https://cleanview.co/reports/behind-the-meter-data-centers), and to Nebius in Vineland, New Jersey, which is [struggling to obtain an air permit to run a 400 MW on-site gas plant on which 7 of the 9 compute tranches of its $17.4 billion Microsoft deal depend](https://newsletter.cleanview.co/p/data-center-developers-want-to-build). Jupiter is a Bloom fuel-cell site, so its bottleneck is neither a turbine nor a GSU. It is a right of way. Texas shows the other pattern: small permits first, the big one later. Floodlight's review of TCEQ records found that Stargate Abilene's [fleet of 10 turbines and 62 backup diesel generators was permitted under permits by rule and standard permits that require no public notice, that the developers then filed a first major permit for 41 more turbines and 18 more generators, and that at least 38 Texas datacenters have received minor permits for on-site power since 2024, covering more than 2,100 backup generators](https://floodlightnews.org/ai-data-centers-texas-pollution-loophole/). A former EPA air enforcement chief told Floodlight the staggered approach "could violate" EPA aggregation policy. Cleanview's timeline for the same site: Crusoe filed for a [360 MW plant five months before Stargate was announced, got approval six days later, ordered 10 GE Vernova turbines in December 2024 and 19 more in June 2025](https://newsletter.cleanview.co/p/openais-stargate-data-centers-are). SemiAnalysis reads the TCEQ filing as [up to 933 MW of gross nameplate, about 665 MW of it the 19 LM2500s](https://newsletter.semianalysis.com/p/us-grid-constraints-towards-40gw). Georgia went the other way: after VoltaGrid installed [8 of 33 planned engines at a Serverfarm site in Covington before the permit issued, Georgia EPD wrote on July 2, 2026 that the company had likely violated state air rules and must pause construction](https://www.gpb.org/news/2026/07/07/data-center-power-plant-company-texas-gets-warned-it-may-be-violating-georgia-law). The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) logs each state's minor-permit threshold because that threshold, not the turbine catalog, sets how many megawatts a developer can energize before anyone can object. ## What the Grid Side Is Doing Meanwhile Three federal actions frame the behind-the-meter decision. The first is DOE's use of Federal Power Act section 202(c) to keep retiring plants running. DOE's own logs show the order numbers running from [202-25-3 for the J.H. Campbell coal plant on May 23, 2025](https://www.energy.gov/ceser/2025-doe-202c-orders) to [202-26-43 on September 3, 2026](https://www.energy.gov/ceser/2026-doe-202c-orders), with the plant-specific orders renewed every 90 days. Eight plants in seven states, 5,349 MW of nameplate by EIA's unit list: Campbell 1,561 MW, Schahfer 847, Eddystone 782, Centralia 730, Stanton 465, Craig 446, Wagner 415 and Culley 104. The Sierra Club's methodology, as reported by Utility Dive in March 2026, put the [ratepayer cost of the first 13 orders at more than $235 million](https://www.utilitydive.com/news/doe-202c-cost-235m-coal-sierra-club/815045/). For a developer the relevance is indirect but real: 5.3 GW of capacity that the RTOs had already planned around is being counted twice, once in the resource plans that retired it and once in the emergency orders that keep it, and none of it is available to a new 500 MW load in the queue. The second is FERC. On June 18, 2026 the commission issued show-cause orders to [all six RTOs and ISOs, finding their large-load interconnection rules appeared inadequate on five points: study process efficiency, cost-shift prevention, co-location and behind-the-meter generation, new transmission services for flexible loads, and a study process for generators serving "electrically proximate" loads, with 60 days to respond](https://www.utilitydive.com/news/ferc-doe-data-center-interconnection/823360/). The third is Texas, where ERCOT's [Batch Zero process, effective July 2026, adds two pathways for large users, the Withdrawal-Limited Private Use Network for customers who build their own on-site generation and the Provisional Controllable Load Resource for those who accept curtailment](https://www.ercot.com/files/docs/2026/06/18/ERCOT-Trending-Topic-New-Batch-Connection-Process-for-Large-Electricity-Users.pdf), against a queue of more than 438 GW. We covered that queue in [The 2% Funnel](https://open-factory.com/articles/ercot-funnel/). SemiAnalysis counts about [2,885 MW of announced ERCOT co-location with existing plants, led by AWS's 1,200 MW at Comanche Peak and CyrusOne's 400 MW at Thad Hill](https://newsletter.semianalysis.com/p/us-grid-constraints-towards-40gw); those are net-metering deals against existing steel, not new iron. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) includes the WLPUN and PCLR terms because a WLPUN is the legal form most Texas projects in our pipeline will take. ## What Actually Energizes by 2028 The forecast is built block by block from [data/btm-pipeline.csv](data/btm-pipeline.csv) and the lead times above, and it is an estimate, marked as such. The rule we applied: a megawatt counts in a year only if the equipment for it was ordered or leased early enough for the published lead time to land before that year ends, the air permit for it exists, and gas is contracted. Cleanview's end-2026 range of 2.8 to 3.2 GW is the anchor; SemiAnalysis's June 2026 title asks whether the market reaches ["40GW+ of behind-the-meter datacenter by 2028"](https://newsletter.semianalysis.com/p/us-grid-constraints-towards-40gw) and forecasts that behind-the-meter will power "well over half of new US datacenters in 2028+"; Cleanview's own high case is [13 GW by the end of 2027 if every project with a signed tenant hits its schedule, and as low as 5 GW if delays like Jupiter's and Vineland's spread](https://cleanview.co/reports/behind-the-meter-data-centers). Base case: 3.0 GW at end-2026, 7.4 GW at end-2027 and 15.0 GW at end-2028, of which 2.0 GW is the existing mobile and aero fleet, 1.4 GW new aeroderivatives, 7.5 GW reciprocating engines, 3.1 GW frame turbines and 1.0 GW fuel cells, with a low case of 10 GW and a high case of 22 GW for 2028. **The 2027 block is engines and jet engines, because those are the only orders placed in 2025 and 2026 that deliver inside the window.** Stargate Frontier in Shackelford County is the largest: Vantage ordered [more than 670 reciprocating engines for 2.58 GW](https://newsletter.cleanview.co/p/openais-stargate-data-centers-are), and Texas filings show [VoltaGrid approved to operate 210 Jenbacher units totaling 700 MW at that site](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/). VoltaGrid says its Oracle program is [2,300 MW of modular engine packs on Energy Transfer gas](https://voltagrid.com/voltagrid-collaborates-with-oracle-to-power-next-gen-ai-data-centers), that it has [ordered 1.5 GW, 300 Jenbacher J624 and J620 engines in 25 MW units, for delivery by 2028](https://voltagrid.com/voltagrid-orders1-5-gw-for-behind-the-meter-power-generation-from-innio), and that its [$5.0 billion financing funds a "4.3+ GW fully contracted power deployment plan through 2028"](https://voltagrid.com/voltagrid-closes-5-0-billion-comprehensive-financing-package-consisting-of-2-0-billion-of-senior-secured-second-lien-notes-and-3-0-billion-asset-based-loan-facility) against about 1.4 GW fielded. We give Frontier 0.7 GW in 2027 and the rest in 2028, the Oracle program 1.5 GW in 2027 and 1.4 GW in 2028 net of the Frontier overlap. The aero block is Abilene's 19 LM2500s, PROENERGY's [13 PE6000 packages for Crusoe](https://www.power-eng.com/gas/data-center-power-crunch-lifts-engines-aeroderivatives-into-larger-role/) and Fermi's 132 MW of TM2500s. We covered why those two technologies beat frames on a 100 MW bridge in [Recips vs Turbines vs Fuel Cells](https://open-factory.com/articles/turbines-recips-fuel-cells/). This is that finding at 10 GW scale. **The 2028 block is where the frame turbines and the GSUs start to bind.** Kilby's first power is "anticipated in 2028" after an FID at the end of 2026. Seven 7HA.02s under a slot reservation and twelve Titan 350s do not all fire in one year, and NGI's reading of Chevron's own capacity chart has the [2.67 GW reached around 2031](https://www.naturalgasintel.com/news/chevron-touts-natural-gas-fired-kilby-project-as-ai-power-demand-outruns-grid/), so we give it 0.7 GW in 2028. Homer City's 14 rows in 860M are all status P with the first 1,256 MW dated July 2028, and the campus is still negotiating its anchor tenant, so we give it 0.9 GW. Fermi's simple-cycle fleet is on site. The gate is the tenant that triggers vertical build, then the substation gear, so 1.0 to 1.5 GW by end-2028 is plausible and we take 1.3 GW. The engine block keeps growing: Joule Capital Partners' Utah campus expects a [1.5 GW first phase of 636 Caterpillar G3520K gensets to ramp by early 2028, fed by Kern River](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/); Nscale's Monarch campus in West Virginia has [2 GW of Caterpillar G3500-series engines](https://www.power-eng.com/gas/data-center-power-crunch-lifts-engines-aeroderivatives-into-larger-role/) tied to Microsoft; Wartsila's Ohio project has [40 34SG engines with delivery aligned to a COD in early 2028](https://www.power-eng.com/gas/data-center-power-crunch-lifts-engines-aeroderivatives-into-larger-role/). Against that, Evercore's David Raso told NGI that [new large engine orders are landing in late 2027 at the earliest, with most slotting into 2028](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/), and Caterpillar's own line is [24 to 30 months for orders placed in Q1 2026](https://www.power-eng.com/gas/data-center-power-crunch-lifts-engines-aeroderivatives-into-larger-role/). Anything not ordered by mid-2026 is a 2029 megawatt. We wrote up the genset side of that allocation in [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/). Fuel cells are the smallest block and the most permit-dependent. Bloom's 10-K says it [plans to double factory capacity from 1 GW to 2 GW by the end of 2026](https://www.sec.gov/Archives/edgar/data/1664703/000162828026006516/be-20251231.htm), and its Q2 2026 revenue was [$1.065 billion, up 166%, with full-year guidance raised to $3.9 to 4.2 billion](https://www.sec.gov/Archives/edgar/data/1664703/000162828026050150/ex991_q226financialresults.htm), so the factory is not the constraint. Jupiter's 2.45 GW is: without the pipeline it is 0 in 2027 and we carry 0.5 GW in 2028 on the assumption that a rerouted line or trucked gas covers the first phase. The other fuel-cell megawatts sit in Tallgrass and Crusoe's Cheyenne campus, which the county approved in January 2026 at [2.7 GW of "natural gas fired turbines, as well as fuel cells," scalable to 10 GW](https://cowboystatedaily.com/2026/01/08/huge-cheyenne-data-center-grows-to-2-7-gigawatts-nearly-triple-state-power-use/), and which Crusoe paused in June 2026 at its customer's request, per DCD and Bloomberg headlines we could not open. We carry 0.5 GW for it in 2028 and would cut that to zero if the pause holds into 2027. The [30% fuel-cell ITC](https://open-factory.com/articles/fuel-cell-itc-last-subsidy/) is why Bloom keeps winning the sites that have gas. ## The Vendors Are Building for the Engines The supply side has already voted. Caterpillar, whose large-engine backlog grew [more than 3.5x since January 2024](https://www.manufacturingdive.com/news/caterpillar-triple-power-generation-capacity-raises-2030-targets-q1-2026-earnings/), is [on track to add 15 GW of annual large-engine capacity on top of the 50 GW it had targeted for 2030, from a 2024 base of about 25 GW; Cummins plans to add 20 GW to reach 55 GW of high-horsepower output by 2030](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/). GE Vernova is going from [20 GW of annual heavy-duty output in Q3 2026 to 24 GW in 2028 and 30 GW in 2030](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm). INNIO booked a [1.1 GW order for more than 200 J624s for one campus in July 2026](https://www.innio.com/en/news-media/press-releases/innio-wins-1-1-gigawatt-gas-engine-order-for-major-data-center-campus/) and a [1.25 GW framework with Rehlko](https://www.innio.com/en/news-media/press-releases/innio-and-rehlko-enter-into-framework-agreement-for-1-25-gw-of-gas-engine-capacity/) in April. Cleanview's OEM count has [Caterpillar at 33% of tracked behind-the-meter projects with more than 8.8 GW of permitted Caterpillar engines and Solar Turbines, and Bloom at 14%](https://cleanview.co/reports/behind-the-meter-data-centers). Caterpillar 25 to 65 GW a year, Cummins 35 to 55, GE Vernova 20 to 30, Siemens Energy at a 13 GW run rate with no stated 2030 figure, Bloom at 2 GW by end-2026: the two engine makers add 70 GW of annual capacity by 2030 and the heavy-duty leader adds 10. The 90 GW was announced as a turbine market and the first 15 GW will be built as an engine market, because engines are 78 to 130 weeks from order, ship in 2.5 to 25 MW increments a permit-by-rule can cover, and plug into 13.8 kV switchgear without a 143-week transformer. Kilby, Homer City, GW Ranch and the back half of Matador are 2029 to 2031 projects with a 2028 press release, and what decides each is the GSU delivery date they have not published. The [Datacenter Electrical Roadmap](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) tracks both dates per campus. ## What to Do Monday If you are a developer holding a 2028 energization promise on a frame-turbine site, get the GSU purchase order date and the transformer OEM's slot in writing from your EPC this week. If the order was not placed by mid-2025, re-baseline to 2029 and tell your tenant before their tranche schedule does it for you. If you are choosing technology for a 100 to 500 MW bridge, price the engine plant at MV distribution against the turbine plant with its GSU in the [Genset TCO](https://open-factory.com/tools/genset-tco/) and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/), and read the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) line for on-site generation with the transformer broken out. If you are a factory or colocation buyer under 75 MW in Texas, file for the WLPUN pathway rather than firm service and check the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) for the security and study fees that come with it. And if you are an investor reading 90 GW, read 15. Behind the paywall: the twelve campuses ranked by what they can have running by end-2028, **8.8 to 14.2 GW against 30 GW announced**, the binding gate on each one, and the vendor-by-vendor winners and payers, all of it maintained in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) behind-the-meter table. --- # Compressed Air Is 10% of Your Factory's Power Bill URL: https://open-factory.com/articles/compressed-air-10-percent/ Date: 2026-04-23 Cooling / How-to About $6.6B a year, 2,962 DOE leak audits at a 4-month payback, 62 CAGI sheets from 17.2 to 22.8 kW per 100 cfm, VSD rebates of $75 to $220 per hp, the 200 hp sheet comparison A 200 hp rotary screw compressor running 6,000 hours a year at the 2025 US industrial rate of 8.62 cents per kWh costs about $70,000 a year in electricity, and over ten years it costs about $1.0M, of which $700,000 is the power bill. The DOE's own tip sheet says [about 10% of a typical plant's electricity goes to making compressed air](https://www.energy.gov/sites/prod/files/2014/05/f16/compressed_air1.pdf), that it takes 7 to 8 hp at the motor to deliver 1 hp of air at 100 psig, and that 76% of the machine's lifetime cost is electricity. This piece covers what the number is worth nationally, what the CAGI sheets say a kW of air should cost, what 2,962 DOE leak audits paid back, how the VSD retrofit pencils with and without a rebate, what heat recovery is worth, how public buyers actually paid for compressors, and how to buy one. The three stacks are $977,585 for a fixed-speed machine leaking 20% of its capacity, $883,577 for a VSD machine with the same leaks, and $800,753 for the VSD machine with leaks cut to 10%, so the leak program is worth $83,000 over ten years and the drive is worth $94,000, before any rebate. Every input is in [data/assumptions.csv](data/assumptions.csv) with its source. ## Where the 10% Comes From The 10% is old and it holds up. DOE's Industrial Technologies Program printed it in August 2004: [a DOE survey found that for a typical industrial facility approximately 10% of the electricity consumed is for generating compressed air, and for some facilities 30% or more](https://www.energy.gov/sites/prod/files/2014/05/f16/compressed_air1.pdf). The number underneath is the Xenergy motor-systems assessment for DOE, which measured [compressed air at 15.8% of all manufacturing motor-system energy, 27.7% in chemicals, and 4.6% in paper](https://energy.gov/sites/prod/files/2014/04/f15/mtrmkt.pdf), in a year when motor systems were 679 billion kWh and 23% of all US electricity sold. Run that forward. EIA's 2022 Manufacturing Energy Consumption Survey, released August 2025, puts [machine drive at 1,652 of the 3,182 trillion Btu of net electricity manufacturers used](https://www.eia.gov/consumption/manufacturing/data/2022/xls/Table5_4.xlsx), 52%. Compressed air at 15.8% of machine drive is 8.2% of manufacturing electricity, about 76 TWh, and at the [2025 US industrial average of 8.62 cents per kWh](https://www.eia.gov/electricity/data/eia861m/) that is $6.6B a year. Take DOE's 10% of a typical plant instead and it is 93 TWh and $8.0B. The [Cooling Roadmap](https://open-factory.com/roadmaps/cooling-roadmap/) includes compressed air as the largest cooling-adjacent electrical load in a plant because it is, and because it is the one with no analyst coverage and the fastest paybacks. At 900,000 kWh a year the system costs $58,950 in Texas at 6.55 cents, $76,680 in Ohio at 8.52, $91,260 in Illinois at 10.14, and $194,580 in California at 21.62. The US industrial average [rose from 6.81 cents in 2019 to 8.62 in 2025](https://www.eia.gov/electricity/data/eia861m/), 27%, which is why a leak that was worth fixing in 2019 is worth fixing sooner now. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) has the rate and capacity line for each of these states for anyone siting a plant. ## What a kW of Air Should Cost: 62 CAGI Sheets Every rotary screw compressor sold in the US by a Compressed Air and Gas Institute participant ships with a standard data sheet: rated capacity in acfm, full-load pressure, total package input power in kW, and the ratio, specific power in kW per 100 cfm, measured to [ISO 1217 Annex C and verified by CAGI's third-party program](https://www.cagi.org/performance-verification/). It is the closest thing this industry has to a miles-per-gallon sticker, and the sheets are public on the manufacturers' sites. We pulled 62 of them from Atlas Copco, Chicago Pneumatic, ELGi, Gardner Denver and Ingersoll Rand at 115 to 130 psig and put them in [data/cagi-sheets.csv](data/cagi-sheets.csv). **The spread inside a size class is wider than the spread between classes: 175 to 215 hp machines run 17.6 to 20.1 kW per 100 cfm, a 14% gap between the best and worst sheet at the same pressure.** The 15 to 30 hp class runs 20.5 to 22.8 and the 120 to 150 hp class 17.3 to 20.4. At 600 acfm and 6,000 hours, one kW per 100 cfm is 36,000 kWh and $3,100 a year at the US rate, so the 2.5 kW per 100 cfm between the [ELGi EG 160 at 20.05](https://www.elgi.com/us/wp-content/uploads/2020/06/EG-160-125.pdf) and the [Ingersoll Rand RS132i at 17.6](https://www.ingersollrand.com/en-us/resources/cagi-data-sheets/contact-cooled-rotary/) is $7,800 a year at full load, before anyone talks about drives. The sheet is the first number to ask for and the [Quote Check](https://open-factory.com/tools/quote-check/) tool takes the kW per 100 cfm as an input next to the price. The VSD sheets publish five or six operating points instead of one. The [Ingersoll Rand RS160n](https://www.ingersollrand.com/en-us/resources/cagi-data-sheets/contact-cooled-rotary/) goes from 18.15 kW per 100 cfm at 1,053 acfm to 17.17 at 564 acfm and back up to 19.12 at 224 acfm; the [Atlas Copco GA 160 VSDs](https://www.atlascopco.com/en-us/compressors/cagi-data-sheets) runs 17.8 at 842 acfm, 18.1 at 417 and 19.9 at 204. The curve is flat where a plant actually lives, 40% to 100% of capacity, and the 60 hp and 75 hp units are 3 kW per 100 cfm worse across the whole range than the 200 hp ones, which is the case for one big machine over two small ones when the load allows it. ## The Leak Audit: 2,962 DOE Recommendations DOE has paid university teams to walk small and mid-size plants since 1976, and the Industrial Training and Assessment Centers publish [every assessment and every recommendation with its cost, savings and whether the plant did it](https://itac.university/download): 23,155 assessments and 170,142 recommendations as of the September 2026 database. We took the FY2015 to FY2025 recommendations with a stated cost and a positive saving. Of the 5,817 assessments in that window, [73% carried at least one compressed-air recommendation](data/iac-compressed-air-measures.csv), which is a fair proxy for how many plants have not looked. Of the 2,962 leak recommendations (ARC 2.4236), 39.8% pay back in under three months, 22.0% in three to six, 17.6% in six to twelve, and 0.7% take more than five years. The medians are a $1,204 repair for a $3,267 annual saving, 42,600 kWh, a 0.35-year payback, and [70% of plants implemented it](data/iac-compressed-air-measures.csv), the highest implementation rate of any compressed-air measure in the database. For plants with 300 to 500 employees the medians are a $1,697 repair and a $5,236 saving. Only 3.8% of the leak recommendations record a utility rebate, against 20.8% of the adjustable-speed-drive recommendations, so the leak program is the one measure here that does not need a program administrator. Lowering the header pressure is the cheapest line in the database: 1,559 recommendations, a median cost of $200, a median saving of $2,063, a 0.10-year payback, and half the plants did it. The DOE sourcebook's rule is [1% of power for every 2 psi of discharge pressure near 100 psig](https://www.compressedairchallenge.org/data/sites/1/media/library/sourcebook/compressed_air_sourcebook.pdf), so 10 psi on the worked plant's 134.9 kW average is 40,500 kWh and $3,490 a year. Upgrading compressor controls (451 recommendations) costs $6,534 for $6,325 a year at 1.25 years and only 38% implement it; buying a right-sized compressor (299) costs $23,000 for $9,523 a year, 2.5 years, 43%. **Two thirds of the money DOE's engineers find in a compressor room costs less than $2,000 to take.** ## The VSD Retrofit, With and Without a Rebate Now the worked plant: 400 people, one 200 hp class machine at 125 psig, average demand 600 acfm, 6,000 hours a year. ITAC plants with 300 to 500 employees (494 of them since FY2018) [report a median $598,853 electricity bill, 10.3 GWh, and $214,152 of demand charges](https://itac.university/download), so the compressor is 12% to 14% of that plant's kWh, right on DOE's number. The fixed-speed baseline comes straight off the sheet. The ELGi EG 160-125 draws [170.46 kW at its 850 acfm rating and 49.43 kW at zero flow](https://www.elgi.com/us/wp-content/uploads/2020/06/EG-160-125.pdf), so a load/unload machine at 600 acfm averages 0.706 × 170.46 + 0.294 × 49.43 = 134.9 kW, and that is a floor because it ignores blowdown and the receiver. The VSD alternative at 600 acfm interpolates to 103.3 kW on the RS160n sheet and 106.8 kW on the GA 160 VSDs sheet, mean 105.1 kW. The saving is 29.8 kW, 178,700 kWh, [$15,400 a year at 8.62 cents](data/plant-model.csv), $11,700 in Texas, $18,100 in Illinois and $38,600 in California. | Line | Value | Basis | |---|---|---| | Fixed-speed average input, 600 acfm, load/unload | 134.9 kW | ELGi EG 160 sheet: 0.706 × 170.46 + 0.294 × 49.43 | | Same, Atlas Copco GA 160s baseline (40.7 kW zero flow) | 123.0 kW | 0.601 × 177.6 + 0.399 × 40.7 | | VSD input at 600 acfm (mean of RS160n and GA 160 VSDs) | 105.1 kW | CAGI part-load points | | Saving vs ELGi baseline | 29.8 kW, $15,401/yr | 6,000 h, 8.62 c/kWh | | Saving vs Atlas Copco baseline | 16.2 kW, $8,390/yr | same | | Leaks at 20% of 850 acfm | 170 acfm, 34.1 kW, $17,629/yr | 20.05 kW per 100 cfm | | Cut leaks to 10% | 17.0 kW, $8,814/yr | DOE sourcebook target | | Header pressure down 10 psi | $3,488/yr | 1% per 2 psi | | Heat recovery, 2,000 heating hours | 427 MMBtu, $2,757/yr gas | 85% heat, 70% recovered, $5.17/MMBtu, 80% furnace | [CSV: plant-model.csv](data/plant-model.csv) The baseline row decides the case. Against the cheapest fixed-speed sheet the drive pays back its assumed $30,000 premium in 1.9 years unaided; against an Atlas Copco GA 160s that unloads to 40.7 kW, the saving halves and the payback is 3.6 years. This is why the Xcel Minnesota program [funds a supply-side study before it will rebate a compressor of 50 hp or more](https://www.xcelenergy.com/staticfiles/xe-responsive/Programs%20and%20Rebates/Business/MN-Compressed-Air-Info-Sheet.pdf): $2,000 plus $20 per hp for a 50 to 199 hp system, the whole study cost covered if you fix 75% of the leaks it finds, and up to $500 per peak kW saved on the custom side. Xenergy found [adjustable-speed drives on 1.7% of compressed-air systems](https://energy.gov/sites/prod/files/2014/04/f15/mtrmkt.pdf) in 1998; the 20.8% rebate flag on VFD recommendations in the ITAC data says the utilities have been paying for the catch-up ever since. On a 200 hp machine the prescriptive rebate is $15,000 in Wisconsin at [Focus on Energy's $75 per hp](https://assets.focusonenergy.com/production/docs/business/BI-Summary-of-Services-and-Incentives-2026.pdf), $25,000 at [Ameren Illinois's $125 per hp, capped at 80% of the drive's cost](https://cdn.amerenillinoissavings.com/savings/2025/12/App-Specialty-Industrial-2026_Rev02.01.pdf), $34,000 at [Consumers Energy's $170 per hp for one compressor running 6,000 hours or more](https://www.clearesult.com/partner-hub/program-resources/consumers-energy-business/Incentive-Application), and $44,000 at the $220 per hp that [ComEd's trade allies publish for integrated VSD compressors up to 200 hp](https://fluidairedynamics.com/blogs/articles/comed-compressed-air-energy-incentives), a figure ComEd's own page does not print. In Michigan and Illinois the rebate is larger than the drive premium; in Wisconsin it halves the payback to a year. The 200 hp comparison behind the paywall ranks the eight sheets and the seven programs side by side. ## Heat Recovery Is a Rebate Play The sourcebook says [80 to 93% of a compressor's electrical input leaves as heat and 50 to 90% of that is recoverable](https://www.compressedairchallenge.org/data/sites/1/media/library/sourcebook/compressed_air_sourcebook.pdf) from an air-cooled screw package with nothing more than ducting and a thermostatic damper. For the worked plant's 105 kW VSD over 2,000 heating hours that is about 125,000 kWh of heat, 427 MMBtu, and at the [2025 US industrial gas price of $5.17 per MMBtu](https://www.eia.gov/dnav/ng/hist/n3035us3m.htm) through an 80% furnace it displaces $2,757 of gas a year. ITAC's 663 heat-recovery recommendations have a median cost of $6,000 and a median saving of $5,708, 1.27 years, and only 23% of plants do it, the lowest implementation rate of any measure here. The ones that pay are the ones with electric heat or a process water load. What changes the arithmetic is that the utilities pay per horsepower, not per therm. [Focus on Energy pays $75 per hp for compressed air heat recovery, up to 50% of project cost](https://assets.focusonenergy.com/production/docs/business/BI-Summary-of-Services-and-Incentives-2026.pdf), Ameren Illinois [$50 per hp for ducting air-cooled compressor exhaust into space normally heated by gas](https://cdn.amerenillinoissavings.com/savings/2025/12/App-Specialty-Industrial-2026_Rev02.01.pdf), and Consumers Energy [$50 per hp for its gas customers](https://www.clearesult.com/partner-hub/program-resources/consumers-energy-business/Incentive-Application). On 200 hp that is $10,000 to $15,000 against a ducting job that ITAC prices at a median $6,000, which is a project that pays for itself the day the rebate check clears. ## What Public Buyers Paid Nobody publishes transaction prices for industrial compressors, so we did what the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) does for transformers and chillers and read the public record. Federal contract awards on USAspending under product code 4310 and the compressor keywords, plus the council records we could reach, give 38 rows in [data/compressor-public-awards.csv](data/compressor-public-awards.csv), of which 23 are one rotary screw package with a stated horsepower and no dryer or installation. The 23 awards run from $387 to $1,628 per hp with a median of $637 (n=23, May 2022 to February 2026). The spread is not size: GSA bought [a 50 hp, 220 cfm rotary screw through Grainger for $37,690 in February 2026](https://www.usaspending.gov/award/CONT_AWD_47QSCC26F1W2E_4732_47QSCC21A0001_4732) and [a 50 hp, 460 V rotary screw through the same Grainger schedule for $81,398 in June 2025](https://www.usaspending.gov/award/CONT_AWD_47QSCC25F5SNM_4732_47QSCC21A0001_4732), a 2.2x gap on one line description, which is the [Grainger pricing story](https://open-factory.com/articles/grainger-confession/) in one product class. The big machines clustered low: [a Kaeser CSD 90, a 125 hp frame, for $57,835](https://www.usaspending.gov/award/CONT_AWD_W50S8L23P0014_9700_-NONE-_-NONE-) to the Army in 2023 at $463 per hp, [an Ingersoll Rand RS55i-A125 at 75 hp for $39,100](https://www.usaspending.gov/award/CONT_AWD_FA481922F0047_9700_47QSWA19D001U_4732) to the Air Force in 2022, and [a 100 hp inverter-type compressor for $52,190](https://www.usaspending.gov/award/CONT_AWD_SPE8E322FJ6K3_9700_SPE8E322D0010_9700) through DLA in 2022, the only VSD unit in the set, at $522 per hp. Add a dryer and the number moves: [a 100 hp air-cooled rotary screw with dryer was $128,080](https://www.usaspending.gov/award/CONT_AWD_140R3024P0078_1425_-NONE-_-NONE-) for the Bureau of Reclamation in Nevada in 2024. Municipal records show the service tail. Riverside Public Utilities paid [Atlas Copco $154,628 for a compressor replacement with a five-year extended warranty](https://riversideca.legistar.com/LegislationDetail.aspx?ID=35272) at its Clearwater plant in October 2021; Silicon Valley Power authorized [up to $1,500,000 for two compressors, two dryers and preventive maintenance with an optional ten-year extension](https://santaclara.legistar.com/LegislationDetail.aspx?ID=25488) in September 2025; and Phoenix bought [a Sullair 1809E for $13,748 after sending the RFQ to three vendors and getting one bid](https://phoenix.legistar.com/LegislationDetail.aspx?ID=13170) in 2019. The [BLS index for air and gas compressor manufacturing rose 55% from January 2020 to March 2026](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCU333912333912) and 6.2% in the twelve months to March 2026, so a 2022 award is 10% to 15% under today's quote. The compressor index outran the pumps-and-compressors group (52%) and its 2023 step, from 276 to 328 in twelve months, is the same one the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) saw in switchgear and transformers. The federal efficiency floor that landed in the middle of this, [DOE's January 2020 compressor standard with compliance required from January 10, 2025](https://www.federalregister.gov/documents/2020/01/10/2019-26355/energy-conservation-program-energy-conservation-standards-for-air-compressors), is the reason CAGI added isentropic efficiency to its sheets; DOE [proposed in May 2025 to withdraw the determination that compressors are covered equipment](https://www.federalregister.gov/documents/2025/05/16/2025-08582/energy-conservation-program-proposed-withdrawal-of-determination-of-compressors-as-a-covered) at all, with comments closed July 15, 2025, so the sheet, not the rule, is what protects the buyer. ## How to Buy a Compressor, and What to Do Monday Buy the sheet before the machine. Ask every bidder for the CAGI data sheet at your pressure, and if a bidder does not participate in the CAGI program treat the missing sheet the way the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) treats a missing lead time: as the answer. Convert kW per 100 cfm to dollars with your rate and hours; at 600 acfm, 6,000 hours and 8.62 cents each kW per 100 cfm is $3,100 a year and $31,000 over ten. Ask for the zero-flow kW on the fixed-speed sheet and the part-load points on the VSD sheet, because the load/unload floor is what the drive competes against. Then price the package the way the [Quote Check](https://open-factory.com/tools/quote-check/) tool does, at $/hp against the $387 to $1,628 public band, and write the RFQ the way [our capital-equipment RFQ piece](https://open-factory.com/articles/rfq-capital-equipment/) lays out, with the dryer, receiver, startup and first-year service as separate lines so the $37,690 and $81,398 versions of the same 50 hp machine cannot hide in one number. Check the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) for the drive and the motor before promising a turnaround date. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) will tell you whether the rebate's pre-approval window closes before the machine lands. Monday, in this order. Pull the last twelve bills and mark the compressor's share. If you cannot, put a $300 clamp meter on the feeder for a week, because the [ITAC data](https://itac.university/download) says 73% of plants have not. Walk the plant on a quiet shift with an ultrasonic gun and tag every leak. The median repair is $1,200 for $3,300 a year and pays back in four months. Drop the header pressure 2 psi at a time until someone complains, then go back up 2: the median cost is $200. If you are in PJM, put the compressors on the receiver during the [five summer peak hours that set the capacity charge](https://open-factory.com/articles/pjm-capacity-factory-bill/): 30 kW off the peak-load contribution is $3,400 a year at the Met-Ed 2026/27 rate of $113.65 per kW on the [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/), and the same 30 kW nominated into ELRP is another $3,400. Then, and only then, get the VSD quote with the sheet, the rebate pre-approval and the [spares and service terms](https://open-factory.com/articles/recommended-spares-margin/) attached, and put the [MRO basket](https://open-factory.com/articles/mro-price-basket/) filters and separator elements into the same negotiation. Behind the paywall: the eight 175 to 215 hp CAGI sheets ranked by kW at 600 acfm, where the gap between the best VSD and the worst fixed-speed machine is **$163,000 of electricity over ten years**, and the seven rebate programs ranked by what they pay per hp, in the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). --- # Five Majors Bought Their Way Into Liquid Cooling. Here Is the Price List URL: https://open-factory.com/articles/cooling-ma-price-list/ Date: 2026-04-19 Cooling / Take Eaton and Boyd $9.5B at 22.5x, Ecolab and CoolIT $4.75B at 29x, Schneider and Motivair $850M, Vertiv and PurgeRite, Trane and LiquidStack, what 29x does to your CDU quote Ecolab agreed on March 20 to pay [$4.75 billion for CoolIT Systems](https://www.sec.gov/Archives/edgar/data/31462/000110465926032446/tm269446d1_ex99-1.htm), 29 times the next twelve months' adjusted EBITDA on about $550 million of CDU and cold-plate sales. Eight days earlier Eaton had closed its [$9.5 billion purchase of Boyd Thermal](https://www.boydcorp.com/boyd-completes-the-sale-of-its-thermal-business-to-eaton-for-9-5-billion.html), agreed in November at 22.5 times projected 2026 EBITDA, and nine days before that Trane closed LiquidStack for a price it will not say. Five majors have now bought their way into liquid cooling: Schneider, Vertiv, Eaton, Trane and Ecolab, for $16.3 billion of disclosed prices in seventeen months. This piece lists every deal with its multiple and date, works out what each buyer must now earn per dollar of sales, lines up the revenue each has promised investors, shows where the dollars sit inside a GB300 rack, and ends with the negotiating consequence for anyone buying CDUs in 2026 and 2027. Nine acquisitions between October 17, 2024 and March 20, 2026, four with a disclosed multiple: a mid-single-digit multiple of revenue, 11.5x EBITDA, 10.0x EBITDA, 22.5x EBITDA and 29x EBITDA, in that order of announcement. [CSV](data/01-deal-table.csv). ## The Price List, October 2024 to March 2026 **The multiple a major pays for a liquid cooling company went from about 5x revenue in October 2024 to 29x EBITDA in March 2026, and the buyers say so in their own releases.** Schneider Electric opened the sequence on October 17, 2024 with [$850 million in cash for a 75% controlling interest in Motivair](https://www.se.com/ww/en/assets/564/document/492190/Schneider-Electric-acquires-Motivair-Corporation.pdf), a Buffalo maker of CDUs, rear-door heat exchangers, cold plates and chillers with over 150 employees. The release says the price "values Motivair at a mid-single digit multiple of projected FY2025 revenue" and that Schneider "expects to acquire the remaining 25% of non-controlling interests in 2028." Flex bought the cold-plate startup [JetCool on November 14, 2024](https://investors.flex.com/news/news-details/2024/Flex-Acquires-JetCool-to-Expand-Data-Center-and-Power-Portfolio/default.aspx) for an undisclosed sum. Vertiv did two deals. It paid [$200 million for Great Lakes Data Racks](https://investors.vertiv.com/news/news-details/2025/Vertiv-To-Acquire-Custom-Rack-Solutions-Manufacturer/) in July 2025 at "approximately 11.5x expected 2026" EBITDA including synergies, then on November 3, 2025 announced [about $1.0 billion for PurgeRite](https://www.prnewswire.com/news-releases/vertiv-announces-intent-to-acquire-purgerite-a-leading-provider-of-specialized-fluid-management-services-to-expand-liquid-cooling-services-portfolio-302602394.html), a Houston loop-flushing contractor, at "approximately 10.0x expected 2026" EBITDA including cost synergies, plus "up to $250 million in cash based on achieving certain 2026 performance metrics." That deal [closed December 4, 2025](https://www.sec.gov/Archives/edgar/data/1674101/000167410125000031/exhibit991vrt-12042025.htm). PurgeRite is a service business; Vertiv already sells the CDU and bought the crew that commissions the loop. Eaton signed for Boyd Thermal on November 2, 2025. Its [8-K](https://www.sec.gov/Archives/edgar/data/1551182/000155118225000033/etn09302025exhibit99.htm) gives the price, $9.5 billion, a headcount of more than 5,000 and an expected close in the second quarter of 2026; the announcement release, as carried by [Pulse 2.0](https://pulse2.com/eaton-to-buy-boyd-thermal-for-9-5-billion-expanding-data-center-liquid-cooling-and-aerospace-capabilities/) and [MergerSight](https://www.mergersight.com/post/eaton-corporation-s-9-5bn-acquisition-of-boyd-thermal), puts Boyd Thermal's 2026 sales at $1.7 billion, $1.5 billion of it liquid cooling, and the price at 22.5x projected 2026 adjusted EBITDA, accretive in the second year. Boyd closed on March 12, 2026, a quarter early. The day after Eaton signed, Daikin Applied bought [Chilldyne](https://www.storagereview.com/news/chilldyne-acquired-by-daikin-applied-americas), a negative-pressure direct-to-chip vendor, terms undisclosed. Trane announced [LiquidStack on February 10, 2026](https://www.dallasinnovates.com/north-texas-liquidstack-to-be-acquired-by-ireland-based-trane-technologies/) and [closed on March 3](https://investors.tranetechnologies.com/news-and-events/news-releases/news-release-details/2026/Trane-Technologies-Completes-Acquisition-of-LiquidStack/default.aspx). LiquidStack sells immersion and direct-to-chip systems and a "GigaModular" CDU rated to 10 MW from Carrollton, Texas. Its last round was a $20 million Series B in fall 2024. No price, no multiple, no revenue. Trane's [10-K](https://www.sec.gov/Archives/edgar/data/1466258/000162828026005731/tt-20251231.htm) lists "data center liquid cooling solutions" among its products. The [Open Factory OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) counts Trane, Daikin and Flex among the buyers that told the market nothing about what they paid. Then Ecolab. KKR had bought CoolIT in May 2023, a price [KKR's release did not state](https://www.coolitsystems.com/resources/news/kkr-to-acquire-coolit-systems/) and trade press put at about $270 million (Data Center Dynamics, May 2023; we could not open the page this session). KKR says the sale to Ecolab returned [15 times its equity](https://www.esgtoday.com/kkr-impact-fund-earns-15x-investment-on-sale-of-liquid-data-center-cooling-provider-coolit-to-ecolab/). [BetaKit](https://betakit.com/ecolab-to-acquire-coolit-systems-for-4-75-billion-usd/) traces CoolIT's revenue from $60 million to $155 million in 2024, against the $550 million Ecolab now expects for the next twelve months. Ecolab expects to close in the third quarter. Outside the five, Samsung paid [EUR 1.5 billion for FläktGroup](https://www.coolingpost.com/world-news/samsung-completes-e1-5bn-flaktgroup-purchase/), an air-handling group with [more than EUR 700 million of revenue and 14 plants](https://www.triton-partners.com/news/triton-completes-sale-of-flaktgroup-to-samsung-electronics), closing November 6, 2025. That is the cheap end of the table, and it is cheap because FläktGroup moves air. ## What 29x Means Per Dollar of Sales CoolIT at 8.6x forward sales, Boyd Thermal at 5.6x, Motivair at about 5x on Schneider's words (grossed to 100%, the stake implies an enterprise value near $1.13 billion and 2025 revenue near $230 million), FläktGroup at 2.1x. The [CSV](data/02-ev-per-sales.csv) and the [implied EBITDA table](data/02b-implied-ebitda.csv) carry the arithmetic. **A 30% EBITDA margin bought at 29x leaves no room for a price concession: every point of CDU price Ecolab gives away costs it about $160 million of the value it just paid.** Divide $4.75 billion by 29 and CoolIT's next-twelve-month adjusted EBITDA is about $164 million on $550 million of sales, a 29.8% margin. Divide by 24 and 2027 EBITDA is about $198 million, so the deal assumes 21% EBITDA growth in one year with the margin held. One point of price on $550 million of sales is $5.5 million of EBITDA; at 29x that is $160 million of enterprise value. Ecolab's [July 2 release](https://investor.ecolab.com/news/news-details/2026/Ecolab-Closes-CoolIT-Acquisition-and-Expands-AI-Cooling-Platform-as-Global-High-Tech-Business-Targets-4-Billion-by-2030/default.aspx) confirms the plan: Global High-Tech is to reach $4 billion of sales by 2030 "with operating income margins of 25%." A chemistry company did not pay 29x to sell CDUs at a discount. It paid to attach fluids, filtration and service contracts to the hardware, which is where the pressure will land. Boyd's arithmetic is different. $9.5 billion at 22.5x is about $422 million of 2026 EBITDA on $1.7 billion of sales, a 24.8% margin, which is exactly Eaton's own [record 2025 segment margin of 24.5%](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000002/etn12312025exhibit99.htm). Eaton cannot make Boyd accretive by squeezing a margin already at group level. It needs Boyd's sales to grow, and its release says accretion arrives in year two. Eaton is a volume buyer. Vertiv paid 10.0x for PurgeRite and 11.5x for Great Lakes, both bolt-ons to a company that [guides 22% to 23% adjusted operating margin for 2026](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm) on $13.25 to $13.75 billion of sales. Vertiv bought the commissioning crew and the rack shop, and its $250 million PurgeRite earnout is written on 2026 metrics. No public list price exists for a merchant CDU, and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds no CDU award yet, which is why the [Open Factory Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) treats the CDU as a v1 cell built from contributed quotes. What the deals give us instead is the floor under the vendor's price: the margin the acquirer has capitalised at 22x to 29x. Read a CDU quote against that floor in [Quote Check](https://open-factory.com/tools/quote-check/), not against the vendor's cost. ## The Revenue Each Buyer Has Promised Vertiv $13.5 billion at the 2026 midpoint, Ecolab's Global High-Tech $4.0 billion by 2030 from about $1.5 billion annualized in mid-2026 and $150 million in 2021, Boyd Thermal $1.7 billion in 2026, Modine's single-customer agreement at more than $4 billion over 2027 to 2029, Munters Data Center Technologies about SEK 6 billion in 2025, CoolIT $550 million, Motivair about $230 million implied. The [CSV](data/03-acquirer-targets.csv) lists the date and status of each figure. **Ecolab needs Global High-Tech to compound at 28% a year from 2026 to 2030 in a liquid cooling market Dell'Oro sizes at $7 billion in 2029.** [Dell'Oro's January 2026 forecast](https://www.delloro.com/news/data-center-liquid-cooling-market-to-approach-7-billion-by-2029-as-ai-deployments-accelerate/) puts the market near $3 billion in 2025, names Vertiv the leader with CoolIT, nVent and Boyd behind it, and calls Aaon the fastest grower. Two of those four were bought inside five months. Vertiv's numbers explain the hurry: [backlog of $15.0 billion at December 31, 2025, up 109%](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm), a fourth-quarter book-to-bill of about 2.9x, organic sales guided up 27% to 29% in 2026, and a [10-K](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm) that says most of that backlog ships within 12 to 18 months. The [Open Factory Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) includes Vertiv's 2.9x next to Trane's applied book-to-bill of 200% from the same quarter. Our [estimate of Vertiv's backlog after it stopped disclosing](https://open-factory.com/articles/vertiv-backlog-estimate/) is the companion piece. Eaton's Q4 release reports [Electrical Americas orders up 16% "driven by data center momentum"](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000002/etn12312025exhibit99.htm), Electrical backlog up 29%, and 2026 organic growth of 7% to 9% on $27.4 billion of 2025 sales. Schneider's Motivair release put Data Center & Networks at [21% of FY2023 group orders](https://www.se.com/ww/en/assets/564/document/492190/Schneider-Electric-acquires-Motivair-Corporation.pdf). Trane and Daikin have published no data center revenue figure at all. nVent, the one Dell'Oro top-four vendor nobody has bought, reported [2025 sales of $3.9 billion, up 30%](https://www.sec.gov/Archives/edgar/data/1720635/000162828026005930/q42025nvtpressrelease.htm), guided 2026 reported growth of 15% to 18%, and set [three-year targets of 10% to 13% organic growth and about 22% adjusted operating margin](https://www.globenewswire.com/news-release/2026/03/18/3257988/0/en/nvent-highlights-portfolio-transformation-and-growth-priorities-at-2026-investor-day.html) at its March 18 investor day. ## Where the Dollars Sit in the Rack $49,860 per GB300 NVL72 rack: 18 compute trays at $2,260 each, $40,680, plus 9 NVSwitch trays at $1,020 each, $9,180. Vera Rubin NVL144 rises to $55,710, with compute trays at $2,660 and switch trays down to $870. The [CSV](data/04-rack-cooling-bom.csv) holds the rows. **The $49,860 Morgan Stanley counts is the ODM's bill, not the developer's: at roughly $420 per kW it is about an eighth of the mechanical dollars in a liquid-cooled hall, and the other seven-eighths is the CDU, the piping and the plant.** [Tom's Hardware reported](https://www.tomshardware.com/pc-components/cooling/cooling-system-for-a-single-nvidia-blackwell-ultra-nvl72-rack-costs-a-staggering-usd50-000-set-to-increase-to-usd56-000-with-next-generation-nvl144-racks) the Morgan Stanley figures on November 5, 2025: cold plates at $300 per CPU or GPU and $200 per NVSwitch ASIC on GB300, $400 on Vera Rubin, and a 17% step from GB300 to Rubin after a 20% step from GB200. Eighteen compute trays at 6.6 kW is about 119 kW, so the in-rack content is about $420 per kW. Turner & Townsend's [2025 index](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends) puts mechanical at 33% of a liquid-cooled US build against 22% air-cooled; at Phoenix's $9.8 per watt that is about $3,230 per kW of mechanical, so the rack-internal content is about 13% of it. The rest is what the five majors bought: the CDUs Motivair, CoolIT, LiquidStack and Boyd sell, the loop PurgeRite flushes, the chillers and dry coolers that reject the heat. Our [chiller piece](https://open-factory.com/articles/chiller-60-week-problem/) covers the plant side. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes the CDU at 30 to 50 weeks (Origin Partners, May 2026). The in-rack part has its own consolidation, in Taiwan rather than Ohio. [AVC holds an estimated 40% to 50% of GB200 and GB300 cold plates](https://t.media/3179423), plans to lift cold plate capacity from 200,000 units at the end of 2025 to one million by the end of 2026, and was named one of four preferred Vera Rubin cold plate suppliers at GTC 2026. SemiAnalysis lists the gold-plated cold plate field for Rubin as [AVC, Delta, Boyd, CoolIT and Auras](https://semianalysis.com/2026/02/25/vera-rubin-extreme-co-design-an-evolution/). Two of those five are now Eaton and Ecolab. The other three sell to the server ODM, not the developer, which is why the developer's cooling money runs through the CDU. The [800 VDC piece](https://open-factory.com/articles/800vdc-strands-frame-agreements/) covers what Rubin does to the rack power side of the same frame agreement. ## What to Do Monday The CDU order is the one line in the mechanical package that five acquirers now need to justify $16.3 billion, and the buyer's response is procedural. Put the CDU on the long-lead list with the chillers: at 30 to 50 weeks it belongs in the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) before the mechanical subcontract is let, per the [OFCI argument](https://open-factory.com/articles/ofci-vs-design-build-chillers/). Price it per kW of rated capacity at your approach temperature, not per unit. Split the RFQ into hardware, fluids and commissioning services so that a 25% margin service tail cannot hide inside a hardware price. The [RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) has the split. Attach [liquidated damages](https://open-factory.com/articles/liquidated-damages/) to the delivery date, because a company that just paid 29x will protect price before it protects your schedule. Run the Taiwan alternates through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) before assuming they land cheaper. Behind the paywall: which of the five acquirers needs volume and which needs margin, deal by deal, the two earnout dates that keep Motivair's and PurgeRite's sales teams hungry through **2028**, and the nine vendors still outside the majors, in the [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # Recips vs Turbines vs Fuel Cells for a 100 MW Bridge URL: https://open-factory.com/articles/turbines-recips-fuel-cells/ Date: 2026-03-24 Power / TCO 16 to 18 months vs 24 to 30 vs 36+, $/kW from Wartsila, Caterpillar, Bloom and GE Vernova deals, heat rate, permitting, resale, Oracle's 2.8 GW Bloom MSA A 100 MW bridge plant run for five years costs $96 per MWh on solid oxide fuel cells, $104 on gas reciprocating engines and $112 on aeroderivative gas turbines. That is at 85% utilization with gas delivered at $3.78/MMBtu, and the ranking holds at every gas price from $2 to $8 and at every utilization from 40% to 95%. It flips twice, and only twice: take away the [30% fuel-cell investment tax credit](https://www.lw.com/en/insights/one-big-beautiful-bill-new-law-disrupts-clean-energy-investment) and fuel cells go to $128 and lose to engines by $24; price the turbines at their [2023 installed cost of $1,606/kW](https://www.eia.gov/analysis/studies/powerplants/capitalcost/pdf/capital_cost_AEO2025.pdf) instead of what GE Vernova charges for an aero package in 2026 and turbines win at $84. The rest of the spread is lead time, which is not in the $/MWh at all: 17 months to first power for fuel cells, 27 for engines, 30 for turbines. The [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes all of it, with every input and its source in one CSV. We cover the three lead times and where they come from, what each technology costs per kW installed after the 2026 repricing, heat rate and fuel at Henry Hub plus basis, fixed and variable O&M, the air-permit route each one takes at 100 MW, resale at year five, the five-year total and $/MWh with sensitivities to gas and utilization, and behind the wall, the technology to buy in Texas, Virginia, Ohio, Arizona and Georgia and the price book of named deals. Fuel cells at 16 to 18 months, engines at 19 to 30 and aeroderivative turbines at 18 to 36 all beat the 36-month grid connection in Dallas or Phoenix and beat Columbus and Northern Virginia, at 84 months, by four to five years. ## Three Lead Times, Three Kinds of Factory The lead times come from a utility planner, not a vendor. Benjamin Jordan of CPS Energy put the delivery, installation and commissioning timeline at 50 MW-plus scale at [16 to 18 months for solid oxide fuel cells and linear generators, 24 to 30 months for large reciprocating engines and 30 months or more for gas turbines](https://avanzaenergy.substack.com/p/data-centers-are-killing-the-grid) at PowerGen International in early 2026, against 36 to 84 months for a grid connection. Those are whole-project numbers. Bloom Energy's own headline is faster for one machine: it [delivered a fully operational system to Oracle in 55 days](https://www.bloomenergy.com/news/bloom-energy-and-oracle-expand-strategic-partnership-to-deploy-up-to-2-8-gw-to-accelerate-ai-infrastructure-build-out/), a month ahead of a 90-day plan, in 2025. A hundred megawatts is a different job, with gas interconnection, switchgear and a substation, which is why the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists the 16 to 18 month figure and not the 55 days. The three factories behind those numbers are in different states. Bloom is expanding Fremont from [1 GW to 2 GW of annual capacity by the end of 2026](https://www.sec.gov/Archives/edgar/data/1664703/000162828026006516/be-20251231.htm) and says each further 1 GW increment takes six to nine months and $100 million to $150 million of capital. Its Q2 2026 revenue was [$1.065 billion, with product revenue up 215% to $935 million](https://www.sec.gov/Archives/edgar/data/1664703/000162828026050150/ex991_q226financialresults.htm), and full-year guidance is $3.9 billion to $4.2 billion. **Fuel cells are the only one of the three technologies where the supplier's capacity is growing faster than its backlog.** Engines are the opposite. Caterpillar's firm backlog reached [$72.1 billion at June 30, 2026](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm), and its large reciprocating engine backlog had grown [more than 3.5 times since January 2024](https://www.manufacturingdive.com/news/caterpillar-triple-power-generation-capacity-raises-2030-targets-q1-2026-earnings/). Evercore's David Raso told NGI in May that [new large engine orders land in late 2027 at the earliest, with most slotting into 2028](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/); Caterpillar is targeting 65 GW a year of engine capacity, Cummins 55 GW of high-horsepower output by 2030. Our [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/) piece has the quarter-by-quarter backlog. The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) updates it each quarter. Turbines split by size. Wood Mackenzie counted [110 GW of global gas turbine orders in 2025 against 60 to 70 GW of manufacturing capacity](https://www.woodmac.com/press-releases/gas-turbine-prices-soar-195-as-market-faces-supply-demand-crisis/). GE Vernova says heavy-duty lead times are [about three years, with roughly 10 GW of slots left across 2029 and 2030](https://www.naturalgasintel.com/news/natural-gas-turbines-arent-gating-data-center-buildouts-ge-vernova-says/); Siemens Energy says [three years or more](https://www.utilitydive.com/news/siemens-gas-turbine-backlog-nears-70-gw-as-company-expands-manufacturing/827390/); Mitsubishi's June-quarter orders [deliver between 2028 and 2030](https://www.utilitydive.com/news/mitsubishi-gas-turbine-backlog-earnings/827761/). A heavy-duty frame is not a bridge. EPRI's Bobby Noble puts the small turbine at [18 to 36 months, against about five years for a large one](https://www.utilitydive.com/news/gas-turbine-supply-crunch-set-to-raise-prices-195-by-2027-woodmac/816904/), and notes that 70% of the turbine units ordered in 2025 were under 100 MW. The bridge turbine is an aeroderivative: two GE Vernova LM6000 PF+ Sprint units at [56.9 MW and 8,328 Btu/kWh LHV](https://www.gevernova.com/gas-power/products/gas-turbines/lm6000), or three LM2500XPRESS at [34.5 MW each](https://www.gevernova.com/gas-power/products/gas-turbines/lm2500), the package GE Vernova is shipping to Crusoe 29 at a time. The [$60 Million Deposit](https://open-factory.com/articles/slot-economy/) piece covers what it costs to hold a heavy-duty slot. This piece is about not needing one. ## What a Kilowatt Costs After the 2026 Repricing The eleven price points run from $600/kW for a heavy-duty turbine as equipment to $7,291/kW for a 10 MW fuel-cell plant installed, and the three we use for the bridge sit at $2,240, $2,670 and $3,003. Start with the turbine, because it is the one where the 2023 reference cost is no longer true. Sargent & Lundy's December 2023 case for EIA prices a [4 x 54 MW aeroderivative simple-cycle plant at $339.0 million, or $1,606/kW](https://www.eia.gov/analysis/studies/powerplants/capitalcost/pdf/capital_cost_AEO2025.pdf), of which $155.9 million, $739/kW, is the turbines and their SCR and CO catalysts. BNP Paribas now estimates GE Vernova gets [~$1,800/kW for an aeroderivative and ~$790/kW for a heavy-duty machine](https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-climbs-to-116-gw/826039/) (July 2026). Swap the equipment line and hold everything else at 2023 levels and the plant is ~$2,670/kW. **The aero turbine's 2026 price alone is higher than the 2023 cost of the whole plant it sits in.** We use $2,670 as the base, $1,606 as the low case and EPRI's [~$3,000/kW top of the range for turbine plants](https://www.utilitydive.com/news/5-year-waits-and-rising-costs-how-demand-is-redefining-the-gas-turbine-mar/813385/) as the high. Engines have no public data-center price. Wartsila's [412 MW order of 40 x 34SG engines for an Ohio hyperscaler](https://www.wartsila.com/usa/media/news/16-04-2026-wartsila-s-34sg-engine-makes-its-data-center-debut-with-new-412-mw-u-s-project-3740972), Caterpillar's [1.5 GW first phase of 636 G3520K gensets for Joule in Utah](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/) and INNIO's [1.25 GW framework with Rehlko](https://www.innio.com/en/news-media/press-releases/innio-and-rehlko-enter-into-framework-agreement-for-1-25-gw-of-gas-engine-capacity/) all come without a dollar figure. So the base is EIA's [$2,240/kW for a 21 MW internal combustion plant in 2022 dollars](https://www.eia.gov/outlooks/aeo/assumptions/pdf/elec_cost_perf.pdf), with a range of $1,800 to $2,800 that is ours. If a reader has a 2026 engine quote, the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) is where it belongs. The cell will print when it has eight. Fuel cells have the cleanest price in the set. AEP's fourth-quarter 2025 release describes a Wyoming generation project [backed by $2.65 billion in fuel cell purchase commitments](https://www.sec.gov/Archives/edgar/data/4904/000000490426000011/a4q20258kpressreleaseex991.htm), and Bloom's 10-K describes the same agreement as [up to 1 GW with an initial 100 MW order](https://www.sec.gov/Archives/edgar/data/1664703/000162828026006516/be-20251231.htm): ~$2,650/kW of equipment. Installation is a separate revenue line at Bloom; in 2025 it was $204.1 million against $1,531.3 million of product revenue, 13.3%. Installed is therefore ~$3,003/kW, and after the 30% ITC, ~$2,102/kW, the lowest net capital of the three. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes the gross number on the equipment line and the credit on a separate line, because the credit has conditions the equipment does not. ## Heat Rate and Fuel: Where Fuel Cells Earn Their Price At $3.78 gas the fuel line is $36/MWh for turbines, $31 for engines and $24 for fuel cells; at $6.00 it is $57, $50 and $39. Heat rates are on a higher heating value basis throughout, because that is how gas is billed. The aero plant runs at [9,447 Btu/kWh net in the Sargent & Lundy case](https://www.eia.gov/analysis/studies/powerplants/capitalcost/pdf/capital_cost_AEO2025.pdf), which includes SCR back-pressure and plant loads; GE Vernova's 8,328 LHV for the LM6000 PF+ is ~9,230 HHV before those losses. The engine plant is [8,295 Btu/kWh in EIA's table](https://www.eia.gov/outlooks/aeo/assumptions/pdf/elec_cost_perf.pdf); an engine-plant developer told NGI new machines run [as low as 7,200](https://www.naturalgasintel.com/news/no-turbine-data-centers-find-faster-path-with-natural-gas-reciprocating-engines/), which we carry as the low case. Fuel cells are 6,469 in the same EIA table, 52.7% HHV. Bloom's marketing claim of 60 to 65% fuel-to-electricity, [as relayed by the CPS Energy presentation](https://avanzaenergy.substack.com/p/data-centers-are-killing-the-grid), is an LHV figure that converts to ~6,300 HHV; we run it as the low case only. The monthly average went from $4.26 in December 2025 to [$7.72 in January 2026 and back to $2.78 by August](https://fred.stlouisfed.org/series/MHHNGSP); the September STEO puts the [2027 annual average at $3.28/MMBtu](https://www.eia.gov/outlooks/steo/pdf/steo_full.pdf), down from $3.43 for 2026. Our base delivered price is that $3.28 plus $0.50 for firm transport, and the $0.50 is an assumption, not a measurement. The measured number is the state basis. EIA publishes monthly citygate and industrial prices by state; over July 2025 to June 2026 the citygate ran [$0.28/MMBtu above Henry Hub in Texas](https://www.eia.gov/dnav/ng/hist/n3050tx3m.htm), $2.26 above in Virginia, $2.37 above in Ohio (six months published), $0.59 below in Arizona and $1.26 above in Georgia. Those five numbers, not the $0.50, drive the paid section. A 100 MW plant at 85% and 8,295 Btu/kWh burns ~6.2 million MMBtu a year, so each $1 of basis is $6.2 million a year on engines and $4.8 million on fuel cells. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool takes the same heat-rate-times-price arithmetic for one machine. ## Permitting: 1,136 Tons of NOx Versus Zero At the federal floor, 100 MW of lean-burn engines emits 1,136 short tons of NOx a year and 100 MW of aero turbines 447; with SCR the engines fall to ~112 and the turbines to 26, against thresholds of 250, 100 and 50 tons depending on where the plant stands. The floors are federal new-source performance standards. A non-emergency lean-burn gas engine of 500 HP or more built after July 1, 2010 must meet [1.0 g/HP-hr of NOx, or 82 ppmvd at 15% oxygen](https://www.ecfr.gov/current/title-40/part-60/subpart-JJJJ) under Subpart JJJJ; at 0.746 kW per horsepower and a 97% generator, that is ~3.05 lb/MWh. A new gas turbine between 50 and 850 MMBtu/h must meet [25 ppm, which the rule itself equates to 1.2 lb/MWh](https://www.ecfr.gov/current/title-40/part-60/subpart-KKKK), and 15 ppm or 0.43 lb/MWh above 850 MMBtu/h. Sargent & Lundy's aero case carries SCR and predicts [0.0075 lb/MMBtu](https://www.eia.gov/analysis/studies/powerplants/capitalcost/pdf/capital_cost_AEO2025.pdf), 0.071 lb/MWh at its heat rate. Bloom's 10-K says its Energy Server has a ["negligible impact on air quality"](https://www.sec.gov/Archives/edgar/data/1664703/000162828026006516/be-20251231.htm), that most states require permits only for its larger installations, and that New York, New Jersey and North Carolina have specific air-permit exemptions for fuel cells. It publishes no lb/MWh figure in the filing, so the chart shows zero and the notes say why. The thresholds decide the route. In an attainment area a new plant that is not one of the 28 listed source categories becomes a PSD major source at [250 tons per year of any regulated pollutant](https://www.ecfr.gov/current/title-40/section-52.21); a simple-cycle turbine or engine plant is not a "fossil fuel-fired steam electric plant," so 250 applies, not 100. In an ozone nonattainment area the [major-source threshold drops to 100 tons per year, and to 50 in a serious area](https://www.ecfr.gov/current/title-40/section-51.165). **A 100 MW engine plant with 90% SCR is still a major source in Phoenix, Northern Virginia, Dallas and Houston, and a 100 MW aero plant with SCR is a minor source in all four.** Major means LAER or BACT, offsets in nonattainment areas, public notice and a review that Sargent & Lundy budgets inside an 18-month development phase. Minor means a permit by rule or a standard permit. Texas has issued minor permits to [at least 38 data centers since 2024, covering more than 2,100 diesel generators](https://www.latitudemedia.com/news/ai-data-centers-are-getting-built-thanks-to-a-permit-common-for-dry-cleaners), including Stargate's 10 turbines and 62 gensets, and one Vantage site permitted at 99.8 tons against a 100-ton line. The third route is the one xAI took in Southaven, Mississippi: [27 portable turbines totaling 495 MW without an air permit](https://www.utilitydive.com/news/doj-intervenes-xai-data-center-gas-turbine-lawsuit/823267/) under the state's reading of the mobile-source provisions, now in federal court with the Department of Justice intervening on xAI's side. We do not model that route. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) follows the case. ## Five-Year Cost and Resale Over 3.72 million MWh the turbines cost $417 million, the engines $388 million and the fuel cells $357 million; capital net of resale and credits is $46, $42 and $48 per MWh, fuel $36, $31 and $24, and O&M $7, $12 and $5. The method is deliberately plain. Installed cost times 100,000 kW, less the ITC where it applies, is the capital. We charge interest on it at [Moody's Baa yield of 6.32% (August 2026)](https://fred.stlouisfed.org/series/BAA) for five years and give back a residual at year five. Fuel is MWh times heat rate times $3.78. Fixed O&M is Sargent & Lundy's [$9.56/kW-year for the aero plant](https://www.eia.gov/analysis/studies/powerplants/capitalcost/pdf/capital_cost_AEO2025.pdf) and EIA's $39.57 for engines and $34.65 for fuel cells; variable O&M is $5.70, $6.40 and $0.66 per MWh, the turbine figure including hours-based major maintenance. Bloom sells maintenance agreements of [5 to 20 years](https://www.sec.gov/Archives/edgar/data/1664703/000162828026006516/be-20251231.htm) separately, and its service gross margin was 13.3% in Q1 2026, so the $34.65 is a floor rather than a quote. Every one of those inputs, with low and high cases and a source URL, is in the [assumptions table](data/assumptions.csv). Resale is the assumption we are least sure of and the one we flag loudest. Dealers report that low-hour used gensets now sell at [60 to 75% of new when new delivery is a year out](https://csdieselgenerators.com/the-ai-data-center-boom-and-generator-shortage-lead-times-prices/) (August 2026), but a bridge engine runs ~37,000 hours in five years, so we take 30%. Aero turbines have a real secondary market: PROENERGY's [13 x PE6000 packages for Crusoe](https://www.proenergyservices.com/2026/04/proenergy-to-deliver-major-generating-equipment-to-power-crusoe-ai-factories/) are rebuilt LM6000-class machines at 50 MW each. We take 35%. Fuel cells have no published secondary market and their stacks are consumed; we take 10%. Moving all three residuals to zero adds $25/MWh to turbines, $18 to engines and $8 to fuel cells and leaves the order unchanged. Two lines are left out on purpose. We do not add a gas interconnection or a substation, because all three technologies need both and Sargent & Lundy prices them at [$3.95 million and $2.4 million](https://www.eia.gov/analysis/studies/powerplants/capitalcost/pdf/capital_cost_AEO2025.pdf) for a 211 MW plant with a half-mile lateral, about $30/kW, which does not move the ranking. We do not add the utility's minimum bill on the bridged load either, because it depends on the tariff. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) includes it by market, and in AEP Ohio it is 85% of contract demand for up to 12 years. From a September 2026 order, fuel cells run 23 months before 2030 and deliver ~1,430 GWh, engines 13 months and ~810 GWh, turbines 10 months and ~620 GWh; at the Virginia industrial price of 9.31 cents the 810 GWh gap between fuel cells and turbines is ~$75 million of electricity. That is the number the $/MWh table hides. A bridge exists because the grid is late, and a bridge that arrives 13 months after another bridge is 13 months of a datacenter or a plant not running. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tool puts the PO date and the gear class in and returns the first-power month, and the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool prices the utility side of the same wait. ## Sensitivity: Gas, Utilization and the Two Flips At $2 gas the three options cost $95, $89 and $85 per MWh; at $8 they cost $152, $139 and $123, so the fuel-cell advantage widens from $10 to $29 as gas rises. At 40% utilization the costs are $191, $179 and $176 per MWh; at 95% they are $105, $97 and $89, and the fuel-cell margin over engines narrows to $3 at the low end because its net capital per kW is the smallest of the three and utilization punishes capital. Utilization is the input the buyer controls least. A datacenter campus ramps, so a 100 MW bridge ordered for a 2028 first phase runs at half load until the second hall fills, and every month at 50% costs ~$140/MWh instead of $96 to $112. Engines and fuel cells are modular, forty 2.5 MW gensets or a field of power modules, so the second half of the order can ship a year after the first; Oracle's 1.2 GW is being [deployed through 2027 in tranches](https://www.bloomenergy.com/news/bloom-energy-and-oracle-expand-strategic-partnership-to-deploy-up-to-2-8-gw-to-accelerate-ai-infrastructure-build-out/) for the same reason. Two aero turbines cannot be bought by halves without giving up the second machine's slot. The right comparison for a phased site is therefore the 60% row for turbines against the 85% row for the other two, which widens the gap by another $30/MWh. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool runs the phased case unit by unit. The ranking flips in two cases and both are about capital, not gas. First, the credit. Section 48E after the July 2025 budget act gives fuel cells [a 30% ITC without regard to emissions, with no prevailing wage or apprenticeship requirement and no adders, for projects beginning construction after 2025](https://www.lw.com/en/insights/one-big-beautiful-bill-new-law-disrupts-clean-energy-investment), phasing down only for construction starting [in 2034](https://www.kirkland.com/publications/kirkland-alert/2025/08/one-big-beautiful-bill-act-brings-big-changes-to-green-energy-tax-credits). Without it, fuel cells cost $128/MWh and engines win by $24. The credit is transferable, so a developer without tax appetite sells it. The FEOC supply-chain rules apply to projects beginning construction after 2025 and Bloom manufactures in California and Delaware. Our [30% Fuel-Cell ITC](https://open-factory.com/articles/fuel-cell-itc-last-subsidy/) piece walks through the eligibility test. Second, the turbine price. At Sargent & Lundy's 2023 installed cost of $1,606/kW the aero plant is $84/MWh and wins outright. At BNP's 2026 equipment price it is $112 and loses to both. A buyer holding a 2024 aero quote, or a rebuilt PE6000-class package at a rebuilt price, is in the first world. A buyer calling GE Vernova today is in the second. The [Quote Check](https://open-factory.com/tools/quote-check/) tool prints the band a turbine or engine quote should fall in for its date, and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) will add public turbine and engine awards as they post. EIA's July 2026 planned-generator file holds 487 natural-gas units and 68.1 GW, of which 5.6 GW of simple-cycle turbines and 0.34 GW of engines are under construction or complete, and just 1.0 GW of engines in any status, against more than 10 GW of announced US datacenter engine orders at Wartsila, INNIO and Caterpillar alone. That gap is the behind-the-meter market not yet reporting. The EIA-860M survey catches utility and merchant generators. The engine plants in Utah, West Virginia, Ohio and Texas will appear as they file, or not at all if they never export. Our [Behind the Meter, In Front of the Queue](https://open-factory.com/articles/behind-meter-90gw/) forecast tracks the announced GW against the surveyed GW each quarter, and the [Datacenter Electrical Roadmap 2026 to 2032](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) includes the resulting demand for switchgear and transformers, since every one of these plants needs a [generator step-up transformer at 144 weeks](https://open-factory.com/articles/transformer-160-weeks/). ## What to Do Monday Pick the technology by the two things this model cannot average away: the credit and the air permit. If the project entity can use or sell a 30% ITC and the site is in a moderate or serious ozone area, the fuel cell is the cheapest bridge and the fastest, and the PO goes to Bloom or an AEP-style utility channel this month, because Fremont's 2 GW comes online at the end of 2026 and Oracle has already contracted 1.2 GW of it. If the credit is not usable, the engine is the cheapest bridge, and the order goes in now for a late-2027 or 2028 slot with a written SCR guarantee and a permit strategy that keeps the plant under 100 or 250 tons. The turbine is the right answer only when someone is offering 2023 equipment, rebuilt or otherwise, or when footprint rules out forty engines. Write the delivery month, the heat rate at site conditions and the emissions guarantee into the purchase order, with [liquidated damages that bite](https://open-factory.com/articles/liquidated-damages/), and ask for the resale-support clause in the same document. The [RFQ for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) template has the language. Then check the state you are in, because the basis and the permit thresholds change the answer in three of the five markets below. Behind the paywall: the technology to buy in Texas, Virginia, Ohio, Arizona and Georgia given each state's citygate basis, ozone classification and grid wait, where **Ohio gas costs $2.37/MMBtu more than Henry Hub** and adds $12 to $18/MWh to every bridge; and the price book of 22 named deals and reference costs from GE Vernova, Wartsila, Caterpillar, INNIO, PROENERGY, AEP and Bloom, in the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # Nox Metals: The Detroit Startup Cutting Aluminum Plate in 36 Hours URL: https://open-factory.com/articles/nox-metals/ Date: 2026-03-18 Procurement / Deep dive $11.5M seed led by Hyperion, Y Combinator and Palmer Luckey, a 30,000 sq ft Ferndale factory, 6061 and 7075 plate to 24.5 in, instant quotes, the 93-cent Midwest premium, how to place a first order Nox Metals quotes a saw-cut aluminum block in seconds and, by its lead investor's count, delivers it in an average of 36 hours, in a trade where the quote alone has taken three to five days. The company is fourteen months old. It launched out of Y Combinator's Summer 2025 batch, closed an [$11.5 million seed on 14 June 2026](https://noxmetals.co/blog/nox-metals-seed-round) led by Hyperion with Palmer Luckey, Jared Friedman, RoboStrategy, Operator Collective, DTX and Alumni Ventures alongside Y Combinator, and is moving into a 30,000 square foot Second World War factory in Ferndale, Michigan, north of Detroit. Its founder, Zane Hengsperger, is 26, grew up on the floor of his father's shop, and describes what he has built as "a software company wrapped around a saw". This piece covers the turnaround Nox Metals publishes against the service centers it competes with, the product range and the machine behind it, the $14 billion incumbent's market it is entering, the tariff-driven metal price that no supplier controls, the twelve months from Demo Day to hundreds of customers, and behind the wall, the landed-cost worksheet and certificate checklist for a first order. Six suppliers, two ways of quoting. Nox Metals, Cut2Size Metals and Industrial Metal Service quote online. Reliance, with about 310 locations, delivers about 40% of orders within 24 hours, Alro states a next-day capability, and the typical service center as Nox's backers describe it takes 3 to 5 days to quote and 3 to 5 days to deliver. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) has tracked switchgear and transformer lead times since launch. Cut plate is the first raw material it lists, because Nox made the number public. ## Nox Metals Started on a Shop Floor in Windsor Zane Hengsperger is from Windsor, Ontario, across the river from Detroit, and [grew up in his father's shop](https://enjoyer.com/detroits-youngest-new-industrialist/), where metal arrived late and was priced by phone. He graduated from the University of Michigan in 2022 and, before Nox, [ran a 30-person fabrication facility](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals). The idea came from the buying side of that job. On the MakingChips podcast in February 2026 he put it as a diagnosis: [the bottleneck in American manufacturing is not machining, it is metal supply](https://makingchips.com/reindustrializing-america-how-zane-hengsperger-is-reinventing-the-metal-supply-chain-510/). Machine shops that are, in his words to SendCutSend's podcast, ["too small to get white glove treatment from traditional service centers but too important to keep waiting"](https://sendcutsend.com/podcast/video/zane-hengsperger/) spend days getting a price on a block of 6061 they could cut in an hour. He wrote about the idea online, and [Y Combinator funded the company within 24 hours](https://makingchips.com/reindustrializing-america-how-zane-hengsperger-is-reinventing-the-metal-supply-chain-510/) of reaching out, with a flight to San Francisco. Nox launched in the [Summer 2025 batch](https://www.ycombinator.com/companies/nox-metals) in August 2025 as "fully automated factories that cut and ship metal fast". The founding team [sold during the day, wrote software, and cut metal in the evening](https://sendcutsend.com/podcast/video/zane-hengsperger/). Hengsperger chose Detroit for the reason a metals company should: the machinists, riggers and saw operators are still there, and so are the buildings. By the end of 2025 he was on [Forbes' 30 Under 30 list for Manufacturing and Industry](https://www.forbes.com/30-under-30/2026/manufacturing-industry/). **The stated mission is to supply every factory in America in under 24 hours at a fair, transparent price, and by March 2026 the measured average was 36 hours.** That figure comes from [Behind Genius Ventures](https://behindgenius.substack.com/p/why-we-invested-in-nox-metals), which led the pre-seed and wrote up the company on 2 March 2026 against what it calls an industry standard of 4 to 5 days. It is an investor's number, not yet a published monthly statistic, and it is the number a buyer should hold the company to, in the way the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) holds Schneider to its published lead-time guide. Few suppliers of any age have said what their average is. ## A Software Company Wrapped Around a Saw The saw is a [Kasto BBS](https://behindgenius.substack.com/p/why-we-invested-in-nox-metals) block band saw that takes a plate of up to 50,000 pounds. The software around it has names. WAYNE is the nesting and optimization engine, which the company and its investors say [evaluates "thousands to millions of possible cut layouts per plate in 30 seconds"](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals) and batches several customers' orders onto one plate so the remnant is smaller. GONDOR, later described as NOX NEST, is the customer portal: instant algorithmic pricing on an uploaded cut list, material lists parsed by the software, automated work orders, order tracking, and [self-service delivery of material certificates](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals). The plant is [ISO 9001:2015 certified](https://noxmetals.co/). Plate in 5000 series, 6061, 7050 and 7075 from 0.25 to 24.5 inches thick, in T651, T7351 and T7451 tempers, saw-cut to length and width and faced flat on request. Round bar in 6061 and 7075 from 0.5 to 17 inches in diameter, flat bar from 0.25 to 4 inches, cut to length with square ends. Inconel and stainless [on request](https://dealroom.co/companies/nox-metals/). That is the range a job shop machining aerospace brackets, tooling plates, fixtures or medical housings actually buys, and it is the range Hengsperger narrowed to on purpose after [early struggles getting access to domestic mills](https://makingchips.com/reindustrializing-america-how-zane-hengsperger-is-reinventing-the-metal-supply-chain-510/) taught him to go deep on aluminum plate first. The team is 35 people as of the [Y Combinator profile](https://www.ycombinator.com/companies/nox-metals), built deliberately as [industry veterans beside software talent](https://makingchips.com/reindustrializing-america-how-zane-hengsperger-is-reinventing-the-metal-supply-chain-510/). The chief operating officer, Vince Mazzei, spent [21 years at Alro Steel](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals), the $2.85 billion Jackson, Michigan distributor. The chief revenue officer, Kelly Gibb, has 15 years selling to the same machine shops. The founding engineer, Inan Syed, built WAYNE, and the company drew more than 500 engineering applicants in its first year. The mix matters because the trade Nox is entering has, by Operator Collective's count, [fewer than 40 software engineers across the entire US sector](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals). ## The Service Center Market Nox Metals Is Entering Metal service centers are the layer between the mill and the machine shop. They buy plate and bar by the truckload, hold it, cut it, and sell it in pieces. In the United States the layer is large and old. Reliance, the biggest, says in its [10-K for 2025](https://www.sec.gov/Archives/edgar/data/861884/000110465926020651/rs-20251231x10k.htm) that it sold 6.4 million tons from about 310 locations in 41 states, about 17% of everything US service centers sold, to more than 125,000 customers on an average order of $3,120, and that it delivered about 40% of orders within 24 hours. Ryerson, founded in Chicago in 1842, [shipped 1.95 million tons at $2,348 a ton](https://ir.ryerson.com/news/news-details/2026/Ryerson-Reports-Fourth-Quarter-and-Full-Year-2025-Results/default.aspx) in 2025. Alro Steel, privately held in Jackson, Michigan, had [2024 revenue of $2.85 billion across 74 locations and 6.3 million square feet](https://ftp.metalcenternews.com/downloads/2509_Top50-Companies.pdf) in Metal Center News' Top 50. Reliance at $14.29 billion and 310 locations, Ryerson at 1.95 million tons, Alro at $2.85 billion and 74 locations, and then the rest of the trade: Behind Genius puts the [average service center at one location, 24 employees and about $47 million a year](https://behindgenius.substack.com/p/why-we-invested-in-nox-metals), and the sector at more than $200 billion. Nox is one building, 30,000 square feet and 35 people. The comparison is the point rather than the problem. Most of the companies Nox sells to are also one building and 24 people, and they have been buying from a supplier shaped exactly like themselves, by phone, for eighty years. **The opening Nox Metals found is that a service center's product is not metal, it is the four days between the request and the truck, and software collapses four days more cheaply than a fifth saw does.** Reliance's 40% within 24 hours is achieved with 310 stocking points and $14 billion of turnover. Nox's 36-hour average is achieved with one plant and a nesting engine, which is a different cost structure for the same promise. The nearest analogue is [SendCutSend](https://open-factory.com/articles/sendcutsend/), which published a 2 to 4 day processing time for laser-cut sheet in 2018, shipped 30 million parts, and raised $110 million at a $1 billion valuation this May. Hengsperger has been on SendCutSend's podcast, and the two companies sell to the same shops: one the sheet, the other the block. ## Who Buys From Nox Metals The customer list the company describes is the American job shop and the OEMs above it. Daily Detroit quoted Hengsperger on where the first year's metal went: ["Our metal has gone to space. It has protected our troops. It is in your car and in the machine that scanned your chest."](https://www.dailydetroit.com/detroit-startup-raises-11-5-in-funding/) Aerospace, defense, automotive tooling and medical imaging are the four sectors named in the seed announcement, and SpaceX is the one customer named in public. The rest are the [16,619 machine shop businesses](https://www.ibisworld.com/united-states/number-of-businesses/machine-shop-services/645/) IBISWorld counts in 2026 and the [265,030 people](https://www.bls.gov/oes/2023/may/naics5_332710.htm) BLS counts working in them, most of them in shops with one building and a handful of spindles. For that shop the block is the start of every job. A [$130,000 Haas VF-2](https://open-factory.com/articles/haas-vf2-true-cost/) machining a fixture plate earns nothing while the 6061 is on a truck, and the [machine-tool record](https://open-factory.com/articles/machine-tool-record-fewer-units/) we wrote in May showed the average new machine getting dearer as unit counts fell, which makes idle spindle-hours more expensive every year. Reliance's own average order is $3,120, which is a few blocks and a bar, and the arithmetic of a service center is thousands of those a day. The [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has no raw-material band yet, which is why the readers' log matters. Nox's bet is that the shop placing the $3,120 order would rather place it on a web page at nine in the evening than on a phone at nine the next morning, and that the saw can be scheduled by software the moment the order lands. The published range fits that buyer. Plate to 24.5 inches covers tooling blocks, fixture bases and the thick aerospace hog-outs that are most of a 7075 buyer's spend, and bar to 17 inches covers the turned parts. What the range does not yet include, steel and stainless plate as stocked items, is the part of the order the incumbent keeps for now, and the paid section below says how to split it. A shop with a [used machine bought at auction](https://open-factory.com/articles/used-cnc-auction-index/) and a two-week backlog gains the most from a 36-hour supplier, because it has the least float in its schedule, and the [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) turns a supplier's promise into the calendar that shows it. ## What the Metal Costs Before Nox Touches It Three layers of an aluminum block's price are set before any service center cuts it, and a buyer comparing Nox to an incumbent should hold them constant. The first is the LME cash price. The second is the US Midwest premium, the delivered-to-Midwest surcharge over LME that every domestic supplier pays. It went from [the high 50s in cents per pound in early June 2025](https://www.aluminummarketupdate.crugroup.com/pulse/midwest-premiums-how-high-is-high/) to [90.30 cents on 15 December 2025](https://www.spglobal.com/energy/en/news-research/latest-news/metals/121525-aluminum-us-midwest-premium-reaches-90-centslb-for-the-first-time), about 93 cents by 6 January 2026, and [$2,182 a tonne, 99 cents a pound, by 25 February 2026](https://investor.wedbush.com/wedbush/article/marketminute-2026-2-25-aluminum-midwest-premium-hits-record-2182-inside-the-2026-price-surge-and-the-bessent-pivot) as reported by Wedbush. The third is the reason for the second: Section 232 aluminum duty at 50% since 4 June 2025, extended by [Proclamation 11021 on 2 April 2026](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states) to the full customs value of derivative articles. The [September 2026 tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) has the rate table, and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) applies it to a bill of materials. High 50s in June 2025, 90.3 cents in December, 93 in January, 99 in February. A 12 x 12 x 2 inch block of 6061 weighs about 28 pounds, so the premium alone moved its raw-metal cost by about $11 in nine months before any mill or service-center margin. ISM's August 2026 survey listed aluminum as [up in price for the 33rd consecutive month](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), the longest run in the table we read in [the 171-day capex piece](https://open-factory.com/articles/171-day-capex-lead-time/). The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) shows the same mechanism on the copper side each week. What Nox controls sits on top of that tape, and it is two things. The first is yield. Its investors say [traditional service centers lose 30 to 40% of a plate to scrap](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals), and the whole purpose of nesting several customers' blocks onto one plate is to sell more of the plate as product and less as remnant. The company's site shows its yield figure live. The second is the days, and days are money to a shop whose [$130,000 Haas VF-2](https://open-factory.com/articles/haas-vf2-true-cost/) is waiting for stock. A buyer who wants to see the two effects separately should put the same cut list through Nox and the incumbent in the same hour and log delivered dollars per pound and calendar days to the dock in the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/), which takes raw-material orders as well as machines. ## From Demo Day to Hundreds of Factories in Twelve Months The pre-seed closed on 23 September 2025 at Demo Day. Its lead, Behind Genius, [reports $5.3 million](https://behindgenius.substack.com/p/why-we-invested-in-nox-metals) from 8090 Industrials, Y Combinator, E1, The Council, Anorak and itself. [TBPN Digest](https://www.tbpndigest.com/company/nox-metals) and Dealroom report $4.6 million for the same round, so we print both. The seed nine months later was [$11.5 million led by Hyperion](https://noxmetals.co/blog/nox-metals-seed-round), the New York firm, with Palmer Luckey of Anduril, Y Combinator's Jared Friedman, RoboStrategy, Operator Collective, Dallas's DTX Ventures and Alumni Ventures. The capital is for technology, machinery and production capacity at the Detroit plant. $5.3 million in September 2025, $11.5 million in June 2026, $16.8 million in total, or $16.1 million on the lower pre-seed figure. What the money bought in between is the more useful record. By 2 March 2026 the company had [shipped more than 100 truckloads](https://behindgenius.substack.com/p/why-we-invested-in-nox-metals) from its first Detroit floor, a 1920s complex in Mexicantown with a [motorized saw platform, a chain conveyor for the blocks, coolant, and a chip-recycling loop](https://enjoyer.com/detroits-youngest-new-industrialist/). By 14 June it was [shipping to "hundreds of American factories"](https://noxmetals.co/blog/nox-metals-seed-round) in aerospace, defense, automotive and medical equipment. SpaceX found the company [through social media](https://www.operatorcollective.com/blog-posts/the-metal-shop-that-thinks-like-a-software-company-meet-nox-metals) and became a customer, and the MakingChips episode tells the story of an order for a [block of more than 18,000 pounds](https://makingchips.com/reindustrializing-america-how-zane-hengsperger-is-reinventing-the-metal-supply-chain-510/). Eight dated steps from the August 2025 launch to the planned Los Angeles plant in early 2027, with the 30,000 square foot Ferndale building at 1350 Jarvis Street coming online in summer 2026. Hengsperger's own summary, on X in October 2025, was that ["everyone else is accepting metal prices and speed"](https://x.com/zanehengsperger/status/1974979119306727927). The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) lists the Los Angeles opening as a date to watch, because a second plant is when a one-building supplier becomes a network, and the [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) is where a second published turnaround would land. ## How to Place a First Order With Nox Metals Start with the block you buy most. Take the 6061-T651 plate order that comes up every month, export the cut list, and put it through the [instant quote at noxmetals.co](https://noxmetals.co/) and through the incumbent's rep in the same hour. Record three numbers: minutes to a price, delivered dollars per pound, and calendar days from order to dock. The alloys, tempers and size ranges in the product table above are the ones to test, and 7075-T651 plate, the alloy that gets scarce first in a defense cycle, is the one to test second. Ask for the certificate with the order rather than after it. Nox's portal delivers mill certificates self-service, so request the ASTM B209 cert with heat number on the first shipment and confirm it matches the plate. If any of the work touches DFARS 252.225-7009 specialty-metals clauses, or the 10% Section 232 tier for goods made from US-melted metal, ask where the plate was melted. The [RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) has the certificate language, and [Quote Check](https://open-factory.com/tools/quote-check/) reads any supplier's quote against a published schedule. Then decide what to move. A shop that buys $3,120 of plate a week, Reliance's average order, can move that order to Nox on the first week the delivered price and the 36 hours both hold, keep the incumbent account open, and let the log in the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) decide the split over a quarter. That is the same test we set for every new supplier in this market, and Nox is the first cut-plate supplier young enough and public enough to be tested in a spreadsheet. Behind the paywall: the landed-cost worksheet for a **28-pound** block from LME to dock, the six certificate and clause items for a seed-stage supplier, when to split the order rather than switch it, and what the Los Angeles plant changes for West Coast shops. The turnaround rows for all six suppliers sit in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/), and the delivered prices readers log go to the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/). --- # The 2 MW Plant Service Upgrade Nobody Budgets Right URL: https://open-factory.com/articles/2mw-plant-service-upgrade/ Date: 2026-03-15 Electrical / How-to Utility-owned pad-mount at 52 to 104 weeks, Ohio's 40% CIAC vs Georgia's revenue allowance, the 30-day AHJ clock, arc-flash study at $6k to $70k, the 20-week sequencing gain, Large-Load Cost Table A 400-person plant adding 2 MW of service pays for five things the switchgear quote never mentions, and together they run $400k to $1.2M with one month of bridge power. The calendar is worse than the money: run the project the way most plants run it, utility estimate first and gear order second, and the crane lands in week 74. The [Open Factory Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) now lists the sub-5 MW tariff sheets that set those numbers, and this piece walks one Ohio plant through them: the utility's 45-day estimate clock and 357-day build clock, the transformer the utility owns and cannot get, four CIAC formulas from four real tariffs, the 30-day permit rule, the arc-flash study, and the cost stack. The paid section is the sequencing that pulls energization forward by 20 weeks and the diesel math for the months you cannot pull forward. Read it back: the utility's firm estimate arrives at week 9, the plant decides inside the 90-day validity window at week 22, and the two 50-plus-week items, the LV switchgear and the utility-owned pad-mount, run from week 22 to weeks 72 and 74. ## Seventy-Four Weeks From Decision to Crane Every step on that chart has a published source, and none of them is the plant's contractor. **The utility owes an Ohio customer a binding cost estimate within 45 calendar days of a complete request, and that estimate is valid for 90 days**, under [OAC 4901:1-9-07(C)](https://codes.ohio.gov/ohio-administrative-code/rule-4901:1-9-07), the line-extension rule every Ohio electric utility mirrors in its tariff. AEP Ohio's [Terms and Conditions of Service, Sheet 103](https://www.aepohio.com/lib/docs/ratesandtariffs/Ohio/September_2026_AEP_Ohio_Tariff_Book.pdf) list what "complete" means: switch size, delivery voltage, total estimated load, connected-load list, operating characteristics, a site survey and first-floor elevations. Show up without the load list and the 45 days have not started. What the rule does not give you is a construction date. Ohio publishes an estimate clock and no build clock, which is why the chart borrows the only published one in the country. California's [CPUC Decision D.24-09-020](https://www.cpuc.ca.gov/-/media/cpuc-website/divisions/energy-division/documents/transportation-electrification/energization/fact-sheet-energization-091224.pdf) (September 2024) set an average target of 182 days and a maximum of 357 days for a Rule 15 distribution line extension, measured from the customer's request. Our [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) uses the 357-day figure as the utility's outer bound for a 2 MW job that needs no circuit upgrade. The model does not assume Ohio's utilities are faster than PG&E's. The switchgear and the transformer sit on top of that. Schneider's [Lead Times Guide](https://www.se.com/us/en/download/document/Monthly_Lead/) publishes 50 weeks for drawout LV switchgear (LVDO frames A through F) and 18 to 34 weeks for switchboards (A4 through F), and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes both as published lead times, OEM and month labelled, not as deliveries. Order at week 22 and the lineup ships at week 72. That is the 74-week schedule, and nobody chose it. It fell out of doing things in the order the paperwork arrives. ## The Utility Answers in 45 Days and Builds in 357 The two clocks matter because the utility's answer is the trigger most plants wait for. In Ohio the answer has three parts, each dated in the rule: a nonbinding good-faith estimate within [ten business days](https://codes.ohio.gov/ohio-administrative-code/rule-4901:1-9-07), a binding firm estimate within 45 calendar days, and a 90-day validity period after which the number can move with rights-of-way and "conditions beyond the reasonable control of the electric utility." Land clearance, trenching and backfill on the customer's own property are the customer's cost regardless. The [Interconnection Cost tool](https://open-factory.com/tools/interconnection-cost/) now runs down to 1 MW so a plant can put those three dates on its own calendar before calling the utility. The bars read 357 days for a Rule 15 line extension, 335 for a Rule 16 service extension, 684 days where the job needs a new or upgraded circuit, 1,021 for a substation upgrade and 3,242 for a new substation. **A 2 MW plant that lands on a circuit with no headroom is a 684-day job by California's own maximum, not a 357-day one**, and the utility's screening answer is the only place you learn which one you are. Ohio has no equivalent numbers, which is a finding in itself and the reason the [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) logs Ohio as "estimate clock only." The CPUC figures are targets set from five years of utility history, enforced by biannual reporting and a customer delay form. The [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/) table treats them the same way, as the ceiling a utility will be asked to explain, not a delivery date. ## The Transformer You Do Not Own Under AEP Ohio's tariff the company "will design, construct, own, operate and maintain the line extension and all other equipment installed to serve the customer's load up to the point of service," per [Sheet 103-5](https://www.aepohio.com/lib/docs/ratesandtariffs/Ohio/September_2026_AEP_Ohio_Tariff_Book.pdf). PG&E's [Rule 16](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_RULES_16.pdf) says the same thing differently: service transformers are PG&E-owned service facilities, and the applicant furnishes the pad, the substructures and the protective structures. The pad-mount that steps 12.47 kV down to 480 V for a 2 MW plant is therefore the utility's purchase, on the utility's queue, and the customer's schedule hangs on it. That queue is the one nobody budgets. The industry coalition NRECA joined in January 2024 told the Senate that a distribution transformer took "an average of one year to order and receive" before the pandemic and that the schedule "has grown to [more than two years](https://www.electric.coop/nreca-joins-coalition-urging-senators-to-fund-distribution-transformers)." A T&D World survey of the utility supply chain quoted Hawaiian Electric's lead time for single-phase pad-mounts jumping from 16 weeks to 142 and reported that "most OEMs are quoting [52 weeks plus for padmount transformers](https://www.tdworld.com/utility-business/article/21243198/transformative-times-update-on-the-us-transformer-supply-chain)." Wood Mackenzie's August 2025 read put the 2025 supply deficit at [10% for distribution transformers and 30% for power transformers](https://www.woodmac.com/press-releases/power-transformers-and-distribution-transformers-will-face-supply-deficits-of-30-and-10-in-2025/), with imports covering half of distribution supply. **The Monitor's working range for a utility-procured 2,500 kVA pad-mount is 52 to 104 weeks from the utility's PO**, and the utility does not place that PO until the CIAC is paid. The transformer index stands at 181 and the switchgear index at 171 in February 2026 against 100 in December 2019, per [BLS series WPU117409](https://fred.stlouisfed.org/series/WPU117409) and WPU1175. That price path is inside the utility's cost estimate too. Whatever share of the job the tariff assigns to you is a share of a 2026 transformer, not a 2019 one, and the [Transformer Lead Times Hit 160 Weeks](https://open-factory.com/articles/transformer-160-weeks/) piece covers why the mills upstream are not going to relieve it before 2028. Both Ohio tariffs give the customer an escape hatch. AEP Ohio Sheet 103-7 paragraph 3.c: "If a substation is required as part of the line extension project to a customer, the customer shall be given the option of building, owning, and maintaining such substation." The [Transformer Sizing and Cost tool](https://open-factory.com/tools/transformer-sizing-cost/) prices that option; the paid section below shows when it is worth taking. ## CIAC: Four Tariffs, Four Answers We pulled the line-extension sheets from four utilities and ran the same 2,000 kW service through each. The formulas do not resemble each other, which is the point of putting them side by side in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). | Utility and sheet | Customer pays | Threshold for a 2,000 kW plant | Tax and refund terms | |---|---|---|---| | AEP Ohio, P.U.C.O. No. 22 Sheet 103-7; OAC 4901:1-9-07(D)(3) | 40% of the total line-extension cost, plus 100% of premium services (underground, oversizing), before construction | None; the 40% applies from dollar one | Premium-service CIAC grossed up for taxes; pro-rata refund if a new customer uses the facilities within 50 months | | Georgia Power, Rules and Regulations Sec. C.2 (book effective Jan 2023) | Net job cost above 3.5x estimated annual base-rate revenue (overhead) or 2.5x (underground) | ~$2.0M underground at a $792k revenue proxy (10,512 MWh at 7.53 c/kWh); base rates alone give a lower figure | Refund if another customer connects within five years; utility not required to start until service contracts are signed | | PG&E, Rule 15 Sec. C.2 and D; Rule 2 Sec. I; Prelim. Statement J | Total estimated installed cost above an allowance equal to net revenue divided by the cost-of-service factor (1.23%/month, 14.76%/year) | ~$6.0M at B-20 secondary distribution demand rates (~$882k/yr net revenue), before PG&E's non-residential multiplier | 24% ITCC on every contribution; refundable amounts refunded over 10 years; applicant builds pad and substructures | | ComEd, Rider DE Sheets 269 to 272.3 | Deposit equal to EXT COST minus the greater of the Standard Extension Cost (5,000 ft of three-phase overhead with transformer) or Five Year Expected Delivery Revenue, if the customer passes a credit screen | ~$1.38M at the 2026 Very Large Load secondary DFC of $11.49/kW | Deposit refundable over 10 years as delivery revenue arrives; letter of credit required at $2M and up | [CSV](data/ciac-four-utilities.csv) Sources for the table are the [Ohio rule](https://codes.ohio.gov/ohio-administrative-code/rule-4901:1-9-07), the [AEP Ohio tariff book](https://www.aepohio.com/lib/docs/ratesandtariffs/Ohio/September_2026_AEP_Ohio_Tariff_Book.pdf), Georgia Power's [Rules and Regulations for Electric Service](https://www.georgiapower.com/content/dam/georgia-power/pdfs/tariffs/rules-regs.pdf), PG&E's [Rule 15](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_RULES_15.pdf), [Rule 2](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_RULES_2.pdf), [Preliminary Statement Part J](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_PRELIM_J.pdf) and [Schedule B-20](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_SCHEDS_B-20.pdf), and ComEd's [2026 Schedule of Rates](https://www.comed.com/cdn/assets/v3/assets/blt3ebb3fed6084be2a/blt86ebee5fe6ed02f8/6aa02102cad5be96d2a2b30d/2026_Ratebook.pdf?branch=prod_alias). The revenue proxies use EIA's December 2025 industrial prices from [Electric Power Monthly](https://www.eia.gov/electricity/monthly/archive/february2026.pdf), Table 5.6.A: Georgia 7.53 c/kWh, Ohio 9.33, Illinois 10.10, California 19.86. At a $600k utility job the Ohio plant writes a $240k check, the Georgia plant $0, the PG&E plant $0 and the ComEd plant posts $0; at $2.5M the figures are $1.0M, $521k, $0 and a $1.12M refundable deposit. **Ohio's 60/40 split is the only one of the four that charges a 2 MW customer anything on a typical job, because it is the only one that ignores the revenue the load brings**. Georgia's 2.5x multiple on base-rate revenue, PG&E's revenue-divided-by-14.76% allowance and ComEd's five-year delivery-revenue threshold all say the same thing in different words: a 2,000 kW load at a 60% load factor pays for its own extension through the bill. Ohio's rule, effective in its current form since [June 2023](https://codes.ohio.gov/ohio-administrative-code/rule-4901:1-9-07) and unchanged in substance since 2009, does not. Three caveats sit in the [Interconnection Cost tool](https://open-factory.com/tools/interconnection-cost/) inputs rather than in the chart. Georgia Power applies its multiple to base-rate revenue, which excludes fuel and riders, so the true allowance is below the $2.0M full-bill proxy. PG&E's Rule 15 lets the company apply its own "non-residential allowances net revenue multiplier," and every dollar above the allowance carries the 24% [income tax component](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_PRELIM_J.pdf). ComEd's credit screen is the gate: fail it and the threshold falls back to a Standard Extension Cost the tariff defines but does not print. The [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) covers what the same 2,000 kW does to the Ohio bill after energization. ## The Gear You Own Ships in 8 to 50 Weeks The customer's side of the meter is the part the plant controls, and the lead times there are published monthly. Schneider's September guide is the one the Monitor includes this month: [50 weeks for LV drawout switchgear](https://www.se.com/us/en/download/document/Monthly_Lead/), 18 to 34 for switchboards, 24 for MTZ power breakers 600 to 6,000 A, 15 to 25 weeks for HVL medium-voltage switches and 40 or more for CBGS0 and GHA metal-clad, 24 weeks for I-Line II busway and 40 for plug-in units, 17 to 27 weeks for a 1,600 to 4,000 A transfer switch, 6 to 27 for an MCC lineup and 8 weeks for a DOE 2016 dry-type above 300 kVA. Siemens' schedule as republished runs a similar spread. Every one of those is a made-to-order figure from release to manufacturing, which starts at approved submittals, not at PO. The chart reads 50 weeks for LV switchgear, 34 at the top of the switchboard range, 40 for the longest MV switchgear and busway, 27 for a large ATS or a full MCC, and a pad-mount transformer spread from 13 to 104 weeks. **The customer-owned path for the transformer is 13 weeks from a stocking distributor at $98k to $104k list, against 52 to 104 weeks when the utility orders it**, on the three vendor listings we could find for a 2,500 kVA pad-mount: [Maddox](https://store.maddox.com/products/2500-kva-padmount-transformer) at $104,000 and in stock, [Bear Power Solutions](https://bearpowersolutions.com/product/2500-kva-3-phase-12470y-7200-277-480-padmount-transformer-copy/) at $98,000, and [Western Powerline Supply](https://www.westernpowerlinesupply.com/shop/p/2500kva) at $98,000 with a "90 day lead time." Those are list prices, n=3, not awards; the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) will replace them with municipal and utility award bands as the rows come in. The switchgear price belongs to [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/), and the [Quote Check tool](https://open-factory.com/tools/quote-check/) is where a spec goes in and a band and a lead time come out. What belongs here is the deposit. Estimators are now told that a [10 to 20% deposit](https://www.electronate.app/blog/switchgear-lead-times-2026-data-center-boom) on the switchgear value to hold a production slot is a standard commercial mechanism, and that the slot clock starts on the deposit while drawings are still moving. That single fact is what makes the paid section's schedule possible. The copper in the feeders moves weekly and lives in the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/). The contractor's hours are covered in [Datacenter Construction Is Eating Your Electricians](https://open-factory.com/articles/electricians-wages-by-metro/). ## The 30-Day Permit Clock and the Arc-Flash Line The permit is the one step where Ohio gives the plant a hard number. Ohio law requires a certified building department to act on a commercial plan submission within 30 days; Cincinnati's Buildings and Inspections department states it directly on its [permit review page](https://www.cincinnati-oh.gov/buildings/building-permit-forms-applications/permit-guide/permit-review-process/): "Ohio law requires that we review your permit application within 30 days from when we receive it." Columbus's Building and Zoning Services publishes its [submission standards](https://www.columbus.gov/Business-Development/Building-Zoning-Services/Building-Plan-Review/Commercial-Plans) and the code editions it reviews to, the 2024 Ohio Building Code and the 2023 NEC, but no turnaround figure of its own; the state clock is the one to plan on. Plans go in under [ORC 3791.04](https://codes.ohio.gov/ohio-revised-code/section-3791.04) with an engineer's seal, and a resubmittal restarts the count. The arc-flash study is the line the insurer adds after the budget is approved. **A study for a plant of this size costs $6,000 to $70,000 depending on bus count and drawings, and takes about nine weeks**: [Zech Engineering](https://www.zechengineers.com/arc-flash-study-cost/) puts mid-size facilities at $6,000 to $14,000 and larger industrial sites at $14,000 to $22,000 or more (2026 pricing), [Herzig Engineering](https://www.herzigengineering.com/post/arc-flash-study-cost-estimate) (June 2025) puts a campus at $15,000 to $35,000, and [e-Hazard](https://e-hazard.com/how-much-does-an-arc-flash-study-cost/) (updated January 2024) reports a large food plant at about $70,000: 15 days of data gathering, five weeks of modeling, seven days of labeling. The number of buses is the primary cost driver, and every panel, switchboard, MCC and transformer the upgrade adds is a bus. It is also not a one-time line. [NFPA 70E requires the risk assessment to be reviewed at intervals not exceeding five years](https://www.zechengineers.com/blog/when-to-update-arc-flash-study/) and whenever the system changes in a way that affects the results, and a new utility service, a new transformer and new protective-device settings are three such changes at once. The labels on the new gear are only correct after the utility confirms the available fault current at the new transformer, which is why the study sits at weeks 63 to 72 on the sequential chart and cannot be pulled earlier than approved submittals. [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/) has the clause that makes the switchgear vendor deliver the settings file the study needs. ## The Cost Stack Outside the Quote Put the five lines together and the size of the miss is clear. We built the stack for the Ohio plant, low case to high case, with the transformer shown as the customer-owned option and the bridge power shown per month. | Line item | Low | High | Mid | Basis | |---|---|---|---|---| | Utility line-extension share, AEP Ohio 40% | $120k | $480k | $240k | 40% of a $300k to $1.2M utility job | | Customer-owned 2,500 kVA pad-mount, list | $98k | $104k | $98k | Three vendor listings, n=3 | | Arc-flash study and NFPA 70E labels | $6k | $70k | $22k | Zech 2026, Herzig 2025, e-Hazard 2024 | | Temporary power rental, 2,000 kW, per month | $34k | $110k | $75k | Ziegler dry rate; Stag full-service range, Dec 2025 | | Temporary power fuel, per month | $133k | $486k | $184k | Cat XQ2000 fuel curve at $4.859/gal, 352 to 720 h | [CSV](data/cost-stack.csv) The bars read $120k to $480k for the Ohio CIAC share, $98k to $104k for the transformer, $6k to $70k for the study, $34k to $110k a month for the rental and $133k to $486k a month for fuel. The fuel line comes from the [Cat XQ2000 spec sheet](https://ohiocat.com/wp-content/uploads/2023/06/XQ2000-Ohio-CAT-Spec-Sheet-2020.pdf), which burns 77.7, 107.8 and 138.9 gallons an hour at 50, 75 and 100% of its 2,000 kW standby rating, priced at the [$4.859 a gallon](https://fred.stlouisfed.org/series/GASDESW) US on-highway diesel average for the week of March 9, 2026. The rental line runs from [Ziegler Rental's](https://www.zieglerrental.com/equipment/lights-generators/generators-diesel/2000kw-diesel-generator/) dry monthly rate of $33,945 for a 2,000 kW unit to the [$75,000 to $110,000](https://stagrentals.com/understanding-the-cost-of-renting-a-2000-kw-generator-what-you-need-to-know/) a month Stag Rentals quoted in December 2025 for a full-service package with fuel management. **One month of a 2 MW diesel bridge at two-shift load costs more than the arc-flash study, the permit and the transformer together**, and the [Genset TCO tool](https://open-factory.com/tools/genset-tco/) is where the run-hours assumption gets tested against your own interval data. ## What to Do Monday Start with the utility, but not with the question most plants ask. Send the complete package Sheet 103-5 lists, ask in writing for both the nonbinding and the binding estimate dates under OAC 4901:1-9-07(C), and ask whether the plan of service puts you on an existing circuit or a new one; that answer decides whether you are on a 357-day clock or a 684-day one by the only published benchmark. Ask for the customer-owned transformer option under Sheet 103-7 paragraph 3.c in the same letter, so the utility's estimate comes back both ways. Then call the switchgear vendor and the insurer in the same week. Get the slot-deposit terms in writing, because the 50-week clock starts on the deposit, not on approved drawings, and get the arc-flash proposal priced per bus with the labels included, because the study has to be redone at the next service change anyway. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) holds the four tariff formulas above and is adding the next eleven utilities in order of the plant counts in their territories. The [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) takes your PO dates and gear classes and returns the energization month. Behind the paywall: the parallel-path schedule that moves the crane from **week 74 to week 54**, what to order before the utility answers and on what paper, and the month-by-month diesel math that tells you whether a bridge at $167k to $520k a month beats waiting, with the run profiles in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) workbook. --- # Chillers Are the 60-Week Problem URL: https://open-factory.com/articles/chiller-60-week-problem/ Date: 2026-03-13 Cooling / Deep dive Air-cooled 40 to 60 weeks, water-cooled 50 to 70, Munters into 2028, Trane and Carrier and JCI backlogs, R-454B's summer, OFCI 18 months out, the 20 MW procurement calendar A water-cooled chiller over 1,000 tons quotes at [50 to 70 weeks](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026) in the United States as of May 22, 2026. An air-cooled machine over 500 tons is 40 to 60 weeks and a coolant distribution unit 30 to 50. None of those is the longest line on a datacenter schedule. The 230 kV transformer is 130 to 170 weeks and everyone knows it, so it gets a slot reservation two years out. The chiller is the item that still gets bought the old way, as part of a mechanical subcontract awarded after the guaranteed maximum price, which puts a 60-week machine on a schedule that has 40 weeks left. That is why the [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) treats chillers as the cooling pillar's headline class and why this piece exists. The Monitor's input for the whole HVAC industry says the same thing the consultants' tracker says about chillers. Census M3 puts unfilled orders at [$18.6 billion for NAICS 3334](https://www.census.gov/manufacturing/m3/) in July 2026, 2.7 months of shipments, against 1.6 months through 2015 to 2019. This piece covers the lead-time table for every class that cools a hall, seven quarters of backlog at Trane, Carrier and Johnson Controls, the order growth behind it, the component shortage under it, what liquid cooling does to the mechanical share of a build, what the R-454B summer did to refrigerant prices, and how to buy the machine 18 months before you need it. The ratio peaked at 3.1 months in July 2022, fell to 2.3 in January 2025 as the residential cycle turned down, and has climbed back to 2.7 by July 2026 with unfilled orders up $4.6 billion, or 33%, in 18 months. ## The Lead-Time Table, Cooling Rows **Every class of equipment that rejects heat from a datacenter hall now quotes at 30 weeks or more, and the two chiller classes sit at 40 to 70.** Origin Partners, a power and datacenter transaction advisor, publishes a quarterly equipment tracker. Its [Q2 2026 edition](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026) (updated May 22, 2026) is the only public source we found that quotes chillers and CDUs by tonnage class alongside transformers and switchgear. The Monitor cites it as theirs. Our own compiled series for cooling is the Census M3 ratio above, plus the OEM backlogs and PPI further down. A contributed panel of MEP quotes is the v1 layer and does not exist yet. Water-cooled chillers over 1,000 tons quote at 50 to 70 weeks, air-cooled over 500 tons at 40 to 60, CDUs at 30 to 50, against 26 to 40 weeks for a 1.5 to 3 MW diesel standby set and 80 to 110 for metal-clad switchgear. The tracker's own gloss on the cooling rows: "The shift to direct liquid cooling for high-density AI loads is producing a parallel supply chain story on CDUs and dry coolers that did not exist 18 months ago." Two OEM disclosures bracket those numbers from the supply side. Vertiv's 10-K for 2025 reports a [$15.0 billion backlog against $7.2 billion](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm) a year earlier and says the majority "is expected to be shipped within the next 12 to 18 months," which is 52 to 78 weeks and matches the tracker's upper rows. Munters, whose Data Center Technologies unit had about SEK 6 billion of sales and 1,000 staff in 2025, published five large orders between November 2025 and April 2026 with delivery windows in each release. | Order | Announced | Scope | Customer | Deliveries | |---|---|---|---|---| | 82 MUSD chillers | Dec 2, 2025 | Geoclima Circlemiser chillers, service, commissioning | US colo, hyperscale AI tenant | mid Q2 2026 to end Q3 2026 | | 840 MSEK chillers | Dec 11, 2025 | Circlemiser chillers, from the Virginia plant | New US colo customer | Q4 2026 to Q1 2027 | | 215 MUSD CRAHs | Nov 2025 | Custom over-the-rack CRAHs, platform-wide | US hyperscaler | end 2026 to early 2028 | | 2.1 BSEK CRAHs, CDUs, chillers | Dec 2025 | Chilled-water CRAHs, CDUs, Circlemiser chillers, start-up | US colo | Q4 2026 to Q1 2028 | | 2.0 BSEK CDUs and CRAHs | Apr 24, 2026 | High-capacity CDUs and over-the-rack CRAHs for an AI factory | US colo | early 2027 to Q1 2028 | *Munters' five largest data center orders deliver into Q1 2028. Source: Munters Group press releases via Cision, November 2025 to April 2026. Data: [08-munters-delivery-windows.csv](data/08-munters-delivery-windows.csv)* The [2.1 BSEK order](https://news.cision.com/munters-group-ab/r/munters-wins-record-orders-for-data-center-equipment-with-a-total-value-of-2-1-bsek,c4287586) for CRAHs, CDUs and chillers "begin[s] in Q4 2026 and continue[s] through Q1 2028"; the [2.0 BSEK CDU and CRAH order](https://news.cision.com/munters-group-ab/r/munters-wins-order-of-bsek-2-0-for-a-modular-ai-cooling-solution,c4339624) booked in April 2026 starts "in early 2027" and runs to Q1 2028; the [840 MSEK chiller order](https://news.cision.com/munters-group-ab/r/munters-wins-840-msek-data-center-chiller-order,c4280172) of December 2025 delivers Q4 2026 to Q1 2027. A colocation buyer who signed in December 2025 is taking chillers 10 to 15 months later. That is the 60-week problem stated as a calendar. ## Seven Quarters of Backlog at Trane, Carrier and JCI **Trane Technologies added $4.3 billion of backlog in the first half of 2026, four times the $1.05 billion it added in all of 2025.** We read the 8-K earnings exhibits for Trane, Carrier and Johnson Controls from Q4 2024 through Q2 2026 and built the series in the [Open Factory Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/). Trane's enterprise backlog was [$6.75 billion at the end of 2024](https://www.sec.gov/Archives/edgar/data/1466258/000146625825000005/exhibit991-q42024earningsr.htm), $7.8 billion at the end of 2025, [$10.7 billion at March 31, 2026](https://www.sec.gov/Archives/edgar/data/1466258/000162828026028547/exhibit991-q12026earningsr.htm) and a [record $12.1 billion at June 30, 2026](https://www.sec.gov/Archives/edgar/data/1466258/000162828026050743/exhibit991-q22026earningsr.htm), up 70% year over year, with Americas Commercial HVAC backlog up 90%. Trane went from $7.8 billion to $12.1 billion between December 2025 and June 2026, Johnson Controls from $14.9 billion to $21.0 billion between September 2025 and June 2026, and Vertiv from $7.2 billion to $15.0 billion between the two year-ends. Two caveats sit inside the chart. JCI [restated its backlog from fiscal Q1 2026](https://www.sec.gov/Archives/edgar/data/833444/000083344426000014/q1ex991xq1fy26earningsrele.htm) to include equipment-only sales for longer-cycle projects, so the jump from $14.9 billion to $18.2 billion between September and December 2025 is partly a definition change; the organic growth rates it reports (+20%, +26%, [+32% in fiscal Q3 2026](https://www.sec.gov/Archives/edgar/data/833444/000083344426000083/q3ex991xq3fy26earningsrele.htm)) are the cleaner read. And Carrier does not publish a backlog in dollars at all, which is why it is missing from the chart and why it scores low on that line of the [Open Factory OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/). The composition matters more than the total. On the Q2 2026 call, Trane's management said the backlog is "almost 95% for commercial HVAC" and that ["we have about $6 billion of the backlog, that's for 2027."](https://www.fool.com/earnings/call-transcripts/2026/07/30/trane-technologies-tt-q2-2026-earnings-call-transcript/) Half of Trane's order book is already spoken for in the year after next. An analyst on the same call framed the arithmetic: the backlog is roughly half of a year's revenue and is attributable mostly to applied equipment, which is 15 to 20% of revenue. Applied equipment is the trade term for the chillers, air handlers and CDUs that a datacenter buys. Carrier's language is the same without the dollars. Q4 2025 Commercial HVAC orders were ["up nearly 50% driven by key data center wins"](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000005/a8-kerexhibit99xq42025.htm); in Q1 2026 data center orders were ["up over 500%" and "backlog fully covers expected 2026 data center sales"](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000023/a99-q12026earningsexhibit.htm); in Q2 2026 total orders were up about 40%, Commercial HVAC about 65% and [data centers more than 300%](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000030/a99-q22026earningsexhibit.htm), with 2026 sales guided to about $23 billion. Carrier told investors in February 2025 it expected [$1 billion of data center sales in 2025](https://www.facilitiesdive.com/news/carrier-unveils-data-center-strategy-growth-hvac-sales-q4-quantum-leap/739923/), which against $21.7 billion of 2025 net sales is 4.6% of the company. A vertical that is 5% of revenue is moving a $22 billion company's order book. ## Data Center Orders Grew 100% to 500% a Quarter **The order growth rates the OEMs disclose for data center and applied equipment have run between 37% and 500% for five consecutive quarters, and the fastest growers are the smallest bases.** Trane reported applied equipment bookings [up over 60% in Q2 2025](https://www.sec.gov/Archives/edgar/data/1466258/000146625825000149/exhibit991-q22025earningsr.htm), over 100% in Q3, [more than 120% in Q4 2025](https://www.sec.gov/Archives/edgar/data/1466258/000162828026003953/exhibit991-q42025earningsr.htm) with an applied book-to-bill of 200%, more than 160% in Q1 2026 and 130% in Q2 2026. Regnery's summary on the Q2 call: "On a 2-year stack, applied bookings are up more than 4x." Trane's applied bookings grew 60%, 100%, 120%, 160% and 130% in the five quarters to June 2026; Carrier's data center orders grew more than 500% and 300% in the first two quarters of 2026; JCI's Building Solutions North America orders grew 56%, 40% and 37% organically, each release attributing the growth to data centers; and Munters' Data Center Technologies order intake grew [416% in Q4 2025](https://news.cision.com/munters-group-ab/r/exceptional-demand--while-earnings-weakened,c4299489), taking the group's 2025 book-to-bill to 1.6x and its backlog up 53%. Modine is the purest read because it now reports data centers as a segment. Data Centers sales were [$348.6 million in the quarter to June 30, 2026, up 90%](https://www.sec.gov/Archives/edgar/data/67347/000110465926088230/mod-20260729xex99d1.htm), 40% of a $874 million company, after [158% growth in the March 2026 quarter](https://www.sec.gov/Archives/edgar/data/67347/000110465926066291/mod-20260526xex99d1.htm). Its CEO reported "three consecutive quarters of record order intake leading to our backlog nearly doubling over the past year." Vertiv's 2025 backlog doubled. Munters' DCT unit is 40% of the group. The buyers who will be told "60 weeks" this year are competing with these order books, not with the plant down the road. ## Components Are the Second Bottleneck **Two chiller and CRAH makers have now told the SEC or their shareholders that components, not factory floor space, are what limits output.** Modine's 10-K for the year to March 2026 is the plainest: ["demand for certain key components is currently outpacing supplier capacity. We began experiencing supply shortages in the fourth quarter of fiscal 2026. These component shortages are negatively impacting our production schedules and are tempering our planned production ramp-up for certain data center products."](https://www.sec.gov/Archives/edgar/data/67347/000110465926066795/mod-20260331x10k.htm) The Data Centers segment's gross margin fell 960 basis points to 20.2% in the June 2026 quarter, which the company attributed to capacity expansion costs plus "production inefficiencies due to supply chain constraints." Munters said the same thing three months later. Its [Q2 2026 report](https://news.cision.com/munters-group-ab/r/exceptional-demand---external-headwinds-currently-affecting-profitability,c4375573) (July 17, 2026) records group order intake up 137% and DCT sales down, because "the production ramp-up and supply chain constraints impacted throughput," with "delays in component deliveries" named alongside tariffs, and the mitigation is the usual one: "qualifying additional suppliers, increasing inventory of critical components." Its US chiller plant in Virginia was to be [operational by the end of Q1 2026](https://news.cision.com/munters-group-ab/r/munters-wins-82-musd-data-center-chiller-order,c4275579) and was still described as ramping in July. Trane, with the largest applied footprint, is the calmest voice and still concedes the point. Asked about suppliers on the Q2 2026 call, Regnery said: "are we seeing constraints? Yes, but we always do. Are we seeing a bit more than normal? Maybe a bit more normal, but nothing that our team can't manage." The company says it has [expanded applied capacity four times over the last three years](https://www.facilitiesdive.com/news/trane-managing-capacity-as-commercial-hvac-demand-surges/826711/), is investing in Grand Rapids, and is "not turning away orders." A buyer should read those three statements together: the assembler with the most capacity is the one saying supplier constraints are "a bit more than normal," and the two smaller assemblers are saying their throughput is already limited by parts. Neither filing names the parts. Whatever they are, a chiller buyer cannot second-source them on the buyer's own schedule, which is the argument for buying the machine, not the subcontract, early. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) tracks the OEM statements quarter by quarter on its cooling page. ## Mechanical Is 33% of a Liquid-Cooled Build **Liquid cooling moves 11 points of a datacenter's construction budget out of shell and electrical and into mechanical, so the 60-week item is also a bigger share of the money.** Turner & Townsend's [2025 Data Centre Construction Cost Index](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends) gives an indicative US cost allocation: mechanical including equipment is 22% of an air-cooled build and 33% of a liquid-cooled one; electrical falls from 54% to 48%; core, shell and architectural from 14% to 9%; general conditions stay at 10%. The same report puts the liquid-cooled premium at 7 to 10% for similar IT capacity and prices Phoenix and Columbus at $9.8 per watt, Atlanta at $9.9, Silicon Valley at $13.3 and Tokyo at $15.2. Mechanical is 22% of an air-cooled US build and 33% of a liquid-cooled one, electrical 54% and 48%, shell 14% and 9%. At Phoenix's $9.8 per watt that is $2.2 million per MW of mechanical for air and $3.2 million per MW for liquid, or $43 million against $65 million for a 20 MW hall before the 7 to 10% premium. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lists each line with the Monitor's lead time beside it and the carrying cost of a month's slip. The price of the equipment inside that share has not stood still either. The BLS producer price index for [air conditioning and refrigeration equipment](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU1148) is 62% above January 2020 as of July 2026 and 4.6% above July 2025. The index rose 40% between January 2020 and July 2022, drifted for two years, and added another 9% from January 2025 to July 2026. A chiller quote from 2021 is not a reference for anything. The [Quote Check](https://open-factory.com/tools/quote-check/) tool applies the PPI path and the published lead time to a spec so the buyer sees both bands before opening a bid. ## R-454B's Summer **The refrigerant transition added a second supply shock to chillers in 2025, and the May 2026 rule change that eased it for houses left the chiller limit untouched.** EPA's Technology Transitions rule of October 2023 ([88 FR 73098](https://www.federalregister.gov/documents/2023/10/24/2023-22529/)) restricts HFCs with a GWP of 700 or more "for chillers--comfort cooling beginning January 1, 2025," for "all compressor types," and sets the same 700 limit for datacenter and computer-room cooling equipment from 2027 and for industrial process chillers from January 1, 2026 or later by temperature. Residential and light commercial AC hit the same 700 limit on the same day, and that is where the trouble started: the field moved to R-454B all at once. A 20 lb cylinder that cost [$345 in 2021](https://aciq.com/aciq-dealer-program/2025-epa-refrigerant-phase-out/) listed at $310 from a Virginia supply house's regular distributor in March 2025 with a six-week wait, against [$730 on the spot market](https://www.achrnews.com/articles/164334-contractors-should-plan-ahead-to-obtain-r-454b) the same week, and traded at [$700 to $2,000 through the summer](https://aristotleair.com/r-454b-refrigerant-shortage-update/) with regional jumps from $485 to $1,050 "in weeks"; all of these are single reported quotes, not a distribution, and the piece says so. The mechanism was cylinders, not molecules. Chemours told ACHR News in March 2025 that it had "lead times of approximately 10 weeks for new orders" and that "the primary issue with R-454B packaged product for the aftermarket is A2L cylinder availability." At ACCA's June 18, 2025 panel, Worthington said the industry had ["probably 10x'ed them, 5x'ed them"](https://hvac-blog.acca.org/r-454b-shortage-crisis-whats-being-done-and-when-will-it-end/) on forecasts, "one entire batch failed testing and couldn't be used," Honeywell had raised blending and down-packing capacity 44% and Arkema had entered the market in mid-May. By October 13, 2025 HARDI's Alex Ayers told the same outlet ["the crisis is over,"](https://www.achrnews.com/blogs/17-opinions/post/165373-hardi-on-r-454b-the-crisis-is-over) with availability restored in late August and September and prices "still elevated" but falling month over month. Lennox said in January 2026 that the [40% of its 2025 residential mix still on R-410A "is gone, and it's all going to be R-454B"](https://www.achrnews.com/articles/165750-lennox-is-cautiously-optimistic-about-2026) in 2026. | Date | Instrument | What it did | |---|---|---| | Oct 24, 2023 | Technology Transitions final rule, 88 FR 73098 | Comfort-cooling chillers: GWP 700 limit from Jan 1, 2025, all compressor types; datacenter and computer-room cooling: GWP 700 from 2027; IPR chillers from Jan 1, 2026 or later | | Dec 26, 2023 | Interim final rule | Residential and light commercial AC install deadline extended to Jan 1, 2026 for pre-2025 components | | Jan 1, 2025 | Compliance date | New comfort-cooling chillers and residential AC must use refrigerants under 700 GWP; R-454B (465) and R-32 (675) become the volume products | | Mar to Sep 2025 | R-454B cylinder shortage | Lead times to 10 weeks at Chemours, cylinders $700 to $2,000, demand 5 to 10x forecasts, one cylinder batch failed test | | Oct 3, 2025 | Proposed reconsideration, 90 FR 47999 | EPA proposes to revisit compliance dates and limits across subsectors | | Oct 13, 2025 | HARDI: "the crisis is over" | Availability restored late August to September; prices elevated but falling month over month | | Nov 20, 2025 | CY2026 allowance allocations, 90 FR 52391 | 2026 HFC allowances issued at the 60% step | | May 26, 2026 | Final reconsideration, 91 FR 31284, effective Jul 27, 2026 | Install deadline removed for pre-2025 residential systems; IPR and semiconductor chiller dates to Jan 1, 2030; supermarket and cold-storage limits raised until 2032. Comfort-cooling chiller limit unchanged. | *The refrigerant rule changed four times in 31 months; the chiller limit never moved. Source: Federal Register (88 FR 73098, 90 FR 47999, 90 FR 52391, 91 FR 31284); ACHR News; ACCA. Data: [11-aim-act-timeline.csv](data/11-aim-act-timeline.csv)* The final reconsideration of May 26, 2026 ([91 FR 31284](https://www.federalregister.gov/documents/2026/05/26/2026-10387/), effective July 27, 2026) removed the installation deadline for pre-2025 residential and light commercial systems, pushed industrial process and semiconductor chillers to January 1, 2030, and raised supermarket and cold storage limits to 1,400 and 700 until 2032; it changed nothing for comfort-cooling chillers, which have been at 700 GWP since January 1, 2025. For a plant or datacenter chiller that means R-1234ze or R-513A on centrifugals and screws, or R-454B and R-32 on smaller machines, and it means A2L machinery-room ventilation, leak detection and technician certification that never appear on the equipment quote. [The R-454B Summer](https://open-factory.com/articles/refrigerant-transition-prices/) runs the pass-through into chiller pricing line by line. The point for this piece is that the OEM ramping a new refrigerant platform in 2025 was also the OEM whose order book doubled, and the [Cooling Roadmap](https://open-factory.com/roadmaps/cooling-roadmap/) tracks which platforms have shipped on which refrigerant. ## Buying the Chiller 18 Months Out **The hyperscalers already buy chillers the way utilities buy transformers: owner-furnished, slot-reserved, sometimes with a deposit on the factory itself, and the lead-time table says every 20 MW developer now has to copy them.** Owner-furnished, contractor-installed procurement moves the chiller purchase order from the mechanical subcontract to the owner's own desk at basis of design, which is the only point on a schedule that is 70 weeks before mechanical completion. The trade's rule of thumb is that it saves [roughly 15% of contractor markup](https://www.irecruit.co/insights/ofci-vs-cfci-procurement-owner-vs-contractor-furnished-equipment-data-centers) on the equipment and costs up to 6% in installer productivity on unfamiliar gear; those are a recruiter's practitioner estimates, not a measured distribution, and [OFCI vs Design-Build](https://open-factory.com/articles/ofci-vs-design-build-chillers/) works through what a colocation developer can and cannot copy from the hyperscaler playbook. The far end of that playbook is the capacity reservation. Modine's 10-K discloses that in March 2026 it "entered into a long-term capacity agreement with one of its strategic data center customers" and "received a [$165 million up-front deposit](https://www.sec.gov/Archives/edgar/data/67347/000110465926066795/mod-20260331x10k.htm) from the customer to support investments necessary to meet the planned sales volume," expecting "to sell more than $4 billion of data center cooling products to this customer during calendar years 2027 through 2029." One buyer has reserved three years of a chiller and CRAH maker's expansion before the expansion is finished. [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/) covers how the same instrument spread from transformers and turbines to cooling. The Modine deposit is the largest one disclosed in cooling so far. Between the mechanical subcontract and the $165 million deposit there is a middle path, and it is what to do Monday. First, price the schedule, not the machine: run the hall through the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) with the Monitor's published 50 to 70 weeks for water-cooled chillers and 30 to 50 for CDUs, and if the chiller purchase order is not dated before the civil permit, the energization month is wrong. Second, issue the chiller and CDU purchase orders as owner-furnished at basis of design, with a named factory slot week and [liquidated damages at 0.5% a week](https://open-factory.com/articles/liquidated-damages/) written against that slot, not against a vague "ship date." Third, write UQD compliance and a second CDU source into the spec, because the CDU is the one class on the table where a second qualified vendor exists at 30 to 50 weeks. Fourth, put the refrigerant on the face of the purchase order (R-1234ze, R-513A or R-454B), with the A2L room ventilation and detection scoped to the mechanical contractor, and run the landed cost of an imported machine through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) before choosing a European or Asian line. Fifth, ask each OEM in the RFQ for its backlog and next open slot in writing, following the [RFQ template for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/), and grade the answers against the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/): Trane and JCI publish a backlog, Carrier does not, and the one that will not tell its shareholders will not tell you. Behind the paywall: the OEM-by-OEM table of where 2027 chiller capacity actually opens and why, from seven quarters of filings and calls, and the week-by-week OFCI calendar for a 20 MW hall that puts the chiller purchase order **78 weeks** before mechanical completion, both built on the [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) and the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # What a Haas VF-2 Really Costs in 2026 URL: https://open-factory.com/articles/haas-vf2-true-cost/ Date: 2026-03-13 Machinery / TCO $70,995 list, 63 used listings and 3 prices, PM at 2 to 3.5% a year, spindle rebuild $8k to $25k, ten-year bill ~$173k before labour, Section 179 at $2.56M, the quote form A new Haas VF-2 lists at [$70,995](https://machinetoolindex.com/machines/haas-vf-2), and that number is 35% of what the machine will cost a one-shift job shop over ten years. The other 65% is maintenance, one spindle, power, coolant, cutting tools, and the vises and holders that the list price does not include. We built the bill line by line from Haas's published list, Haas Tooling's catalog, Hymson's maintenance benchmarks, EIA's industrial tariff and IRS Publication 946, and we opened 63 used listings to see what a five-year-old machine actually asks. This piece covers the sticker and what it excludes, the used market that hides its prices, maintenance and the spindle, power and tooling, the utilization curve, the Section 179 cash position, and behind the wall the configuration that wins, five checks before buying used, and what the [Open Factory Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) can and cannot yet say about machine tools. Read left to right: $70,995 of machine, ~$10,000 of rigging and startup tooling, $19,524 of preventive maintenance, $16,000 of spindle and ball screw, $11,004 of electricity, $15,000 of coolant and lube, $60,000 of cutting tools, a gross outlay of $202,523, and a net of $172,523 after a $30,000 resale in year ten. The low and high cases in the CSV run $122,590 to $266,981 net. ## The Sticker Is a Third of the Bill Every line in the waterfall has a source or an (A) for assumption, and the CSV under the chart carries both. The sourced lines are the list price ([Haas via MachineToolIndex, verified 31 Aug 2026](https://machinetoolindex.com/machines/haas-vf-2)), preventive maintenance at [2.0 to 3.5% of replacement value a year](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/), an integral spindle rebuild at [$8,000 to $25,000](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/), a ball screw regrind at $2,500 to $5,000 per axis from the same table, electricity at the [EIA industrial rate of 9.17 cents per kWh in June 2026](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a), and the vises and holders priced from [Haas Tooling](https://www.haastooling.com/c/vise-kits) in September 2026. The assumptions are freight, rigging and install ($6,000), coolant and lube ($1,500 a year), consumable cutting tools ($6,000 a year), a 6 kW average draw, and a $30,000 resale. **The base case puts the ten-year net cost at $172,523, or 2.4 times the sticker, before anyone is paid to stand in front of the machine.** Operator labour is excluded on purpose: it dwarfs every line here and it is a shop decision, not a machine decision. Floor space, insurance and a CAM seat are excluded for the same reason. What is left is the cost of owning and feeding one VF-2, which is the number a buyer should compare against a used machine, an SS, or a competitor's quote through [Quote Check](https://open-factory.com/tools/quote-check/). The two lines that move the total most are the ones with an (A). Cutting tools at $6,000 a year are 30% of the gross outlay, and a shop cutting titanium will spend double while a shop cutting 6061 spends half. Resale at $30,000 is anchored on one data point we could verify, a [2008 VF-2 that KD Machinery marked sold at $22,750](https://surplusrecord.com/listing/haas-vf-2-30-x-16-y-20-z-10000-rpm-24-automatic-tool-changer-cat-40-haas-cnc-2008-1088411/), and on Machinio's asking band for 2016 to 2022 machines, which we return to below. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) exists to replace both assumptions with contributed numbers. ## What $70,995 Buys and What It Does Not The list price buys the standard machine: 30 by 16 by 20 inch travels, a 22.4 kW (30 hp) 8,100 rpm 40-taper spindle, a 20-pocket carousel changer at 4.2 seconds tool to tool, 1,000 ipm rapids, a 3,539 kg machine holding 208 litres of coolant, and a service requirement of [220 VAC three-phase at 70 A or 440 VAC at 35 A](https://machinetoolindex.com/machines/haas-vf-2). The same page includes the sentence that matters: "Options, tooling, freight and installation are extra." Haas is one of the very few builders that publishes list prices at all, which is why it scores where it does on the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/), and why this article can exist for a VF-2 and not for a Mazak. The ladder reads $70,995 for the [VF-2](https://machinetoolindex.com/machines/haas-vf-2), $78,995 for the [VF-2YT](https://machinetoolindex.com/machines/haas-vf-2yt) with 20 inches of Y, $83,995 for the [VF-2SS](https://machinetoolindex.com/machines/haas-vf-2ss) with a 12,000 rpm inline spindle and a 30+1 side-mount changer, and $87,995 for the [VF-3](https://machinetoolindex.com/machines/haas-vf-3). The SS premium of $13,000 is 18% of the base machine and is the one option Haas prices as a model rather than a line item. **Probing, through-spindle coolant, a chip conveyor and 4th-axis wiring are not in the $70,995, and Haas's Build and Price tool sits behind a bot check we could not read in this pass.** So the waterfall carries the standard machine and states the options gap rather than inventing it. The used listings we opened show what shops actually order: six of the ten Premier Equipment machines were SS models with 12,000 rpm spindles, three of the Surplus Record machines listed a WIPS or Renishaw probe, and one 2021 machine at Resell CNC had a [spindle probe, through-spindle coolant and a programmable coolant nozzle](https://www.resellcnc.com/products/used-cnc-verticals/01tVS00000EgDnBYAV/haas_vf-2ss). A configured machine, not the base one, is what resells. Freight and rigging are the second unpriced line. Hymson's rate card gives [$150 to $225 an hour for distributor service labour and $1.50 to $2.00 a mile for travel](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/), and a VF-2 install is a day for two technicians plus a rigger and a truck. We carry $6,000 and mark it (A). The [RFQ template for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) asks the distributor to break freight, rigging, install and training into four lines, and the answer is usually between $4,000 and $8,000 for a machine this size, a spread [Quote Check](https://open-factory.com/tools/quote-check/) flags when a distributor folds it into one line. ## Machine Tool Prices Are Up 30% Since 2016 We wanted to chart the VF-2's list price year by year from archived Haas price lists. The Internet Archive was offline for the whole of our research window, so the history we can show is the industry's, not the model's. The BLS producer price index for machine tool manufacturing tells the shape. The index sat at 109.3 in January 2016 and 116.4 in January 2021, a 6.5% rise in five years, then climbed to [142.6 by July 2026](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCU333517333517), a further 22.5% in five and a half. Scaled by that ratio, today's $70,995 machine would have carried a list price near $54,400 in 2016 and $57,900 in early 2021, and the CSV under the chart carries the implied series with the arithmetic labelled as ours. **A VF-2 bought in 2021 and sold today has lost less to depreciation than list inflation has added, which is why used asking prices sit so close to new.** The mechanism is the same one we described in [Japan Is Eating the US Capex Boom](https://open-factory.com/articles/japan-eats-capex-boom/): builders raised prices into a demand wave, and the [first half of 2026 was the strongest half-year for US machine tool orders since 1998 at $3.44 billion](https://www.americanmachinist.com/news/news/55397089/machine-tool-demand-is-booming-in-2026-us-manufacturing-technology-orders-report-june-2026), with 11,243 units and an average order near $306k that we unpacked in [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/). Haas builds the VF-2 in Oxnard, California, so the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) returns zero Section 232 exposure on the machine itself. The [September 2026 tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) still matters for the Taiwanese vises and Japanese carbide that go on it. ## The Used Market Hides Its Prices A buyer who wants to know what a 2018 VF-2 is worth can find dozens for sale and almost no prices. We opened every VF-2 family listing we could reach on 10 September 2026 and counted. [Surplus Record](https://www.surplusrecord.com/search/?q=haas+vf-2) showed 51 VF-2 family listings, nine dealers in the 32 rows we captured, and 3 of them carried a price: a 1994 machine at $5,950, the 2008 machine sold at $22,750, and a 2011 parts machine at $3,069. [KD Machinery](https://www.kdmachinery.com/?s=haas+vf-2&post_type=product) listed 20, every one "Call for Price". [Premier Equipment](https://premierequipment.com/search/vf-2/) listed 10 with dates from May to September 2026 and no prices. [Resell CNC](https://www.resellcnc.com/products/used-cnc-verticals/01tVS00000EgDnBYAV/haas_vf-2ss) showed a 2021 VF-2SS behind "Please login to see price". Machinio and eBay blocked automated access outright. The 63 rows we captured are in the CSV under the chart. Premier's ten listings are the only ones with dates, and they show how fast the good machines move. A 2022 VF-2SS posted on 9 June 2026 and a 2023 VF-2SS with a rotary table posted on 21 May 2026 were both marked sold by the time we opened the page in September. The 2016 standard machines posted on 10 August and 2 September were still for sale, as were a 2010 SS and a 2014 VF-2 with the older 20 hp spindle. Late-model SS machines clear in weeks at a price nobody publishes, while base machines from 2014 to 2016 sit. That pattern is the resale argument for buying the configured machine in the first place, and it is why our resale assumption is $30,000 for a standard VF-2 and not the $41,900 floor of Machinio's band. Of the 63 machines, 31 were model year 2015 or newer and 12 were 2020 or newer, including a 2025 VF-2SS at KD Machinery and a 2023 VF-2SS with a rotary table that Premier marked sold in May 2026. The supply is young because shops trade up to an SS or a UMC and because the machines hold value. The machines hold value because the used channel keeps the clearing price private. **The only public asking-price series we found is Machinio's model page, and it put 2016 to 2022 machines at $41,900 to $62,900 when we read it on 9 September 2026, 59% to 89% of new list.** The page blocked our attempt to re-open it the next day, so the bands below are as read once, and marked as such. The three bands read $8,900 to $12,500 for 1994 to 2008 machines, $34,500 to $39,950 for 2010 to 2015, and $41,900 to $62,900 for 2016 to 2022, against the $70,995 reference line. Asking is not sold, and the two priced Surplus Record machines from 2008 and 1994 came in at $22,750 sold and $5,950 asked, both outside the Machinio band for their years. [Building the Missing Index](https://open-factory.com/articles/used-cnc-auction-index/) lays out how we intend to turn realized auction hammers into a monthly series; until it exists, the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) contribution form is the only place a buyer can put the price they actually paid. ## Maintenance: 2 to 3.5% a Year Plus One Spindle Preventive maintenance is the most predictable line and the one most shops skip. Hymson's benchmark table, [updated 24 June 2026](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/), puts a standard 3-axis VMC at 2.0 to 3.5% of new replacement value a year, which is $1,420 to $2,485 for a VF-2, and a [comprehensive PM contract for one machining center at $1,500 to $4,000 a year](https://www.in-housecnc.com/cnc-machine-repair-cost/) in In-House CNC's July 2026 cost guide. The two sources agree, and our base case takes 2.75%, or $1,952 a year and $19,524 over ten. The VF-2 sits at the bottom of the table: 2.0 to 3.5% for a 3-axis VMC, 2.5 to 4.0% for a 2-axis lathe, 4.0 to 7.0% for a 5-axis or mill-turn, and 5.0 to 8.0% for grinders. A UMC-500 costing $120,000 at 5.5% costs $6,600 a year to maintain against the VF-2's $1,952, which is a $46,000 gap over ten years before anyone counts the crashes. **The spindle is the one component that turns a maintenance budget into a capital event, at $8,000 to $25,000 for an integral spindle and an expected interval of 8,000 to 15,000 spindle hours.** At 2,000 hours a year that is one rebuild between year four and year eight, and at three shifts it is two. In-House CNC's numbers are lower, [$1,500 to $4,000 for a rebuild of common assemblies and $5,000 to $15,000 for a new high-speed spindle](https://www.in-housecnc.com/cnc-machine-repair-cost/), so our $12,000 sits between the two sources. We asked the spindle shops for a price and got none: [GTI Spindle](https://www.gtispindle.com/spindle-repair/) publishes a 10 to 20 business day turnaround and a one-year warranty, [Gilman Precision](https://www.gilmanprecision.com/spindle-repair/) publishes a repair evaluation form, [PDS](https://spindlerepair.com/) publishes a 24-month warranty, and not one publishes a dollar figure. Haas sells the cartridge through its parts store, which sits behind the same bot check as the price list. Ball screws are the other wear item. Hymson prices a regrind at $2,500 to $5,000 per axis and a new OEM screw at $4,000 to $9,000, and we carry one regrind over ten years at $4,000. A shop that crashes the machine twice a year will replace a screw instead, and a shop that never crashes it will not touch them. The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) tracks the probing and tool-breakage-detection options that cut the crash rate, which is the best maintenance spend on the list. ## Power, Coolant and Cutting Tools Electricity is the line buyers worry about and should not. The VF-2 draws a 26.7 kVA service at 220 V and 70 A, but the spindle is rated 22.4 kW and runs nowhere near it on average. We assume a 6 kW mean draw over spindle hours, mark it (A), and multiply by the [EIA industrial rate of 9.17 cents per kWh in June 2026](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a). One shift is 12,000 kWh and $1,100 a year at the US average, $790 in Texas at 6.58 cents, and $2,489 in California at 20.74 cents. The US industrial rate ran [6.76 cents in 2016, 6.67 in 2020, 8.62 in 2025 and 8.89 year to date in 2026](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_3), a 31% rise that adds $260 a year to a one-shift VF-2. Ten years of power at the US average is $11,004, 5% of the gross outlay; in California it is $24,888 and still only 11%. The shops that should care are three-shift shops in PJM states, where [the same table shows Ohio at 9.89 cents](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a) and rising. **Cutting tools and workholding are the lines the sticker hides best, and they are 32% of the ten-year gross outlay in our base case.** Startup is sourced: Haas Tooling lists the [Essential Mill Workholding Kit with a 6 inch manual vise at $999.95](https://www.haastooling.com/c/vise-kits) and a [second 6 inch fixed-jaw vise at $589.95](https://www.haastooling.com/c/fixed-jaw-vises), a [CT40 ER32 collet chuck at $79.95](https://www.haastooling.com/c/er-collet-chucks), and [CT40 end mill holders at $80.95 to $92.95](https://www.haastooling.com/c/end-mill-holders), all list, all read on 10 September 2026 with IMTS sale prices 5 to 25% under. Twenty holders, two vises, collets and knobs come to roughly $4,000 before one end mill. Consumable carbide at $6,000 a year is our assumption and the widest range in the model, $4,000 for aluminum job work to $10,000 and beyond for steel and titanium. Coolant, way lube and filters at $1,500 a year are the last (A). A 55 gallon sump refilled twice a year with a 5% emulsion, a drum of way oil, and mist filters land in that range for a one-shift shop. A shop that lets the sump go rancid will spend more on disposal than on concentrate. The [50-SKU MRO basket](https://open-factory.com/articles/mro-price-basket/) covers what those consumables cost at Grainger, McMaster and Amazon Business, and the spread between the three is wider than the spread between coolant brands. ## Utilization Decides the Hourly Rate A machine cost is a rate, not a total, and the rate depends on hours. We split the base case into fixed costs (capital less resale, 60% of PM, the ball screw) and variable costs (40% of PM, a spindle rebuild every 12,000 hours, power at 6 kW, coolant at $0.75 an hour, cutting tools at $3.00 an hour) and ran it at four utilizations. At 1,000 spindle hours a year the VF-2 costs $12.36 an hour all-in and $9.36 for ownership alone; at 2,000 hours it is $9.03 and $6.03; at 4,000 hours $7.36 and $4.36; at 6,000 hours $6.80 and $3.80. **Going from one shift to two cuts the hourly cost 19%, and the ownership component falls 28%, because $51,000 of capital and $16,000 of fixed maintenance are spread over twice the hours.** Past two shifts the curve flattens: cutting tools are $3.00 of every hour regardless, and the second spindle rebuild arrives in year five instead of year ten. Two things the chart does not show. Resale falls with hours, and a 6,000-hour-a-year machine at year ten has 60,000 spindle hours on it against 20,000 for the base case, so the $30,000 residual is optimistic there and the three-shift bar is a floor, not an estimate. And the spindle line is linear in the model but lumpy in life: Hymson's 8,000 to 15,000 hour interval puts a three-shift shop's first rebuild in year two, when the machine is still under a payment plan, which is the cash event to budget for rather than the average. The number a job shop should carry into a quote is the ownership rate, $4 to $6 an hour at one to two shifts, plus its own tooling and labour. The $75 to $95 an hour that job shops typically quote for 3-axis time (a rule of thumb, not a survey) is not recovering machine cost. It is recovering a machinist, overhead and margin, and the machine is a rounding error at two shifts. That is also why an under-utilized VF-2 is the most expensive machine in the building, and why the [used-refurbished market](https://open-factory.com/articles/used-refurbished-market/) is full of machines that ran 800 hours a year. ## Section 179 Beats Any Discount The 2026 Section 179 limit is [$2,560,000 with the phase-out starting at $4,090,000](https://www.irs.gov/publications/p946), per IRS Publication 946, and the same publication restores the [100% special depreciation allowance for property acquired and placed in service after 19 January 2025](https://www.irs.gov/publications/p946). A VF-2 is 2.8% of the limit. A shop buying one, or ten, expenses the whole price in the year the machine is placed in service, and the year-one tax saving is 21% of $70,995 for a [C corporation](https://www.law.cornell.edu/uscode/text/26/11) and up to 37% for a [pass-through owner in the top bracket](https://www.law.cornell.edu/uscode/text/26/1). A cash buyer at 21% is out $56,086 net in year one; at 37% the figure is $44,727. A buyer who finances 100% over 60 months at 8.75% (the [prime rate of 6.75% on 2 September 2026](https://fred.stlouisfed.org/graph/fredgraph.csv?id=DPRIME) plus 2 points, our assumption) pays $1,465 a month, $17,582 in year one, and deducts the full $70,995 plus $5,745 of interest, for a year-one net of minus $1,466 at 21% and plus $10,812 at 37%. Total interest over the loan is $16,913, 24% of the price, and the year-one shield alone is $14,909 to $26,268. **Expensing the machine in year one is worth $14,909 to $26,268, which is more than any discount a Haas Factory Outlet will put on paper for a single VF-2.** The 100% bonus rate removes the old reason to time purchases around a phase-down, so the only calendar constraint left is "placed in service" by 31 December, and a VF-2 built to stock in Oxnard makes that easy. We could not read Haas Financial Services' rate sheet, so ask for it in the RFQ and compare against prime plus 2 before signing. The rate is one of the eleven items the [switchgear quote anatomy](https://open-factory.com/articles/switchgear-quote-anatomy/) taught us to pull onto one line. ## What to Do Monday Price the whole bill, not the sticker. Take the base case CSV under the first chart, replace our three assumptions (cutting tools, coolant, resale) with your own shop's numbers, and you have a ten-year cost that a used machine, an SS, or a Brother or DN Solutions quote must beat. Ask the distributor for freight, rigging, install and training on four lines, and for the option prices Haas will not show a bot, then run the total through [Quote Check](https://open-factory.com/tools/quote-check/) so the number sits next to the published list and the tooling catalog. Write two things into the PO. First, the option list by Haas part description, so the machine that arrives is the machine that was quoted and the resale listing in 2032 can say so. Second, the placed-in-service date, because the $14,909 to $26,268 of year-one tax shield depends on the machine running before 31 December, not on the invoice date. Both are lines the distributor will accept without argument and both are missing from most machine-tool POs we have read. If you bought or sold a VF-2 in the last three years, put the price into the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) form. It takes two minutes, the price is never shown, and the cell publishes only when five contributors and eight quotes are in. Sixty-three listings and three prices is the state of the used market today, and it will stay that way until buyers, not dealers, supply the number. Behind the paywall: the configuration that minimizes ten-year cost, which is not the SS and not the used 2018 that looks **$11,769 cheaper** on paper; the five checks that separate a $50,000 used VF-2 from a $50,000 problem; and what the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) can and cannot say about machine tools this year. --- # What 2,439 Public Change Orders Show About Electrical and Mechanical Contractors URL: https://open-factory.com/articles/change-orders-212-invoices/ Date: 2026-03-04 Procurement / Deep dive Median 1.8% of award, 9.5% of the dollars, 69% of mods give no reason, five line items that grow, labor units vs billed hours, clause language and contractor ratios The median electrical or mechanical contract in our set of 1,096 federal awards closed 1.8% above its award price. The dollars closed 9.5% above it. That gap between the typical contract and the typical dollar is the whole story of change orders: most jobs finish close to the bid, and a quarter of them finish more than 10% over, and the quarter carries the money. We pulled [2,439 priced change orders](data/change-orders.csv) from two kinds of public record: 2,417 modifications to federal construction contracts coded to electrical (NAICS 238210) and plumbing, heating and air-conditioning (238220) contractors on [USAspending.gov](https://api.usaspending.gov), and 22 change orders that city councils and utility boards in Fort Worth, Riverside, Kansas City, Sacramento, El Paso and San Antonio approved with the memo attached. This piece sizes the change-order load by project type, size and buyer, sorts the reasons that recur, names the five line items that always grow, checks billed hours against a published labor unit, and ends with the contract language and the contractor ratio table. The [Open Factory Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds the dataset: for every contract the award, the contractor, each change order's date, value and text, and the URL of the public record. Of the [1,096 contracts](data/change-orders-projects.csv), 383 (35%) closed with no priced change order, 125 finished 0 to 2% over, 148 finished 2 to 5% over, 126 finished 5 to 10% over, 131 finished 10 to 20% over, 106 finished 20 to 50% over, 33 more than doubled or came close to it, and 44 finished below award after deductive changes. ## What the 2,439 Records Are The federal rows come from the [USAspending API](https://api.usaspending.gov): every definitive contract (award type D) awarded in fiscal years 2019 to 2024 with a non-change-order value between $1M and $60M, coded to NAICS 238210 or 238220. That returned 1,162 awards, of which 1,096 fit the value window; for each we pulled every modification. A change order is a modification whose FPDS action type is `CHANGE ORDER`, `SUPPLEMENTAL AGREEMENT FOR WORK WITHIN SCOPE`, `DEFINITIZE CHANGE ORDER` or `ADDITIONAL WORK`, with a non-zero dollar value, dated on or before March 4, 2026, whose text is not a close-out, de-obligation or option exercise. Deducts count. Method and keyword rules: [research/notes.md](research/notes.md). The set covers [51 federal buyers and 486 contractors](data/change-orders.csv) in 54 states and territories. Four buyers dominate: the Army (263 contracts), Veterans Affairs (242), the Air Force (150) and the Navy (74), then GSA's Public Buildings Service (58), the Bureau of Prisons (45) and the FAA (33). Pricing is almost all firm fixed price: 1,088 of 1,096 contracts, with four time-and-materials and two cost-no-fee awards. On a fixed-price job every dollar in this file is a negotiated modification, not an hours ticket. The municipal rows are the opposite kind of evidence: few, each with a paragraph explaining why. Riverside's Board of Public Utilities has to approve any change order that takes a contract past 10% of its original price under [Purchasing Resolution 24101, Section 1104](https://riversideca.legistar1.com/riversideca/attachments/93eed94b-cd49-4947-bf83-dc96775f6560.docx), so every overrun above 10% gets a memo. Fort Worth's council communications list each prior change order with its cause; El Paso's coversheet has checkboxes for differing site condition, user request, design error and quantity adjustment, and prints the contractor's first proposal against the negotiated figure. Those 13 contracts carry [22 change orders](data/06-municipal-cases.csv) and every reason category we use. ## The Median Is 1.8%. The Dollars Are 9.5% The [1,096 federal contracts](data/change-orders-projects.csv) had a combined non-change-order value of $3.54B and collected $338.0M of net change orders: $352.9M additive across 2,263 modifications and $15.0M deductive across 175. The median contract's net change was 1.8% of award; the mean was 8.8%; the 75th percentile contract sat at 9.9% and the 90th at 23.7%. Among the 715 contracts that had at least one priced change order, the median net change was 6.6% and the median count was two. The median single change order was $45,619, or 1.37% of its contract. **The electrical trade runs lighter than the mechanical trade on every measure.** The 445 [NAICS 238210](data/change-orders.csv) contracts (base $1.49B) carried a median 1.4% and a dollar-weighted 7.4% ($109.4M), and 22% of them ended more than 10% over. The 651 mechanical contracts (base $2.06B) carried a median 2.3% and a dollar-weighted 11.1% ($228.6M), and 26% ended more than 10% over. The median electrical change order was $41,644. A 10% contingency covers the 75th percentile electrical contract with room; on a mechanical package it covers the 75th percentile exactly. Contracts of [$1M to $3M](data/03-change-by-size.csv) (n=728) had a median net change of 1.0%, a dollar-weighted 8.3% and 1.45 change orders each, with a median change order of $38,065; $3M to $10M (n=332) had a median 3.7%, dollar-weighted 10.6%, 3.2 change orders and a median change order of $45,619; $10M to $60M (n=36) had a median 3.9%, dollar-weighted 9.5%, 8.3 change orders and a median change order of $84,084. Size buys more change orders, not a bigger share. The 9.5% dollar-weighted figure, not the 1.8% median, is the change-order allowance to put against the electrical and mechanical lines of the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/): it is the number the money follows. ## Change Orders by Project Type and Size We classified each contract from its award description into seven types. Generator and backup-power jobs carried the highest median, HVAC the second, and lighting the lowest. [Generator and backup power](data/02-change-by-project-type.csv) contracts (n=39) had a median net change of 2.6% and a 75th percentile of 9.7%; HVAC and mechanical (n=577) 2.3% and 10.7%; substation and MV distribution (n=69) 2.2% and 9.5%; fire alarm, security and communications (n=91) 1.6% and 11.4%; controls and SCADA (n=69) 1.6% and 10.6%, with 72.5% of them carrying at least one change order, the highest share of any type; general electrical renovation (n=197) 1.4% and 9.0%; lighting (n=54) 0.8% and 4.3%. Dollar-weighted, fire alarm and security ran 11.5%, HVAC 10.9%, controls 10.8%, general electrical 7.8%, substation 6.9%, generators 5.0% and lighting 4.7%. Lighting is the one electrical package where a 5% contingency is enough: the scope is countable from the drawings. Controls and SCADA is the package where a change order is almost certain even when small: seven in ten of those contracts were modified, because integration scope (points lists, graphics, network drops, the PLC another project did not finish) is written after award. Fort Worth's [SCADA replacement](https://fortworthgov.legistar1.com/fortworthgov/attachments/4f8ffa35-5044-4575-920d-8728dc01d535.pdf) is the type specimen: Change Order 1, $125,732.39 on an $8,233,836 contract, for "one additional Category, a PLC that was not part of the original scope of work ... from a project that was not complete at the time this project was bid." Buyer matters as much as type. Among [agencies with 15 or more contracts](data/11-buyers.csv), the National Institutes of Health carried a median 9.9% (n=15, 80% of contracts modified, 20.2% dollar-weighted), the FAA 6.5% (n=33, 88% modified), the Agricultural Research Service 4.4%, Veterans Affairs 3.7% (n=242, 77% modified, 10.9% dollar-weighted), the National Park Service 3.0%, GSA's Public Buildings Service 2.4% (11.2% dollar-weighted), the Bureau of Prisons 1.5% and the Army 1.1% (n=263, 62% modified, but 10.1% dollar-weighted). State, the Air Force (n=150), the Navy (n=74) and Customs and Border Protection had a median of zero; fewer than 55% of their contracts were modified at all. The [buyer table](data/11-buyers.csv) says the owner's habits set the rate before the contractor arrives. NIH's [central utility plant chiller repairs](https://www.usaspending.gov/award/CONT_AWD_75N99023C00048_7529_-NONE-_-NONE-) went from $1,766,165 to $4,016,661 (127%) through in-scope supplemental agreements; the Air Force's 150 contracts ran 4.6% dollar-weighted with a median of zero. Same trades, often the same contractors, different owners. The single-contract extremes are owner stories more than contractor stories. SES Electrical LLC's [Fort Bliss hospital renovation design-build](https://www.usaspending.gov/award/CONT_AWD_W9127S20C6001_9700_-NONE-_-NONE-) went from $2,305,533 to $8,426,580 (265%) through 17 modifications; Precision Mechanical's [Launch Complex 39B environmental control system](https://www.usaspending.gov/award/CONT_AWD_80KSC020C0017_8000_-NONE-_-NONE-) went from $7,368,751 to $19,158,198 (160%) through 31; Midlands Mechanical's [Denver Mint air handler replacement](https://www.usaspending.gov/award/CONT_AWD_2031JG22C00013_2044_-NONE-_-NONE-) added a $4,339,862 "Phase II" to a $3,883,087 contract in one modification. Read the count and the text before the ratio: 31 supplemental agreements on a launch-pad HVAC job is a design that was not finished at award, and the price at the end is the price the design should have carried at the start. ## Who Writes the Reason Down **Of [2,439 change orders](data/04-reason-categories.csv), 1,693 (69%) have no reason at all**: the modification text repeats the project title, or reads "MOD 3", or names a CLIN. That is a finding about federal contracting officers, not about contractors, and it is why the municipal records matter out of proportion to their count. The federal changes clause, [FAR 52.243-4](https://www.acquisition.gov/far/52.243-4), requires the contractor to "assert its right to an adjustment ... within 30 days" of a written change order and promises an "equitable adjustment". It does not require the reason to be recorded or say how the adjustment is priced. Where a reason is written, the distribution is stable across buyers. Of the [746 change orders with a stated reason](data/04-reason-categories.csv), 461 (61.8%) were the owner adding or changing scope ($71.5M additive), 93 (12.5%) were unforeseen or differing site conditions ($18.7M), 68 (9.1%) were design errors, omissions or RFI resolutions ($6.6M), 60 (8.0%) were delay, suspension or time-related costs ($9.1M), 42 (5.6%) were deductive credits (net -$1.5M), 11 (1.5%) were code or authority-having-jurisdiction changes, 9 (1.2%) were price escalation and 2 were utility company relocations. The categories are ours ([build_dataset.py](research/scripts/build_dataset.py)); the text is the government's. Escalation is smaller than the trade press would suggest. Nine federal modifications in seven years said "price increase" in so many words, the largest a [$113,840.55 increase](https://www.usaspending.gov/award/CONT_AWD_36C25622C0011_3600_-NONE-_-NONE-) (7.9%) on a VA access-control contract in April 2022 and a [$23,983.30 increase](https://www.usaspending.gov/award/CONT_AWD_47PJ0023C0013_4740_-NONE-_-NONE-) (0.7%) on Sturgeon Electric's Denver Federal Center substation transformer in July 2023. Firm-fixed-price federal work has no escalation clause, so the copper and steel moves of 2021 to 2025 landed inside the bid or the contractor's margin. The municipal record shows where they landed when the contract allowed it: Riverside's [Daikin chiller](https://riversideca.legistar1.com/riversideca/attachments/295d684d-0964-483a-a053-40a47922c5b3.docx) ordered in April 2021 at $173,845 was repriced to $247,833 (42.6%) in November 2022 because the vendor "was unable to place the order with the Daikin factory at the previously quoted ... pricing", and Riverside's [69 kV steel poles](https://riversideca.legistar1.com/riversideca/attachments/5a940cbf-f077-48cd-8486-045c6d0104d7.docx) carried an estimated $222,704 (33.4%) after the fabrication date slipped from June 2021 to January 2022 "which has subjected RPU to rising steel prices". The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) exists so that an escalation clause names an index and a base date instead of a vendor letter. ## The Reasons That Recur Unforeseen conditions are 12.5% of stated reasons and 17% of the stated additive dollars, and the text tells you what they are. RHI LLC's steam condensate replacement at the Tomah VA Medical Center took a [$1,714,015 modification](https://www.usaspending.gov/award/CONT_AWD_36C25220C0170_3600_-NONE-_-NONE-) (42.4% of a $4.04M contract) in September 2022 for "extra asbestos abatement, plywood and plastic safety barriers", after a $561,854 modification three months earlier for "corrections for design oversight, unknown existing conditions". Brice Builders' Barter Island job took [$1,675,512.80](https://www.usaspending.gov/award/CONT_AWD_W911KB20C0026_9700_-NONE-_-NONE-) (23.0%) for "differing site conditions" in December 2021, sixteen months after a $921,278 descope. C.D. Whitfield's chiller renovation at the DFAS Supply Center in Columbus took [$800,000 and then $858,541](https://www.usaspending.gov/award/CONT_AWD_W912QR21C0039_9700_-NONE-_-NONE-) (32.8% and 35.2% of a $2.44M contract) in 2023 and 2024 for "replace duct and mold mitigation". Asbestos, mold and what is behind the wall are the unforeseen conditions of renovation; on greenfield work it is soil and utilities. The governing clause is [FAR 52.236-2](https://www.acquisition.gov/far/52.236-2): "subsurface or latent physical conditions at the site which differ materially from those indicated in this contract", with notice required "before the conditions are disturbed". Design errors and omissions are 9.1% of stated reasons and the category owners most under-count, because the text rarely says "error". It says "correct quantity of CLIN 1001", as in Lanotte Refrigeration's [$446,548 modification](https://www.usaspending.gov/award/CONT_AWD_12905B22C0011_12H2_-NONE-_-NONE-) (43.4%) for the Agricultural Research Service, or "modification to address RFIs 06, 16, 17 and 18", as in B2 Constructors' [$257,333](https://www.usaspending.gov/award/CONT_AWD_36C24923C0109_3600_-NONE-_-NONE-) (9.7%) for a VA energy management system. Riverside's version is the clearest on the record: on the [Limestone Drive cable replacement](https://riversideca.legistar1.com/riversideca/attachments/93eed94b-cd49-4947-bf83-dc96775f6560.docx), a cap-and-grind step on slurry-seal streets "had not been accounted for during the design phase and was not included in the scope of work when the project was bid. This oversight resulted in a doubling of the unit price for cap-and-grind work", and with 80 tons of asphalt for pavement delamination the change order came to $115,484 on a $704,840 contract (16.4%). Owner scope is the majority and it is not a scandal. Fort Worth's [athletic field lighting](https://fortworthgov.legistar1.com/fortworthgov/attachments/4dff6ab0-870e-4c67-afa1-3539b33507c9.pdf) came in $498,690 under budget and the city spent $368,000 of the savings on a seventh ball field with Change Order 1 (20.1%). Kansas City's [Richards Road electrical upgrade](https://kansascity.legistar1.com/kansascity/attachments/003bf739-de8f-48c8-8b31-dc07368f9c1f.pdf) added a new S&C switch with micro switching to feed two hangar transformers for $308,174 (25.4% on a $1,214,435 contract) in November 2022, and the form notes "current lead time for this equipment is 42 to 44 weeks, contract time will be increased at a later date". Fort Worth Water's [$23.1M generator project](https://fortworthgov.legistar1.com/fortworthgov/attachments/bd7d8e5c-a6a2-42d0-8354-1ed5f5380ee5.pdf) relocated a diesel generator from one fence line to another, rerouted the duct banks, and replaced an RTU and PLC for SCADA compatibility in one $221,671.33 change order. These are decisions priced after award. The buyer's only question is whether they were priced at bid rates or change-order rates. ## The Five Line Items That Always Grow We tagged each change order with the [line item its text names](data/05-line-items.csv). Across the whole set the additive dollars go to HVAC equipment and piping first ($56.0M across 346 change orders), then controls, SCADA and PLC ($13.0M, 107), panels, switchgear and transformers ($9.8M, 112), fire alarm, security and communications ($9.8M, 64), conduit, duct bank and raceway ($8.8M, 61), lighting ($8.3M, 52), civil and finishes ($6.0M, 58), feeders, wire and cable ($5.5M, 49), generators and backup power ($5.1M, 42), time and general conditions ($4.1M, 34) and hazmat abatement ($2.6M, 12). Restrict to the [853 electrical-contractor rows](data/change-orders.csv) and the ranking becomes the list a plant buyer should carry into every electrical negotiation. Of the 334 additive electrical change orders that name a line item, panels, switchgear and transformers appear in 70 ($5.64M, 16.9% of the named dollars), feeders, wire and cable in 37 ($4.84M, 14.5%), fire alarm, security and communications in 50 ($4.63M, 13.9%), generators, ATS and UPS in 33 ($4.58M, 13.7%) and conduit, duct bank and raceway in 39 ($2.66M, 8.0%). **Those five items are 229 of the 334 named electrical change orders and 67% of the named electrical dollars.** Lighting, the item every owner expects to grow, is 20 and $2.59M. Each of the five grows for a reason that can be written into the bid. Switchgear and panels grow because the one-line is redrawn after award: M.C. Dean's Buckley plant recapitalization carries a [$146,849.44 modification](https://www.usaspending.gov/award/CONT_AWD_W912DR21C0023_9700_-NONE-_-NONE-) titled "XFMR AND PANEL ONE-LINE UPDATES" and a $36,170 one for "PANEL UPSIZING". Feeders grow because sizes and set counts change: the same contract has a $100,562.50 credit for "MV FEEDER CABLES SIZES & SETS" and a $59,017 add for a "BUILDING 201 NEUTRAL FEEDER". Fire alarm and low-voltage grow because the AHJ reads the drawings last: Broadway Electric's Cleveland VAMC fire alarm upgrade lists "ALT CONDUIT RUN TO COMPLY WITH NFPA" among [eight modifications](https://www.usaspending.gov/award/CONT_AWD_36C25022C0032_3600_-NONE-_-NONE-). Generators grow because the paralleling and fuel scope is discovered at submittal: DLJ Management's [$582,943 modification](https://www.usaspending.gov/award/CONT_AWD_W912PL21C0026_9700_-NONE-_-NONE-) (15.9%) to "replace additional switches" added 42 calendar days. Conduit grows because quantities are estimated from plan and paid from the field: Riverside's Doty Bros. job found an [unmarked cable-television conduit](https://riversideca.legistar1.com/riversideca/attachments/908d202e-c497-45e1-9216-1e5277cae11f.docx) at the new vault, and El Paso's [Change Order 13](https://elpasotexas.legistar1.com/elpasotexas/attachments/184e300d-7b90-493e-beba-d570216d7c81.pdf) added 9,515 linear feet of conductor and 1,495 feet of 3-inch PVC and bores when El Paso Electric set its service points "far away from the anticipated location" at eight stations. [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/) walks the one-line clauses that decide the first of the five. The [2 MW plant service upgrade](https://open-factory.com/articles/2mw-plant-service-upgrade/) shows the utility service point deciding the fifth. ## When the Change Orders Land Change orders are not front-loaded. Of the [$352.9M of additive federal change orders](data/09-co-timing.csv), 8.0% were signed in the first six months after award, 19.6% in months 6 to 12, 19.1% in months 12 to 18, 17.8% in months 18 to 24, 18.6% in months 24 to 36, 9.9% in months 36 to 48 and 6.9% (60 change orders, $24.3M) more than four years after award. The cumulative line crosses 50% in the 12 to 18 month band and 83% at 36 months, so the contingency has to survive past substantial completion: what arrives after month 36 is delay cost, suspension lifts and the second year of off-site storage. Greenland Enterprises' [$774,560.75 modification](https://www.usaspending.gov/award/CONT_AWD_36C10F21C0009_3600_-NONE-_-NONE-) (8.4%) in December 2024 is titled "delay costs associated with modification nos. P00032 and P00034"; Spees-Hernandez JV's [$999,439](https://www.usaspending.gov/award/CONT_AWD_36C26322C0017_3600_-NONE-_-NONE-) (14.2%) in January 2026 is "2nd year of offsite storage required due to current work suspension". Equipment lead time turns a scope change into a time change. The [Lead-Time Monitor's](https://open-factory.com/models/lead-time-monitor/) published figures for switchgear and generators belong in the schedule clause before award, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) puts them on a calendar. ## Thirteen Contracts With the Memo Attached The council and board records are few, and they are the only place the reason, the markup and the contractor's first price appear together. | Buyer | Contractor | Original | Change orders | Net change | % of award | Recorded reason | |---|---|---|---|---|---|---| | Riverside General Services | NSWC Mechanical Service | $173,845 | 1 | $73,988 | 42.6% | Factory repricing, steel and copper | | San Antonio via J3 Company | SAECO Electric & Utility | $331,403 | 1 | $126,531 | 38.2% | CPS duct bank redesigned 4-way to 9-way | | Riverside Public Utilities | Wireless Structures Consulting | $666,680 | 1 | $222,704 | 33.4% | Design changes, steel escalation (estimate) | | Kansas City Aviation | Citadel Electric Group | $1,214,435 | 2 | $322,494 | 26.6% | Owner-added S&C switch, 42 to 44 week lead time | | Sacramento Public Works | David Engineering Construction | $223,000 | 2 | $57,731 | 25.9% | Paving added; ADA van access | | Fort Worth Park and Recreation | Groves Electrical Service | $1,832,140 | 1 | $368,000 | 20.1% | Seventh ball field funded from bid savings | | Riverside Public Utilities | Asplundh Construction | $704,840 | 1 | $115,484 | 16.4% | Cap-and-grind omitted from design; delamination | | Riverside Public Utilities | Doty Bros. Equipment | $118,780 | 2 | $13,043 | 11.0% | Paving discrepancy; unmarked CATV conduit | | Riverside General Services | Stilwell Construction | $378,707 | 2 | $41,421 | 10.9% | Structural redesign at plan check; ADA drawings | | El Paso Capital Improvements | Venegas (VEMAC) | $19,583,784 | 13 | $462,323 | 2.4% | CO 13: El Paso Electric moved service points | | Fort Worth Water | Red Group | $8,233,836 | 1 | $125,732 | 1.5% | PLC from an unfinished adjacent project | | Fort Worth Water | ACME Electric (Winston Electric) | $23,119,632 | 3 | $321,399 | 1.4% | Generator relocation, fence, insulation | | Fort Worth Water | ACME Electric (Winston Electric) | $8,647,846 | 3 | $104,941 | 1.2% | Pump starters, cables, deleted control panels | [CSV: data/06-municipal-cases.csv](data/06-municipal-cases.csv) The [13 contracts](data/06-municipal-cases.csv) split the same way the federal set does: the three above 30% are equipment-heavy contracts under $700,000 where the change was escalation or a redesign, and the four below 2.5% are contracts above $8M with a full-time inspector and a memo for every change. The El Paso coversheet is the model form. It records that the contractor's original proposal on new items was $31,656.69, that the negotiated amount was $134,309.49 after quantities were reconciled to the field, that an independent cost estimate was prepared, and that pricing used "unit bid prices previously approved" where they existed ($4.04 per linear foot for #4 AWG, $11.76 for #3/0, $10,077.94 per overhead service point) and negotiated unit prices where they did not ($6.24 for #2, $8.90 for #2/0, $10.48 for #4/0, $19.35 per foot of 3-inch PVC, $32.57 per foot bored). A unit-price schedule turns a change order into arithmetic. ## Labor Units Against Billed Hours Every change order on a fixed-price contract is priced as if it were time and materials: hours times rate plus material plus markup, and the hours are the argument. Contractors price them from a labor-unit manual. The one contracts reference is NECA's Manual of Labor Units, which [NECA](https://www.necanet.org) sells and does not publish. The public proxy is Craftsman's [National Electrical Estimator](https://www.craftsman-book.com/media/static/previews/2025_NEE_book_preview.pdf), whose 2025 preview pages print the units and the method: "The Craft@Hrs column shows the recommended crew and manhours per unit ... Costs in the Labor Cost column are the result of multiplying the manhours per unit by the rate of $46.59 per hour." Craftsman's [units](data/08-craftsman-labor-units.csv) run 3.25 hours per 100 ft of 1/2-inch EMT in slab or on a trapeze, 3.50 for 3/4-inch, 4.00 for 1-inch, 4.50 for 1-1/4, 5.50 for 1-1/2, 7.00 for 2-inch, 9.00 for 2-1/2, 10.0 for 3-inch, 11.0 for 3-1/2 and 12.0 for 4-inch; concealed in walls and ceilings adds 0.25 to 0.5 hours per 100 ft on the small sizes. Device covers are 0.05 to 0.10 hours each. The check is mechanical: quantity times unit times a difficulty factor equals hours, and the factor has to be named (occupied hospital, overhead in a live switchgear room, night shift). The only itemised hours in our municipal set are Stilwell Construction's two [Cost Change Estimate Breakdowns](https://riversideca.legistar1.com/riversideca/attachments/68d52c13-1233-4878-bbc8-8f2f28f53e95.pdf) for Riverside: 3 hours at $109 for field measurement and ADA drawings, and 64 hours at $109 (16 for engineering coordination, 8 for potholing, 24 for sawcutting and removing asphalt at four foundations, 16 for drypack and patching) plus $29,817 of subcontracts and a 15% markup for a structural redesign. None of those tasks has a unit in a conduit table, which is the general case: the hours that get argued about are the ones the manual does not cover. Federal records never show hours at all. The rate is the second argument, and it is public. Stilwell billed $109 an hour in Riverside in 2025. The May 2025 median electrician wage was [$30.38 an hour nationally](https://open-factory.com/articles/electricians-wages-by-metro/), $28.37 in Dallas-Fort Worth and $49.21 in Chicago, so a $109 change-order rate carries a 2.2x to 3.8x load of burden, overhead and profit over the median wage, before the markup line. A contract that states the change-order labor rate by classification, and the burden it includes, removes the second argument. The third argument is the markup, and five public forms show the range. | Record | Labor | Equipment | Materials | Subcontracts | Bond | Method | |---|---|---|---|---|---|---| | J3 Company, San Antonio, May 2019 | 55% burden + 25% markup | 15% | 25% | 5% | 2% | Lump sum built from the sub's quote | | Stilwell Construction, Riverside, RFC 2, Feb 2025 | $109/hr, 15% on total | 15% | 15% | 15% | none listed | T&M estimate | | Stilwell Construction, Riverside, RFC 1, Jan 2025 | $109/hr, 10% on total | n/a | n/a | 10% | none listed | T&M estimate | | Venegas (VEMAC), El Paso, CO 13, Dec 2021 | in unit prices | in unit prices | in unit prices | n/a | n/a | Bid unit prices plus negotiated new items | | Citadel Electric, Kansas City, CO 5-R1, Nov 2022 | in lump sum | in lump sum | in lump sum | in lump sum | n/a | Lump sum, all impact claims released | [CSV: data/07-markup-schedules.csv](data/07-markup-schedules.csv) The [J3 form](https://legistar.granicus.com/sanantonio/attachments/01427b1f-624e-4573-8447-bac6524dbae9.pdf) is the one to study: 25% on materials, 15% on equipment, 5% on the $126,530.63 subcontract and 2% bond produced $8,983.67 of markup on a $135,514.30 change, and the unused labor line is printed at 80%, a 55% burden plus 25% markup. A markup schedule in the contract, one rate on labor and material and a lower one on subcontracts, settles that line before the first change order. The paid section below writes it out. ## What to Do Monday Set the contingency from the dollars, not the median. On a fixed-price electrical package budget 7 to 8% for change orders and on a mechanical package 11%, which is what the $3.54B of federal work in the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) actually consumed, and hold the last third of it past substantial completion, because 17% of the additive dollars arrive after month 36. Then write four things into the RFQ before the bid opens: a unit-price schedule for the five items that grow (switchgear and panel modifications, feeder conductors by size, fire alarm and low-voltage devices, generator and ATS accessories, conduit and duct bank by size and installation type), a stated change-order labor rate by classification with the burden named, a markup schedule with one rate on labor and material and a lower one on subcontracts, and a T&M cap that converts to lump sum at a threshold. The [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/) piece has the template. The [Quote Check](https://open-factory.com/tools/quote-check/) tool prices the equipment lines against the Bid-Tab bands so the base bid is right before the changes start. Behind the paywall: the model language for allowances, unit-price schedules, T&M caps and the markup schedule, and the [Bid-Tab Price Book's](https://open-factory.com/models/bid-tab-price-book/) ranking of the 91 contractors with three or more federal contracts in the set, from **eight contractors that closed $55.8M of work with zero priced change orders** to the one that closed 51.6% over. --- # Copper at $6.77: How a 50% Tariff Shows Up in a 200-Amp Panel URL: https://open-factory.com/articles/copper-tariff-panel/ Date: 2026-02-14 Procurement / Take COMEX record, $1,000 per ton premium over LME, copper wire PPI +17.9%, busway and cable quotes, aluminum substitution, the escalator clause COMEX copper settled at a record [$6.77 per pound on August 7, 2026](https://investingnews.com/daily/resource-investing/base-metals-investing/copper-investing/highest-price-for-copper/), the day after the Democratic Republic of Congo banned concentrate exports. The 50% Section 232 copper tariff has been in force since August 1, 2025, and it does not apply to the cathode that price is quoted on. It shows up anyway. **A 200-amp residential service that carried $270 of copper metal at 2024 prices carries $440 today, and about $30 of that is the tariff expectation priced into COMEX rather than the metal itself.** This piece weighs the panel, traces the price from cathode to distributor counter, and shows what the number means for a 400-amp commercial feeder. The global monthly average was [$4.15 per pound across 2024](https://fred.stlouisfed.org/series/PCOPPUSDM), $5.35 in December 2025 and $6.14 in July 2026, 48% above the 2024 average. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) converts that series to $/ft by conductor size weekly and carries the copper wire PPI beside it. The [Lead-Time Monitor's published lead time for LV switchboards is 32 to 41 weeks](https://open-factory.com/models/lead-time-monitor/) (Siemens schedule as republished, Aug 2026), which is why a copper quote from today is installed against a price from next spring. ## The Tariff That Is Not on the Metal [Proclamation 10962 of July 30, 2025](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states) imposed a 50% duty on "semi-finished copper products and intensive copper derivative products" entered from 12:01 a.m. on August 1, 2025, and left refined copper, ores, concentrates and scrap out. Commerce had recommended a phased universal tariff on refined copper of 15% from 2027 and 30% from 2028. The President did not adopt it and instead ordered an update on the domestic market by June 30, 2026. The market had priced a 50% tariff on everything, and when the exemption landed the COMEX premium over LME collapsed from [about $2,937 a tonne in late July 2025](https://think.ing.com/articles/whats-next-for-us-copper-import-tariffs/) to an [average of $130 a tonne through mid-September](https://www.metalminer.com/2025/09/16/comex-price-of-copper-holds-premium/), against an $18 average before 2025. The spread was $110 a tonne on [April 10, 2026](https://www.mining.com/web/copper-traders-get-deja-vu-as-comex-arbitrage-reopens/), about $400 in [early June](https://think.ing.com/articles/whats-next-for-us-copper-import-tariffs/), nearly $600 on [July 21](https://www.mining.com/copper-price-nearing-us-record-as-premium-doubles-shanghai-stockpiles-drop-82/), and the December 2026 contract traded [close to $1,000 a tonne over LME by mid-August](https://www.cnbc.com/2026/08/14/copper-trump-tariffs-metal-commodities-trade-war.html). Societe Generale reads that premium as a 14.6% probability of the 15% refined-copper tariff by January 2027 and a 37% probability of 30% by January 2028. The US imported more than 200,000 tonnes of copper in July 2026, the most in twelve years, to get ahead of it. **US building-wire pricing keys off COMEX, so a $1,000 premium is $0.45 a pound on every conductor sold in the country, tariff or no tariff.** The June 30 deadline passed with no refined-copper action. Border States' July 13 commodity update says so plainly and reports that [copper suppliers announced a 7.5% price increase on July 10, 2026](https://tedmag.com/border-states-commodity-update-july-2026/) with COMEX at $6.28. What did change is the derivative regime: [Proclamation 11021 of April 2, 2026](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states) applies the Section 232 duty to the full customs value "regardless of metal content" from April 6, 2026, at 50% for the copper articles in Annex I-A and 25% for the derivatives in Annex I-B, with 10% where the copper was smelted and cast in the US. A [June 4, 2026 proclamation](https://www.federalregister.gov/documents/2026/06/04/2026-11314/further-adjusting-the-tariff-regimes-for-imports-of-aluminum-steel-and-copper-into-the-united-states) cut the US-content threshold from 95% to 85% from June 8; ING reads the same instrument as extending the copper scope to electrical conductors and cables, which we could not confirm in the published annexes and treat as unverified. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) includes each of these instruments by HTS line with its Federal Register citation, and the [Tariff Bill of Materials](https://open-factory.com/articles/tariff-bom-electrical/) (October 2025) worked the 50%-on-content rule through a UPS, a switchgear lineup and a Megapack before the April rewrite. ## What a Pound of Copper Becomes in Wire The BLS producer price index for copper wire and cable, [WPU10260314](https://fred.stlouisfed.org/series/WPU10260314), stood at 578.2 in July 2026, up 17.9% on July 2025 and 95% on December 2019. It ran 23.6% year over year in June. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) lists this series because wire is the one electrical commodity that reprices weekly. The [Lead-Time Monitor launch note](https://open-factory.com/articles/lead-time-monitor-launch/) (September 2025) set out the transformer and switchgear PPI series the [Monitor](https://open-factory.com/models/lead-time-monitor/) includes beside the Schneider and Siemens schedules. The ISM has recorded copper "up in price" for [14 consecutive months through August 2026](https://www.textileworld.com/textile-world/2026/09/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-manufacturing-expanded-for-the-eighth-consecutive-month/) and electrical components "in short supply" for the same 14. A 2/0 AWG copper conductor weighs about 411 lb per thousand feet bare, so its metal is worth $1.71 a foot at $4.15, $2.30 at $5.60 and $2.78 at $6.77; 600 kcmil is 1,852 lb per thousand feet and moves from $7.69 to $12.54 over the same range. Those are metal values, and finished wire sells for well above them. Buildermuse's April 2026 sheet put [12 AWG THHN at $298 per thousand feet with copper at $6.01](https://buildermuse.com/economy/electrical-material-prices-2026/); 12 AWG carries 20.2 lb of copper, or $121 of metal, so the cathode was 41% of the counter price and drawing, insulation, mill margin and distributor margin were the other 59%. The same sheet claims 12 AWG is "~90% copper by cost", which its own numbers contradict. **The smaller the conductor, the less the copper price explains; on 12 AWG the metal is two fifths of the price, on 600 kcmil it is most of it.** Distributor margin on wire is disclosed, if you know where to look. Graybar's 2025 10-K reports a [gross margin rate of 19.3%, down from 20.1% in 2024, "primarily due to competitive pricing pressures"](https://www.sec.gov/Archives/edgar/data/205402/000020540226000015/c402-20251231x10k.htm). WESCO's cost of goods sold was [78.9% of net sales in 2025 against 78.4% in 2024](https://www.sec.gov/Archives/edgar/data/929008/000092900826000008/wcc-20251231.htm), a 21.1% gross margin. Those are blended rates; wire runs thinner than lighting or gear, and our [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) will publish wire awards by conductor size once the cell rules are met, as it does for pad-mounts and switchgear sections today. ## The 200-Amp Panel, Weighed NEC 310.12 lets a single-phase dwelling service use conductors rated at 83% of the service rating, so a 200 A service needs 166 A of ampacity: 2/0 copper at 175 A in the 75°C column, or 4/0 aluminum at 180 A, per the [NEC 310.16 values Cerrowire republishes](https://www.cerrowire.com/products/resources/tables-calculators/ampacity-charts/). We modeled two ungrounded conductors and a full-size neutral in 2/0 copper over a 50 ft run from meter to panel, plus 25 ft of #4 copper grounding electrode conductor. That is 65 lb of copper: $270 at the 2024 average, $364 at the end-2025 price and $440 at $6.77. Against a 200 A load center at $285 and twenty breakers at about $180 (Buildermuse, April 2026, vendor pricing), the conductor copper is 37% of the material bill at 2024 prices and 49% at the August 2026 record, and the whole material bill rose from $734 to $904, or 23%. The load center itself carries bus bars and lugs we did not weigh; Buildermuse puts a 200 A load center at 40 to 60% copper by weight and its price up 12.2% year over year, so the true copper share is higher than the chart shows. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) prints the copper pounds for any conductor schedule, which is the one number the escalation clause below cannot work without. | Scenario | Copper $/lb | Conductor copper (65 lb) | Load center | Breakers (20) | Material total | Copper share | COMEX premium in the copper | |---|---|---|---|---|---|---|---| | 2024 average | $4.15 | $270 | $285 | $180 | $734 | 36.7% | $29 | | End 2025 | $5.60 | $364 | $285 | $180 | $828 | 43.9% | $29 | | Aug 7 2026 record | $6.77 | $440 | $285 | $180 | $904 | 48.6% | $29 | *Source: Open Factory Copper and Wire Price Sheet; [CSV](data/service-200a-material-copper-share.csv)* Here is what the 50% tariff itself does to that panel. At a $1,000 a tonne COMEX premium, $0.45 a pound, the tariff expectation is $29 of the $440 of copper in the service. The other $141 of the increase since 2024 is the world price: Grasberg, the DRC, sulphuric acid, datacenters. **On a residential 200 A service the tariff is a $30 line; the copper market is a $170 line; and the labor to install it, which we do not model here, is the largest line of all.** I read a lot of panel-upgrade quotes during the Mister Sparky diligence, and none of them itemized copper. They should now. ## Who Absorbs It: Mills, Distributors, Contractors Atkore, which buys copper rod from AmRod, SDI LaFarga and Nexans per its [FY2025 10-K](https://www.sec.gov/Archives/edgar/data/1666138/000162828025054049/atkr-20250930.htm), reported for the quarter ended June 26, 2026 that [average selling prices added $22.4 million while input costs added $48.9 million](https://www.sec.gov/Archives/edgar/data/1666138/000162828026051510/atkr3q26exhibit991.htm), taking gross margin from 23.4% to 22.2%. Two days later it [agreed to be acquired by Prysmian at $95.00 a share, an enterprise value of about $3.8 billion](https://www.sec.gov/Archives/edgar/data/1666138/000166613826000016/atkr_991xpressrelease.htm), a 30% premium to the July 31 close. Prysmian bought Encore Wire in 2024, so Encore's building wire and Atkore's conduit and armored cable will sit under one Italian parent. Southwire is private and publishes no filings. Its price letters reach the market through distributors, which is why the Border States bulletin is the best public record of a wire price move. Distributors pass copper through fast and absorb little. Border States reported a [7% to 10% increase on finished aluminum goods on March 9, 2026](https://tedmag.com/border-states-commodity-update-march-2026/) and the 7.5% copper increase on July 10; Graybar and WESCO both gave up 50 to 80 basis points of gross margin in 2025 on "competitive pricing pressures", not on copper. The [Rockwell take](https://open-factory.com/articles/rockwell-250bps/) (December 2025) found 100 basis points of tariff pass-through in PLC quotes; on wire the pass-through is closer to 100% because the mills reprice weekly against COMEX and the distributor's price sheet follows within days. The contractor is the buyer who holds a bid open for 45 to 120 days while the price moves. The [switchgear cost decomposition](https://open-factory.com/articles/switchgear-50-percent/) (November 2025) found the same pattern one tier up, with switchgear PPI up 14.5% year over year and lead times of 52 to 104 weeks. ## What to Do Monday Price wire off the sheet, not the bid. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) publishes COMEX and the $/ft metal value by conductor size weekly; put the date and the price in the bid and make the conductor line move with it. Buy conductors the week the permit issues; wire stores, and a 45-day gap between bid and purchase at the end-2025 to August 2026 slope cost about $0.10 a foot on 2/0. Check the gear lines of the electrical package against [Quote Check](https://open-factory.com/tools/quote-check/), which returns the Bid-Tab band and the Monitor's published lead time for a switchboard or transformer spec, with n and date. For a service over 400 A, run the aluminum case before the copper one, and run it with the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) open, because the aluminum Section 232 rate is 50% on full value and the US delivered price carries its own premium. Watch the [Monitor](https://open-factory.com/models/lead-time-monitor/) for LV switchboard lead times, because a 32-week switchboard turns a copper quote into a copper forecast, and contribute the wire lines from your last three quotes to the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/), which publishes nothing until a cell has five contributors and eight quotes. Behind the paywall: the escalation clause language we would sign, indexed to COMEX with a 5% deadband, and the aluminum math for a 400 A feeder, where **two parallel sets of 250 kcmil aluminum carry $473 of metal against $2,803 for the copper equivalent** at August 2026 prices, before conduit upsizing takes some of it back. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) runs both calculations on live prices. --- # The Used and Refurbished Market Nobody Indexes URL: https://open-factory.com/articles/used-refurbished-market/ Date: 2026-02-04 Procurement / Deep dive 15 bids for one 2,500 kVA pad-mount, used gensets at $47 to $335 per kW, eight oil and insulation limits, the Charlotte refurb line, dealer stock at 1.8x the median bid, five checks before wiring money Fifteen vendors bid on the same 2,500 kVA pad-mount transformer for Fayetteville's public utility in September 2025 and the unit prices ran from $41,200 to $101,253, a 2.5x spread on one spec sheet, with quoted lead times from 10 to 35 weeks. The dealer that will ship a 2,500 kVA pad-mount from stock this week lists it at $104,000, which is 1.77x the median of those fifteen bids. The premium is for weeks, not for iron, and nobody publishes the curve. We built the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) to carry the bids, and this piece adds the other side: 105 dated public listings, lots and bid lines for used and remanufactured transformers, diesel gensets and 15 kV breakers, the test numbers that decide whether a unit is worth the money, what Siemens Energy is doing in Charlotte, and the arithmetic for when paying 1.8x is the cheap option. KLLEN Group America bid $41,200 at 16 to 18 weeks, ABB-Hitachi through Wesco bid $58,871 at 17 to 26 weeks, ERMCO through Tarheel EMC bid $69,950 at 14 to 16 weeks, and Meta Power Solutions bid $101,253 at 18 to 22 weeks. The median is $58,871 and the mean $60,429. ## Fifteen Bids, One Transformer, a 2.5x Spread The [Fayetteville PWC tabulation](https://www.faypwc.com/wp-content/uploads/2025/09/PWC2526016-Bid-Tab-rv-9.19.25.pdf) is the single most useful public document we have found on what a new pad-mount costs in America, because it holds fifteen manufacturers pricing nine identical line items on the same day. The 2,500 kVA unit is 12,470GrdY/7,200 x 24,940GrdY/14,400 primary to 480Y/277 secondary, quantity three. The 1,500 kVA line drew bids from $29,880 (Action FMR through PulseMac, 120 to 180 days) to $77,012, median $37,979. The 500 kVA line ran $15,728 to $52,209, median $22,157. The 150 kVA line ran $10,808 to $34,885, median $14,199. Every one of those medians is a row in the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) and the full 100-cell transcription is in this article's data folder. **The spread inside one bid tab is wider than the spread between new and stock.** The high bid at 2,500 kVA is 2.46x the low bid. The in-stock dealer price is 1.77x the median. A buyer who only ever sees one quote cannot know which end of that range they are standing at, which is the whole reason the [Quote Check](https://open-factory.com/tools/quote-check/) tool exists: it returns the band and the n for a spec, not any number. The tab also carries the lead times the bidders wrote down, and they do not correlate with price. Calvary Energy quoted 10 weeks at $66,750; Hyundai's D2 Engineering quoted 31 to 35 weeks at $54,310; KLLEN quoted 16 to 18 weeks at the lowest price on the sheet. Speed was not what the expensive bidders were selling. Greenville Utilities Commission's tabs tell the same story two sizes down. In [August 2025](https://www.guc.com/sites/default/files/uploads/bid-tab-sheets/25-43-bid-tab.pdf) eleven vendors bid on ten 150 kVA pad-mounts, from Kllen at $10,101 and 14 weeks to Uttam at $24,845; Border States quoted 63 weeks. In [December 2025](https://www.guc.com/sites/default/files/uploads/bid-tab-sheets/25-80-bid-tab.pdf) the 300 kVA award went to Kllen at $11,763 and 18 weeks over AR Transformers at $12,200 and 14 weeks, and GUC's sheet awards on "lowest total ownership cost," which adds the utility's own loss evaluation to the unit price. One more line on that December sheet matters for this article: Maddox Transformers is listed as a bidder with the note "did not submit a complete bid." The largest public reseller of remanufactured transformers showed up at a utility bid and did not compete on the factory-order price. It competes on a different axis. ## What Stock Costs Against a Factory Order Maddox Industrial Transformer is the one dealer in this market that publishes a price list. Its web store carries one price for every voltage variant of each kVA size, in stock, [2,500 kVA at $104,000](https://store.maddox.com/products/2500-kva-padmount-transformer), 1,500 kVA at $74,100, 1,000 kVA at $63,700, 500 kVA at $39,761, 150 kVA at $31,841, with the caveat "Price and availability may change. Submit RFQ for current details." The pages do not say whether a given unit is new or remanufactured; Maddox [stocks both](https://www.maddox.com/products-and-services/padmount) and sells "same-day deliveries" from stock, remanufactures in 4 to 6 weeks, and builds new to order in under 20 weeks. So the web price is the price of a transformer that ships now, whatever its history. Per kVA, the fifteen-bidder medians fall from $94.7 at 150 kVA to $44.3 at 500 kVA and $23.5 at 2,500 kVA, while Maddox's stock prices fall from $212 to $79.5 to $41.6. The stock-to-median ratio is 2.24x at 150 kVA, 1.79x at 500 kVA, 1.69x at 750 kVA and 1.77x at 2,500 kVA. **The stock premium is the price of a queue position, and it holds at 1.7x to 2.2x across every size we can match.** That is a stronger claim than Maddox makes about itself. Its own 2023 article says remanufactured units offer "a cost reduction between 10% and 40% for the same transformer" against new, and that while "lead times on new, factory-built transformers are stretching out to more than 80 weeks, in-stock, remanufactured transformers can be made available as quickly as 1 to 4 weeks" ([Maddox, May 2023](https://www.maddox.com/resources/articles/beat-supply-chain-delays-recondition-transformers)). Both statements can be true at once if the "new" Maddox compares against is its own built-to-order price, not a fifteen-bidder utility tab. A utility buying 39 units of one size at a time with a loss-evaluated award gets a factory-order price nobody buying one unit will ever see. The [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool prices both ends so a buyer knows which one they are being quoted. The only place Maddox publishes a clean used-versus-new pair is dry-type. On its [600 D to 208Y/120 page](https://store.maddox.com/products/3-phase-600v-delta-208-y-120) a 150 kVA unit is $10,215 new and $6,640 used, 65% of new, and the ratio holds at 65% down to 15 kVA ($2,588 new, $1,682 used); on its 480 D pages used sits at 80% of new. Those are the only public used-to-new ratios for transformers we have found anywhere, and they are for indoor dry-type units that carry a five-year warranty and ship in "3-5 business days" ([Maddox](https://www.maddox.com/resources/articles/what-to-expect-when-buying-a-used-dry-type-transformer)). For liquid-filled pad-mounts, the market publishes one number per unit and lets the buyer guess what condition it is in. ## Used Gensets Are Priced by Model Year, Not Hours Generator dealers publish more prices than transformer dealers do, and the prices sort by model year with almost no regard for hours. We captured 27 dealer listings for 500 to 2,000 kW diesel gensets between June and September 2026 and 18 of them are priced, used, standby-rated, diesel and not rebuilt. A [2011 Cummins QSK60 2,000 kW with 667 hours](https://www.depco.com/diesel/cummins-qsk60-g6-2000kw-stand-by-generator-set-item-20827/) asks $669,900 at Depco, $335 per kW. A [2000 Cummins DQKC 2,000 kW with 393 hours](https://www.depco.com/ready-to-ship/cummins-qsk60-g6-2000kw-stand-by-generator-set-item-19469/), fewer hours on the same engine family, asks $349,900, $175 per kW. A [1998 Generac 2,000 kW with 58 hours](https://generatorsource.com/inventory/2000kw-generac-unit-091552/) asks $120,000 at Generator Source, $60 per kW. Fifty-eight hours in twenty-eight years is not a selling point to a buyer who knows what sitting does to seals, hoses and injectors. The market prices it as such. The four 2018 to 2019 units ask $216 to $307 per kW with a median of $237, the four 2011 to 2017 units $180 to $335 with a median of $239, and the eight units built 1996 to 2003 ask $47 to $175 with a median of $106, against $780 per kW for a new 2,500 kW CAT 3516C Tier 2 sitting in the same dealer's yard. **A dealer claim that low-hour used gensets "now command 60 to 75%" of new does not survive the listings. The newest used iron we can find asks 30 to 43% of new Tier 2 stock.** The 60 to 75% figure comes from [Central States Diesel Generators](https://csdieselgenerators.com/the-ai-data-center-boom-and-generator-shortage-lead-times-prices/), which also states that a tested used unit ships in 1 to 4 weeks against 52 weeks or more for a new 1,250 to 3,250 kW set. The ratio depends on what "new" means. Against Depco's [new 3516C at $1.95 million](https://www.depco.com/diesel/caterpillar-3516c-2500kw-stand-by-generator-set-item-19944/), the 2011 QSK60 is 43% and the 2015 QSK60 with 100 hours at $595,000 is 38%. Against a new Tier 4 Final unit, which our [genset piece from September 2025](https://open-factory.com/articles/gensets-sold-out-2028/) priced at more than $1.4 million for 2,000 kW, the same units are 34% and 30%. There is no public $/kW series for new gensets, so every used ratio is a ratio to a quote, and the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool asks for the quote rather than assuming one. The tier question is what makes old iron legal. Most of these units are Tier 2 or have no tier at all, which limits them to emergency service. Under [40 CFR 60.4211(f)](https://www.ecfr.gov/current/title-40/chapter-I/subchapter-C/part-60/subpart-IIII/subject-group-ECFR8fe0e3a5a3e5b6b/section-60.4211) an emergency stationary engine may run "for a maximum of 100 hours per calendar year" for maintenance and readiness testing, of which "up to 50 hours per calendar year" may be non-emergency, and none of those 50 for peak shaving or paid demand response. A used Tier 2 set is a backup generator, full stop, and the [Genset TCO](https://open-factory.com/tools/genset-tco/) model treats its run hours accordingly. Our September piece walked through the [emergency-engine route by state](https://open-factory.com/articles/gensets-sold-out-2028/). This one only adds that the listings confirm the price penalty for the constraint: the one new emergency-only Tier 2 unit we found, a [Cummins DQFAD 1,000 kW at $269,900](https://csdieselgenerators.com/new-cummins-dqfad-qst30-g5-nr2-diesel-generator--epa-tier-2-5036.html), is $270 per kW, inside the used 2011 to 2019 band. ## The Surplus Auction Floor Below the dealers sits the auction floor, and it is where the "cents on the dollar" stories come from. On the morning of September 10, 2026 GovDeals had 19 lots in its transformers category and 77 in generators. The one that matters is a [Cooper 5,000 kVA OA/FFA substation transformer](https://prod-seo.govdeals.com/en/asset/317/16014), 13,800 V to 4,160Y/2,400, consigned by an industrial seller in Norco, Louisiana, closing September 17. At 07:30 UTC it had seven bids and stood at $100; an hour later it stood at $2,100. Maddox lists a 5,000 kVA pad-mount from stock at [$174,200](https://store.maddox.com/products/5000-kva-padmount-transformer). A 75 kVA pad-mount in Questa, New Mexico sat at $100 against a [$3,193 new-unit award](https://www.guc.com/sites/default/files/uploads/bid-tab-sheets/25-80-bid-tab.pdf) at GUC three months earlier. Six open GovDeals lots on September 10, 2026 carried current bids of $100 to $10,500. The 5,000 kVA Cooper at $2,100 was bid at 1.2% of the nearest in-stock dealer price for the size, with seven days left. **Realized auction prices for electrical gear are not public, and that is the gap this market needs filled most.** GovDeals shows closed prices only to logged-in accounts and its main site returned "Access Denied" to every route we tried; Purple Wave, HiBid and PublicSurplus render results in scripts a reader cannot paginate; Ritchie Bros wants an account. What is public is the terms, and the terms are the discount. The Cooper lot carries a [15% buyer's premium](https://prod-seo.govdeals.com/en/asset/317/16014), removal "within ten (10) business days from the time and date of the close of the auction," a refinery gate that requires "a valid driver's license, copy of registration, proof of insurance and name of your company on both sides the truck," and no test report of any kind. A 5 MVA tank is a permitted oversize haul with oil on board, and the ten days start the moment the hammer falls. That is why a $2,100 bid is not a $2,100 transformer. We covered the same asymmetry for [used CNC machines](https://open-factory.com/articles/used-cnc-auction-index/), where the auction houses also publish lots and never publish hammer prices as a series, and for the [EV plants' machinery](https://open-factory.com/articles/ev-plant-machines-second-hand/) that hit those auctions in 2025. Electrical gear has the extra problem that a buyer cannot see the oil. ## Refurbishment Is a Ladder, Not a Line Item No dealer publishes a price list for refurbishment. What they publish is a price for each rung of condition, and the gap between rungs is what refurbishment costs the buyer. Bullock Breakers lists a used [GE VB1-13.8-1000-3 15 kV, 2,000 A vacuum breaker at $11,500 and a reconditioned VB1-13.8-1000-2 at $19,500](https://bullockbreakers.com/air-vacuum-breakers/general-electric/vacuum-breakers/15kv/), an $8,000 or 70% premium for the teardown, and on the same page 1,200 A units at $10,500 and a 3,000 A unit at $23,500. EMSCO lists [Cutler-Hammer 150 DHP-VR-1000 breakers](https://www.emscomn.com/vacuum-breakers/) at $12,500 for 1,200 A and $15,500 for 2,000 A as "used electrically OK," and a Westinghouse 150DHP500 retrofit conversion at $6,500, with a one-year warranty on refurbished units. Complete lineups, an [ABB ADVAC main-tie-main with 17 breakers](https://bullockbreakers.com/products/abb-advac-15kv-1200a-main-tie-main-switchgear-lineup-188) or a GE PowerVac seven-section main-lug, are quote only. Our [switchgear price piece](https://open-factory.com/articles/switchgear-50-percent/) put the new-price context around those numbers. The used market for MV gear is priced by the breaker because the breaker is what gets reconditioned. At 2,500 kVA the rungs read $41,200 low bid, $58,871 median bid, $69,950 ERMCO, $101,253 high bid and $104,000 dealer stock; for diesel gensets per kW they read $67 for a 2001 unit, $180 for a 2014 unit, $335 for a 2011 QSK60, $530 for a rebuilt rental-grade 2007 C15 and $780 for new Tier 2 stock; for the breaker, $11,500 used and $19,500 reconditioned. **What a buyer is paying for at each rung is disclosed by exactly one seller.** Maddox publishes its remanufacturing test list: "Insulation Resistance / Megger, Transformer Turns Ratio, Winding Resistance, Load and No Load Losses, Excitation Current, Impedance, DGA (for oil filled), PCB (for oil filled), Power Factor (Available on Request), HIPOT (Available on Request)," a "rigorous 45-point quality control inspection," and a unit "left overnight for a leak and pressure test," under a PEARL accreditation and a [3-year warranty](https://www.maddox.com/remanufactured). Depco publishes its genset test standard: two load banks at 750 kW and 2 MW, and "All of our tests are performed at full capacity for a minimum of one hour" ([Depco](https://www.depco.com/our-process-and-guarantee/)). Generator Source publishes a stepped procedure, "Apply 25% -> 50% -> 75% -> 80% load; hold each stage to stabilize and log readings" ([Generator Source](https://generatorsource.com/generator-service/load-bank-testing/)), and posts a load bank video per unit. North American Switchgear describes Class A reconditioning as "teardown to the smallest component" with a one-year warranty ([NASWGR](https://naswgr.com/Shop.html)). The rebuilt rental-grade rung on the genset ladder, Depco's [2007 CAT C15 at $265,000](https://www.depco.com/diesel/caterpillar-c15-500kw-rental-grade-generator-set-item-20837/) or $530 per kW, sits three times above a used 2014 unit of the same rating and two-thirds of the way to new; that gap is the market's price for a rebuilt engine and a rebuilt SR4 generator end, and no dealer itemizes it. ## Charlotte: Refurbishment at Utility Scale The same ladder exists at 100 MVA, and one manufacturer built a factory for it. Siemens Energy announced in February 2024 that it would spend $150 million to expand its Charlotte, North Carolina plant into the first Siemens Energy transformer production site in the United States, with output of 24 new large power transformers a year rising to 57 and, in the same building, 12 refurbished units a year rising to 24 ([APPA, Feb 16 2024](https://www.publicpower.org/periodical/article/siemens-energy-expanding-nc-facility-new-refurbished-large-transformers); [Siemens Energy release, Feb 14 2024](https://www.siemens-energy.com/us/en/home/press-releases/siemens_energy_addresses_shortage_US_powertransformers_invests_new_factory.html)). The release put the problem plainly: "only 20% of U.S. large power transformer demand is met by domestic supply with lead times of up to five years." By June 2025 the start date had moved. Board member Tim Holt told Reuters the first Charlotte units would "run off the factory lines in early 2027," a year later than the release, and that "more than 80% of large power transformers (LPT) are currently imported" ([Transformer Magazine, Jun 30 2025](https://transformer-magazine.com/magazine/siemens-energy-to-start-large-power-transformer-production-in-the-us/)). On February 3, 2026 the company added a further $421 million across North Carolina, most of it generators and gas turbines in Charlotte plus a grid parts and service center near Winston-Salem ([WFAE, Feb 3 2026](https://www.wfae.org/charlotte-area/2026-02-03/siemens-energy-will-invest-over-400m-in-north-carolina)). **A refurbished LPT is a new core-and-coil decision deferred, and Siemens Energy has priced a slot for it at one refurb for every two new units.** The refurb line is the manufacturer's answer to the same arithmetic the pad-mount buyer faces: when a factory order is [143 weeks out for a GSU](https://open-factory.com/articles/transformer-160-weeks/), as Wood Mackenzie's Q2 2025 survey put it, a unit that can be rewound, re-gasketed, re-bushed and re-oiled in a domestic shop is a queue position the utility already owns. The fleet supports the business. NREL estimates the country runs 60 to 80 million distribution transformers, "about half are more than 33 years old," and replaces 2.5% to 3% of them a year ([pv magazine, Jan 7 2025](https://pv-magazine-usa.com/2025/01/07/half-of-distribution-transformers-are-approaching-end-of-life/)). Half a fleet past its design life is a used market whether or not anyone indexes it, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes the factory-order lead times that set the price of the alternative. ## The Numbers on the Oil Report The reason used electrical gear trades without an index is that its condition lives in a lab report the buyer has to ask for. The good news is that the acceptance numbers are public, and federal. The Bureau of Reclamation's transformer manual reprints the Doble in-service oil limits by voltage class: dielectric breakdown by ASTM D1816 of at least 20 kV up to 288 kV and 25 kV above; oil power factor at 25 C no more than 0.5%; water no more than 35 ppm up to 69 kV, 25 ppm to 288 kV and 20 ppm above; interfacial tension at least 25 dynes/cm; neutralization number no more than 0.2 mg KOH/g up to 69 kV and 0.15 above ([USBR, Transformers: Basics, Maintenance, and Diagnostics, Apr 2005, Table 19](https://www.usbr.gov/tsc/techreferences/mands/mands-pdfs/Trnsfrmr.pdf)). The manual also says what a failure means in dollars: filtration and drying "may be required if the dielectric breakdown voltage or water content do not meet these limits," while "reclamation (clay filtration) or replacement of the oil may be required" when power factor, IFT, acidity or sludge fail. Eight limits sort a used transformer into three bins: pass, process the oil, or walk, with oil power factor at 0.5% and 1.0%, TDCG at 720 ppm and 4,630 ppm and overall insulation power factor at 0.5% as the thresholds. **Oil power factor above 1.0% is the one number that ends the negotiation.** FIST 3-31 puts the bands in order: below 0.5% at 25 C is the in-service limit; between 0.5% and 1.0% "further investigation is required; the oil may require replacement or Fuller's earth filtering"; above 1.0% "the oil may cause failure of the transformer; replacement or reclaiming of the oil is required immediately"; above 2% "oil should be removed from service and replaced because equipment failure is imminent. The oil cannot be reclaimed" ([FIST 3-31, Jun 2003](https://www.usbr.gov/power/data/fist/fist3_31/fist3-31.pdf)). The same manual carries the IEEE C57.104 dissolved-gas conditions, with Condition 1 at total dissolved combustible gas below 720 ppm and Condition 4 above 4,630 ppm where the guidance is "consider removal from service," and the Doble field-test grades of G, I, D and B, with the rule that overall insulation power factor "should not exceed 0.5% at 20 EC." On moisture it is blunt: "Each time the moisture is doubled in a transformer, the life of the insulation is cut by one-half." Maddox's own [2020 buyer checklist](https://www.maddox.com/resources/articles/benefits-and-risks-of-buying-reconditioned-transformers) asked sellers to "Provide ANSI C.57 test reports" and "Provide Dissolved Gas Analysis (DGA) of the transformer oil results prior to rebuilding," and that is the right order: the DGA before the rebuild tells you what the rebuild had to fix. For gensets the equivalent document is the load bank log, and the standard the dealers set for themselves is a floor: Depco's one hour at full capacity, Generator Source's four steps to 80%. A unit with 58 hours in 28 years has never been through either, and the buyer should budget the test as part of the price rather than take a "low hours" label as a condition report. ## Weeks, Not Cents, Is the Product Everything above collapses to one chart. The used and refurbished market does not sell a discount. It sells the weeks between a factory order and an energized bus, and the discount or premium follows from how many weeks a buyer is short. Fifteen PWC bidders quoted 10 to 35 weeks with a median of 18; GUC's vendors quoted 10 to 63; Maddox's own comparison point for factory-built units is 80 weeks; remanufactured pad-mounts ship in 1 to 6 weeks, used dry-type in a week, and used gensets in one to four. **The median gap is 17 weeks, and 17 weeks has a price a buyer can compute before asking for a quote.** The Lead-Time Monitor's published figures for the new units are the long side of that gap: [50 to 66 weeks for a 1,500 to 2,000 kW diesel](https://open-factory.com/articles/gensets-sold-out-2028/) and, for the transformer classes above pad-mount, the [Wood Mackenzie survey figures](https://open-factory.com/articles/transformer-160-weeks/) we walked through in October. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tool takes the PO date and the class and returns the energization month, which is the number the stock premium has to beat. The dealers know this. Depco names an inventory bucket "ready in seven." Bullock's pitch is "a fraction of the lead time." EMSCO's is "when lead times won't work." Woodstock Power promises "zero lead time." None of them leads with price, and the price list, where one exists, confirms they should not. ## Where It Trades The market has perhaps a dozen public venues that a buyer can read without a login, and what they disclose varies more than what they charge. Of the ten venues, six print prices on at least some units, three state a warranty (Maddox 3 years with a 5-year option, EMSCO and North American Switchgear 1 year), two publish a load-bank procedure, one publishes a transformer test list, and the two auction houses sell "as is, where is" with a 10% to 15% buyer's premium and no test of any kind. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades the manufacturers on whether they publish lead times, list prices and backlog. The used market deserves the same treatment and this table is the v0 of it. The scoring is simple. A seller that prints a price, a warranty and a test procedure has told the buyer everything the buyer needs to price the risk. A seller that prints a price and nothing else has told the buyer where the negotiation starts. An auction has told the buyer where the truck goes. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/), once it has contributions, will be the first place the closing prices from that third category appear as a distribution, because the auction houses will not publish them and the buyers who paid them are the only other people who know. ## What to Do Monday Pull the bid tabs before the quotes. Fayetteville PWC and Greenville Utilities publish theirs as PDFs and so do dozens of other municipal utilities, and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) includes the ones we have transcribed with n and date. A buyer who knows the fifteen-bidder median for a 2,500 kVA pad-mount is $58,871 walks into a dealer conversation knowing that $104,000 is a 1.77x premium for delivery this month and not a market price. Ask every used or remanufactured seller for the eight numbers on the oil report, the Doble letter grades and the factory test sheet, in that order, before discussing price. The ones that publish a test list will hand them over and the ones that do not will tell you what the discount is for. For gensets, ask for the load-bank log at 100% for at least an hour and treat model year, not hours, as the price variable, because that is how the dealers price it. And run the delay arithmetic first: the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) gives the weeks, the section below gives the formula, and together they say whether the stock premium is a cost or a saving. Behind the paywall: the five checks we would run before wiring money for a used transformer, genset or breaker, each tied to a published limit, and three worked examples that show why the **$104,000** in-stock 2,500 kVA pad-mount beats the $58,871 median bid for any project that gates more than **$2.2 million** of idle capital, built on the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands. --- # 1872, Foundational Industries and Hadrian: Three Blueprints for the Autonomous American Factory URL: https://open-factory.com/articles/1872-foundational-industries-hadrian/ Date: 2026-01-11 Machinery / Deep dive 1872's $15M and Path Robotics welding in a 1903 Cincinnati plant, Foundational's $25M AI-native rack factory, Hadrian's $1.37B at $7.87B and 3 million sq ft, the 330,000-welder gap, what a buyer can order and when Three companies are building the autonomous American factory from three directions, and between July 2025 and August 2026 they raised $1.41 billion to do it. 1872 took [$15 million](https://theohiofund.com/press-releases/1872-launches-autonomous-steel-fabrication-factory-model-in-cincinnati-and-closes-15-million-seed-funding-round/) to run robotic steel fabrication out of a 1903 building in Cincinnati. Foundational Industries took [$25 million](https://fortune.com/2026/07/30/forget-robots-assembly-lines-foundational-industries-ai-to-run-entire-factory-manufacturing-china-robotics/) to build a rack-enclosure plant that already runs in software. Hadrian added [$1.37 billion](https://www.prnewswire.com/news-releases/hadrian-raises-1-37b-series-d-to-build-highly-automated-factories-to-accelerate-americas-industrial-renewal-302844408.html) on top of four operating plants totalling just under 3 million square feet. One is a steel shop, one is a datacenter supplier, one is a defense industrial base in its own right. They share a bet: that a factory run by software, with people trained in weeks rather than years, can quote in days and deliver in weeks where the incumbent quotes in weeks and delivers in months. This piece covers the three founding teams and their machines, the welding gap that makes 1872 possible, the emulated factory Foundational will pour concrete for by December, Hadrian's jump from 600,000 to 3 million square feet, the Cherokee submarine plant and its $35.24-an-hour jobs, what a buyer can order from each today, and behind the wall, the questions to put to each before the purchase order. Five plants across three companies, two of them producing today, one operational since early 2026, one opened on 20 March 2026 with full-rate production promised inside 24 months, and one to be sited and opened by the end of 2026. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) will add a delivered-lead-time row for each as the first public orders ship. ## Three Blueprints, One Idea The idea is the same one that took [Nox Metals](https://open-factory.com/articles/nox-metals/) from a Y Combinator batch to a Ferndale plate saw and [SendCutSend](https://open-factory.com/articles/sendcutsend/) from a Reno garage to a billion-dollar valuation: put the software in front of the machine, so the quote, the schedule and the material plan are computed rather than phoned. 1872's version is called Factory OS. Foundational's is a "common intelligence layer" that every work cell reports to. Hadrian's is Opus, which the company describes as an [AI platform for factory autonomy](https://www.hadrian.co/) that reads legacy designs and automates machining and inspection. The three differ in what comes out the door, and in scale by two orders of magnitude, but a buyer can read them with one checklist: what is the product, where is the building, how many people does it take, and when does the first delivery date land. The money tells you how far along each is. 1872's $15 million is one of the largest seed rounds in Ohio history and pays for production at one plant. Foundational's $25 million is, per Fortune, meant to prove the model at one facility rather than fund a network. Hadrian's $1.37 billion Series D, co-led by WCM Investment Management, Washington Harbour Partners, Valor Equity Partners, 137 Ventures and Baillie Gifford with JPMorgan Chase's Strategic Investment Group as anchor, funds new factories, production lines and R&D for a company that already runs four. The [Factory Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) puts all three on the same 2026 to 2028 calendar as the robot cells and machine tools a shop buys today. ## 1872: Launch-Grade Operations in a 1903 Cincinnati Plant The name is the year Andrew Carnegie built his first steel mill, and the founders come from the company that made "launch-grade" a manufacturing adjective. Dan Summers, the chief executive, led Raptor engine integration and test engineering at SpaceX. Michael Grant, the chief technology officer, built flight-software infrastructure for Starship. Brian Mongilio, head of operations, [ran Merlin engine production](https://x.com/GrishinRobotics/status/2081097531241501103). On 22 July 2026 they announced a $15 million seed led by private funds advised by The O.H.I.O. Fund and opened for production in the [1903 David Hummel Building Company plant](https://local12.com/news/local/former-spacex-engineers-steel-factory-cincinnati-15-million-investment-1872-camp-washington-david-hummel-building) on Spring Grove Avenue in Camp Washington, a building put up by the masonry firm that built Cincinnati City Hall and Union Terminal. What they make is heavy steel fabrication for energy, utility, industrial and modular-construction customers: the weldments, frames and structural assemblies that today come from job shops with a six-month backlog, and whose imported alternatives now attract the Section 232 steel duty the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) applies at full customs value. Factory OS, their own software, [ingests a 3D design, plans material and logistics, schedules machines and coordinates the welding robots](https://www.ohiotechnews.com/1872-launches-autonomous-steel-factory-cincinnati-seed-round/), and also runs purchasing, estimating and shipping, so the whole order from drawing to truck sits in one system. Early automations are already running. The company targets full autonomy in 2027. Summers's stated goal is a plant where "our software orchestrating robotics" delivers "a faster, more efficient, and safer fabrication process", and the claim on lead time is specific: from months or longer to weeks. **1872's arithmetic is a $350 billion fabricated-metal market served by shops that cannot hire, and its answer is to make the weld with a robot that has already seen tens of millions of inches of it.** The company chose Ohio for the labor pipeline and the engineering programs at the University of Cincinnati and Ohio State, and for the robot supplier ninety minutes up I-71. The Camp Washington plant is the prototype. The release calls it the model to validate before scaling to further sectors. A buyer of structural weldments should read that as a plant with capacity to sell in 2026 and 2027, and one worth a first order sized to learn from. We priced a conventional cobot MIG cell at $113,336 in [the robot cell piece](https://open-factory.com/articles/robot-61k-cell-150k/), and the difference between that and 1872 is that the buyer does not have to own it. The [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has the public-award prices for fabricated steel that a first 1872 quote can be read against. ## Path Robotics, Obsidian and the 330,000-Welder Gap The robots come from Path Robotics in Columbus, and the model inside them is Obsidian, which Path announced on 8 September 2025 as [the first purpose-built foundational AI model for welding](https://www.path-robotics.com/news/path-robotics-announces-obsidian-foundational-model-welding), trained on tens of millions of welded inches. Path's founder and chief executive Andy Lonsberry put the case plainly at 1872's launch: welding remains one of manufacturing's most difficult processes to automate, because every fit-up is a little different and a fixed program cannot see the gap. Obsidian is built to see it. In April 2026 Path put the same model on a Boston Dynamics quadruped and called it Rove, with Saronic among the first shipbuilders to evaluate it, which tells a buyer where this technology is headed: off the fixed cell and into the hull. 82,500 a year for four years is 330,000 by 2028 on the American Welding Society's projection [as The Fabricator reports it](https://www.thefabricator.com/thewelder/article/arcwelding/outlook-trends-and-pay-for-the-welding-workforce-according-to-aws-data). Path's own release cites 320,500 by 2029 and says 80% of welding is still done by hand. The two figures differ by a year and a rounding, and both say the same thing: the constraint on American steel fabrication is not steel and not machines, it is people holding torches. Every weld a robot makes at Camp Washington is a weld that did not need one of the 82,500. The [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) tool has the current quoted weeks for fabricated steel from conventional shops, which is the number 1872's "weeks" is measured against. ## Foundational Industries: A Factory Written in Software First Jonathan Winer spent six years at Alphabet's Sidewalk Infrastructure Partners, deployed more than $1 billion of capital there and, by his own site's count, [$3 billion across 25 years](https://foundational.industries/mission) of technology-enabled infrastructure. On 30 July 2026 he announced a $25 million seed for Foundational Industries, led by BoxGroup and Zigg Ventures with Abstract Ventures, Adverb Ventures, Buckley Ventures, Oceans and Offline Ventures, to build what he calls an AI-native factory in the United States. The team, per Dealroom's summary, [draws on Google's physical AI work and the leadership of America's two largest contract manufacturers](https://dealroom.co/news/142088-foundational-industries-raises-25m-seed-to-build-ai-native-us-factories/). The thesis is the reverse of the usual retrofit. Rather than bolting robots onto a line designed for people, Foundational designs the plant for the software from the first drawing. Winer's words: every process, every work cell, every material movement, every inspection step, every scheduling decision and every quality loop is built around one intelligence layer, so the result is a factory orchestrated by AI rather than assisted by it. The company has built the whole plant in software emulators before building it in steel, and each factory it deploys is meant to improve the model and extend the operating playbook for the next. The first physical facility opens by the end of 2026, described on the company's site as the first node in a network. The first product is the one the datacenter boom cannot get enough of: custom rack enclosures for developers, neoclouds and chipmakers whose new silicon [runs at different voltages and needs different cooling](https://fortune.com/2026/07/30/forget-robots-assembly-lines-foundational-industries-ai-to-run-entire-factory-manufacturing-china-robotics/) than the racks in the catalog. Winer frames the national case in numbers: China deploys more than 300,000 industrial robots a year, makes about a third of the world's goods and has put more than $1 trillion into advanced manufacturing over a decade, while the United States holds about 75% of the world's AI compute and is spending an estimated $7 trillion to build the datacenters that run it. The rack is where those two facts meet. We showed in [the 800 VDC piece](https://open-factory.com/articles/800vdc-strands-frame-agreements/) how a voltage change strands a white-space frame agreement, and a factory that can retool a rack design in software is the supplier that change creates. The [MEP Cost Table](https://open-factory.com/models/mep-cost-table/) gives the installed cost per rack position a colo buyer will compare the enclosure price against. **A buyer of rack enclosures in 2027 will have a US-built option designed for the new silicon, and the date to put in the calendar is the end of 2026.** Foundational has not named a site or a customer, which is ordinary for a company nine weeks past its seed announcement, and the one question a purchasing team should hold ready is the delivery date of the first production enclosure. The [datacenter electrical roadmap](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) lists the rack and power-shelf transitions that Foundational's first customers are buying for. ## Hadrian: From 600,000 to 3 Million Square Feet in a Year Chris Power founded Hadrian in Torrance, California, in 2021 to automate precision machining for aerospace and defense, and the company has grown by a factory a year since. Lux Capital and Andreessen Horowitz led a [$90 million Series A in 2022](https://en.wikipedia.org/wiki/Hadrian_Automation). Founders Fund and Lux led a [$260 million Series C in July 2025](https://www.prnewswire.com/news-releases/hadrian-raises-260m-to-build-ai-powered-factories-for-america-adds-full-product-manufacturing-opens-arizona-site-302508179.html), with a factory-expansion loan arranged by Morgan Stanley, to build Factory 3 in Mesa, Arizona: [270,000 square feet, $200 million, 350 jobs, operational early 2026](https://www.gpec.org/news/press-releases/hadrian-announces-mesa-manufacturing-software-hub/), plus a 400,000 square foot headquarters search. In January 2026 a round led by T. Rowe Price [valued the company at $1.6 billion](https://www.therobotreport.com/hadrian-brings-in-additional-funding-bringing-its-valuation-to-1-6b/) and the company launched Hadrian Additive, a division to industrialize metal printing for defense programs, and disclosed a [$360 million revolving credit facility](https://www.prnewswire.com/news-releases/hadrian-launches-additive-manufacturing-division-to-expand-us-defense-production-capacity-302668241.html). $90 million in 2022, $260 million in July 2025, $1.37 billion in August 2026. The Series D, announced 6 August 2026 at a $7.87 billion post-money valuation, had 1789 Capital, Morgan Stanley Wealth Management, Apollo funds, T. Rowe Price, CapitalG, Andreessen Horowitz, Founders Fund, Lux Capital, Altimeter and Construct Capital behind the five co-leads and JPMorgan. Power's line in the release: "Production is now the frontline of deterrence. America's ability to lead will depend on whether we can build, train, and scale faster." $1.6 billion in January, $7.87 billion in August, a 4.9x step in seven months, on a company whose footprint went from three plants and about 600,000 square feet in January to four plants and just under 3 million by August. The [Book-to-Bill Table](https://open-factory.com/models/book-to-bill-table/) tracks the listed defense and electrical OEMs Hadrian sells into, and the pattern there, backlog growing faster than shipments, is the demand the valuation is pricing. About 330,000 square feet across two Torrance sites (our subtraction of Mesa from the 600,000 three-site total), 270,000 in Mesa and 2.2 million in Cherokee, Alabama. Hadrian sells three ways: precision components from prototype to production, manufacturing as a service (dedicated cells at its plant or the customer's), and factories as a service, the design and operation of a whole plant for a defense prime with a production problem. Its site says a technician with limited manufacturing experience [reaches full productivity in 30 days or less](https://www.hadrian.co/), and every employee gets stock. Anduril is a named partner. On the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) Hadrian is unusual among private companies for publishing square footage, headcount plans and wages plant by plant. ## Cherokee, Alabama: 1,000 Jobs at $35.24 an Hour Factory 4 opened on 20 March 2026 at 1200 Haley Drive in Cherokee, Alabama, in the former FreightCar America railcar plant that closed in 2021, once the largest railcar factory in the country. It is [2.2 million square feet](https://aeshoals.com/hadrian-opens-advanced-manufacturing-facility-in-cherokee-bringing-1000-jobs-to-the-shoals/), dedicated to components for the Navy's Columbia-class and Virginia-class submarine programs, and it is a $2.4 billion public-private partnership: $1.5 billion of Hadrian's capital and $900 million in Navy appropriations. Secretary of the Navy John Phelan said at the opening that "private-sector partnership is not optional. It is foundational." Power's version was that industry "has to answer that call with real execution, and the window to do it is now." $1.5 billion private, $900 million public, full-rate production within 24 months of the March 2026 opening, which puts the date in the first quarter of 2028. The town of Cherokee has about 1,000 residents. The plant will employ [more than 100 people by the end of 2026 and about 900 by the end of 2029](https://yellowhammernews.com/hadrian-starts-hiring-at-the-shoals-submarine-plant-35-an-hour-jobs-in-a-town-of-1000/), up to 1,000 at capacity, and the Shoals Economic Development Authority's estimate is 888 jobs at an average of $35.24 an hour. Hiring began in September 2026. 100 by December 2026, 900 by December 2029, 1,000 at full rate. At $35.24 an hour and 2,080 hours, a full staff is a wage bill of about $73 million a year in a county that lost its largest employer five years ago, and the 30-day training claim is what makes 1,000 hires in a town of 1,000 a plan rather than a hope. [Datacenter construction is eating electricians](https://open-factory.com/articles/electricians-wages-by-metro/) in the metros. Hadrian's answer is to train technicians where the electricians are not. A 2.2 million square foot plant also draws power on a scale the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) was built for, and the [Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool prices the utility side of a plant that size in a TVA territory. ## What a Buyer Can Order, and When Monday's list is short and dated. From 1872, request a quote on one structural weldment you already buy, with the drawing, the AWS D1.1 procedure requirement and your current supplier's quoted weeks written on the RFQ, and ask for the delivery date in writing. The plant is producing now, and a first order sized at one truckload teaches you what Factory OS does with a real drawing. Run the quote you get back through [Quote Check](https://open-factory.com/tools/quote-check/) beside the incumbent's to see which line items the software removed. From Foundational Industries, there is nothing to order yet, and that is a scheduling fact rather than a judgment: put the end of 2026 in the calendar, send the rack specification for your 2027 white-space build now so the emulated plant has your design when the real one opens, and price the enclosure against the [MEP Cost Table](https://open-factory.com/models/mep-cost-table/) per-rack-position figure when the quote arrives. From Hadrian, if you are a prime or a Tier 1 with a machined-part backlog, the manufacturing-as-a-service model is the door: dedicated cells at Torrance or Mesa, or at your plant, with a 30-day-trained workforce. If you are a commercial shop, Hadrian is not your supplier, but it is hiring your machinists at $35.24 an hour with equity, and the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) is where to log what that does to the quotes you get from the shops that remain. The [RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) has the clauses for a first order with a new plant, and the [liquidated damages piece](https://open-factory.com/articles/liquidated-damages/) explains what a date is worth when a supplier is new. Behind the paywall: the order book for each of the three, the certification each has said it holds or is pursuing, the **24-month** Cherokee ramp read against what the Navy's two submarine programs need by year, and the five questions to ask before a first purchase order at a plant that is younger than its backlog. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) will publish the first delivered lead times from all three as they become public. --- # Japan Is Eating the US Capex Boom URL: https://open-factory.com/articles/japan-eats-capex-boom/ Date: 2026-01-11 Machinery / Take JMTBA North America JPY 40.7B, +44%, 37 cents of every USMTO dollar, the yen at 158, one 15% rate for Japan, Taiwan and Germany, US assembly plants, a Taiwanese 5-axis in Ohio Japanese machine tool builders booked JPY 40.65 billion of North American orders in June 2026, 44% more than in June 2025, and over the first half of the year the United States alone gave them the dollar equivalent of $1.29 billion against the $3.44 billion that US buyers placed for all metalworking machinery: 37 cents of every dollar of the strongest half-year since USMTO began in 1998 ([JMTBA, June 2026 release](https://www.jmtba.or.jp/english/wjmtbap/wp-content/uploads/2026/08/JMTBA-NOTES-2026-07.pdf); [American Machinist, July 2026](https://www.americanmachinist.com/news/news/55397089/machine-tool-demand-is-booming-in-2026-us-manufacturing-technology-orders-report-june-2026)). That is before the machines Mazak builds in Kentucky and DMG Mori builds in California, which appear in neither series. The yen at 158 does most of the work, the tariff schedule none of it, and because the [Open Factory Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) still has zero machine-tool awards, every dollar figure below is public data, converted, with the arithmetic shown. This piece covers the JMTBA and builder numbers, the import data that has not caught up, the 35% currency discount that nobody passed on, the one 15% rate that now applies to Japan, Taiwan and Germany alike, a Taiwanese 5-axis landed in Ohio, and, behind the wall, which builders assemble in the US and where the room sits in a 2026 quote. North America ran JPY 24.2 billion in January 2025, JPY 41.5 billion in December, and between JPY 34.6 billion and JPY 43.1 billion in every month of 2026 through July; Asia went from JPY 44.8 billion to JPY 81.4 billion over the same span, and Europe from JPY 13.2 billion to JPY 23.1 billion. The JPY 81.42 billion, +174.4% figure that circulated for North America in June is JMTBA's Asia line ([JMTBA, August 5, 2026](https://www.jmtba.or.jp/english/wjmtbap/wp-content/uploads/2026/08/JMTBA-NOTES-2026-07.pdf)); a trade-press table swapped the rows ([Seisanzai Japan, July 30, 2026](https://seisanzai-japan.com/article/p6294/)) against its own text, and [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/) has the correction. North America was JPY 40.65 billion, up 44.0%. The United States alone was JPY 36.0 billion, up 44.4%. ## Thirty-Seven Cents of Every Dollar **Japanese builders' US orders were JPY 203.7 billion in the first half of 2026, up 36.1% on the first half of 2025, and USMTO's US total was up 36.0% over the same six months.** The US figure is JMTBA's destination table (North America less Canada and Mexico, JPY 149.6 billion a year earlier), transcribed in [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/). The whole continent was JPY 232.5 billion, up 36.5%. The two series differ: JMTBA counts builders' bookings at the builder's price, USMTO counts US distributors' bookings at the distributor's price. Converted month by month at the [FRED EXJPUS](https://fred.stlouisfed.org/graph/fredgraph.csv?id=EXJPUS) average, Japanese builders' US orders were $1.01 billion in the first half of 2025 and $1.29 billion in the first half of 2026, a 27.3% rise in dollars against USMTO's 36.0%, because the yen fell from 148.4 to 158.1 between the two halves. The Japanese share held near 37 cents on the dollar. What changed is the price, which is the subject of the third chart. The record half-year itself, $3.44 billion on 11,243 machines and a $306,000 average order, is unpacked in [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/). The listed builders confirm the shape. DMG Mori booked JPY 335.2 billion of orders in January to June 2026, up 34.8% and a record for a half, with the April to June quarter at JPY 179.8 billion, up 40.5%. Its average machine order was JPY 81.8 million, about $517,000 at its own 158.2 first-half rate, and its machine backlog reached JPY 303.0 billion at the end of June against JPY 240.0 billion in December ([DMG Mori IR conference, August 4, 2026](https://www.dmgmori.co.jp/corporate/en/ir/ir_library/pdf/fy2026_2shihanki_ex_e.pdf)). It names "global key accounts in Europe and North America" as the buyers, and its EBIT bridge credits gross margin to "lower discounts for MX products." Okuma's Americas orders were JPY 25.7 billion in April to June 2026, up 102.2% on the year and 34% of a group total of JPY 75.3 billion that was itself a first-quarter record; its Americas orders for the year to March 2026 were JPY 78.7 billion, up 25.5%, and its backlog stood at a record JPY 124.8 billion at the end of June ([Okuma Q1 FY2026 results, August 5, 2026](https://contents.xj-storage.jp/xcontents/61030/d6f666ab/0ec3/4e31/9d27/17d28746b9de/20260805133257045s.pdf); [FY2025 results, May 28, 2026](https://contents.xj-storage.jp/xcontents/61030/b216cc37/cdfd/4a14/80a6/736e06448223/20260528192540264s.pdf)). Yamazaki Mazak is private and publishes nothing, which is why it sits where it does on the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/). The four builders' disclosures are in [data/builder-disclosures.csv](data/builder-disclosures.csv). ## The Import Data Has Not Caught Up **US imports of machine tools were $11.5 billion in 2025 and Japan supplied $1.88 billion of them, 16% of the total, down from 25% in 2015 and behind Germany for the second year.** Customs data run a year behind orders and cannot see a machine assembled in Florence, Kentucky, so they miss the 2026 wave, but they show what a decade of yen did. Japan shipped $2.32 billion of HS 8456 to 8466 machinery to the US in 2015 and $1.88 billion in 2025; at the average exchange rate of each year that is JPY 281 billion both times, to the billion, so a decade of "decline" in the dollar series is a flat line in yen ([UN Comtrade](https://comtradeapi.un.org/), US-reported imports, CSV in [data/us-machine-tool-imports-by-country.csv](data/us-machine-tool-imports-by-country.csv)). Germany went from $1.51 billion to $2.22 billion, Taiwan from $0.86 billion to $1.13 billion, South Korea from $0.44 billion to $0.78 billion, and China from $1.20 billion to $0.73 billion. On machining centers alone (HS 8457) Japan and Germany tied in 2025 at $505 million and $504 million, Korea took $152 million and Taiwan $90 million; on lathes (HS 8458) Japan led with $451 million to Korea's $264 million ([data/us-imports-8457-8458-8459-by-country-2025.csv](data/us-imports-8457-8458-8459-by-country-2025.csv)). The gap between JMTBA's JPY 313 billion of 2025 US orders, about $2.09 billion, and $1.88 billion of 2025 imports is Kentucky, Davis, and a backlog that DMG Mori says is now 6.6 months of sales. ## A 35% Discount Nobody Passed On **A machine priced at JPY 40 million cost $385,000 in January 2021 and $252,000 in August 2026, a 35% cut delivered entirely by the exchange rate.** The yen was 103.8 to the dollar in January 2021, 145 by October 2022, 158.8 in August 2026, and 156.1 on September 4, 2026 ([FRED EXJPUS and DEXJPUS](https://fred.stlouisfed.org/graph/fredgraph.csv?id=DEXJPUS)). Read the line: $385,000 at the January 2021 peak, $307,000 in January 2023, and $246,000 to $256,000 in every month of 2026. Nothing like that appears on a US price list. The BLS producer price index for machine tool manufacturing rose 22.5% between January 2021 and July 2026, from 116.4 to 142.6 ([BLS PCU333517333517 via FRED](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCU333517333517)), and the builders' own planning rates tell you where the currency went: DMG Mori budgets JPY 150 per dollar for the second half of 2026 against an actual 158.2 in the first half, and Okuma moved its assumption from 150 in May to 159 in August ([Okuma, August 5, 2026](https://contents.xj-storage.jp/xcontents/61030/d6f666ab/0ec3/4e31/9d27/17d28746b9de/20260805133257045s.pdf)). Okuma's president writes of "establishing high value-added prices in each market" and DMG Mori of lower discounts. Neither has cut a US list price. The yen's 35% became the builders' margin, the distributors' margin, and the discount that was not offered, in a proportion nobody publishes, which is the gap the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) exists to close once it has five contributors and eight quotes per cell. ## One Rate for Three Origins **A machining center from Japan, Taiwan or Germany pays 15% at the border today, all-in, and the same machine pays 29.2% from January 1, 2028; the brand's passport stopped mattering on April 9, 2026.** Taiwan was published at 32% in the April 2, 2025 reciprocal order, Japan at 24% and the EU at 20% ([EO 14257, Annex I](https://www.federalregister.gov/documents/2025/04/07/2025-06063/regulating-imports-with-a-reciprocal-tariff-to-rectify-trade-practices-that-contribute-to-large-and)); the August 7, 2025 revision cut Taiwan to 20% and Japan to 15% ([EO 14326, Annex I](https://www.federalregister.gov/documents/2025/08/06/2025-15010/further-modifying-the-reciprocal-tariff-rates)), and the Japan agreement made Japan's 15% an all-in cap that absorbed the 4.2% column 1 rate ([EO 14345](https://www.federalregister.gov/documents/2025/09/09/2025-17389/implementing-the-united-states-japan-agreement)). On August 18, 2025 Commerce added HTS 8457.10.00 to the Section 232 derivative list, so the steel and aluminum content of every imported machining center paid 50% while the rest paid the country rate ([BIS inclusions notice, 90 FR 40326](https://www.federalregister.gov/documents/2025/08/19/2025-15819/adoption-and-procedures-of-the-section-232-steel-and-aluminum-tariff-inclusions-process)). The Supreme Court ended the reciprocal tariffs on February 20, 2026, a 10% Section 122 surcharge filled the gap, and Proclamation 11021 moved machining centers into Annex III at 15% all-in on April 9, exempt from Section 122 and from the July 24 Section 301 tier ([Tariffs, September 2026](https://open-factory.com/articles/tariff-schedule-2026/)). Three machining-center bars at 15.0%, which is 4.2% column 1 plus 10.8% of Section 232; three lathe bars at 12.5% from Japan and 10.0% from Taiwan or Germany, because HTS 8458 has no metal tariff and the July 2026 caps do the work; and one bar at 29.2% for any of the three origins on January 1, 2028, when the Annex III tier lapses and the machine falls to column 1 plus 25% ([data/tariff-stack-machining-center.csv](data/tariff-stack-machining-center.csv)). A Mazak from Florence, a DMG Mori NHX from Davis or a Haas from Oxnard pays none of it, which is why the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) asks for the plant before it asks for the brand. The Taiwan number matters for one more reason: a Hurco is a Taiwanese machine. Hurco's 10-K says "the vast majority of our products are shipped from our manufacturing facility in Taiwan from the Port of Taichung," a 370,100 square foot plant, with Indianapolis doing final assembly on certain VMX models ([Hurco 10-K, FYE October 31, 2025](https://www.sec.gov/Archives/edgar/data/315374/000110465926002512/hurc-20251031x10k.htm)). ## A Taiwanese 5-Axis Landed in Ohio **A Hurco VM10Ui class 5-axis at MachineToolIndex's $120,000 to $160,000 estimate lands in a Dayton shop at about $180,000 before tooling, and $21,000 of that is duty only if the entered value equals your price, which it does not.** The line items and their sources are in the table. The machine price is an estimate for the class, not a Hurco quote, and the two lines marked (A) are our assumptions. | Line | $ | Basis | |---|---|---| | Machine, Hurco VM10Ui class 5-axis, new, configured | 140,000 | [MachineToolIndex estimate](https://machinetoolindex.com/machines/hurco-vm10ui), $120,000 to $160,000, not a quote | | Duty, HTS 8457.10.00, Taiwan, 15% all-in on entered value | 21,000 | Upper bound; assumes entered value equals dealer price | | Ocean freight, one 40 ft container, Shanghai to Los Angeles as proxy | 7,185 | [Drewry WCI, September 3, 2026](https://www.drewry.co.uk/supply-chain-advisors/supply-chain-expertise/world-container-index-assessed-by-drewry); oversize flat-rack costs more (A) | | Inland freight to Ohio, rigging, install, training | 12,000 | (A) $8,000 to $16,000 | | Ohio sales tax | 0 | Exempt, manufacturing machinery, [ORC 5739.02(B)(42)(g)](https://codes.ohio.gov/ohio-revised-code/section-5739.02) | | Landed in Ohio, before tooling and workholding | 180,185 | Sum; $165,000 to $205,000 across the ranges | | Comparison: Haas UMC-500, list, built in Oxnard, duty 0 | 156,995 | [Haas list via MachineToolIndex](https://machinetoolindex.com/machines/haas-umc-500); options, freight, install extra | *Source: Open Factory compilation from the documents cited in each row, September 2026. [CSV: data/landed-cost-taiwan-5axis-ohio.csv](data/landed-cost-taiwan-5axis-ohio.csv)* The duty line is the one to argue about. CBP assesses the 15% on the entered value, which for a machine imported by a builder's US subsidiary is the intercompany price, not the price the distributor quotes you, so a 15% "tariff surcharge" on your invoice recovers more duty than was paid. Drewry's $7,185 is a Shanghai to Los Angeles box. A trunnion machine often ships on a flat rack at a premium we cannot source, so the freight line is a floor. Set against a UMC-500 at $156,995 list with no duty, the Taiwanese machine's landed premium is $23,000 on our midpoint, and its whole case has to be made on the 12,000 rpm spindle and the control, which is a comparison [Quote Check](https://open-factory.com/tools/quote-check/) can run once both quotes are on the same Incoterm. ## What to Do Monday **Write three fields onto every machine-tool RFQ this quarter: the ten-digit HTS line, the plant of origin, and the entered value the importer will declare, and refuse a quote that folds them into one "tariffs included" line.** The [RFQ template for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) already has the fields; the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) returns the rate and the citation for the pair, and the difference between 15% on the entered value and 15% on your invoice is the first number you negotiate. Second, ask for the US-built equivalent by name: a DMG Mori NHX horizontal from Davis instead of the same model from Iga, a Kentucky-built Mazak instead of an Inabe-built one, and get the plant written on the order acknowledgment, because the Annex III clock runs to December 31, 2027 and a 40-week delivery ordered in March 2027 lands on the wrong side of it. Third, price the yen: the builder is planning at 150 and buying at 158, and a quote in dollars that has not moved since 2024 has 5 to 8 points of currency in it before the first discount. Fourth, put the price you actually paid into the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/); Haas is the only public list in the market. Behind the paywall: the five builders' US plants, including the one that can build **100 machines a month** in California with zero duty, the imported parts that still pay the stack on a US-assembled machine, and the five lines in a 2026 quote where the room sits, mapped to the [Bid-Tab Price Book's](https://open-factory.com/models/bid-tab-price-book/) machine-tool worked example. --- # Who Bought the EV Plants' Machines URL: https://open-factory.com/articles/ev-plant-machines-second-hand/ Date: 2026-01-10 Machinery / Debrief $34.8B of cancellations in 2025 across 61 projects, Intel Ohio to 2030-31 and TSMC's 3 nm fab to 2027, the 510-lot Lion Electric catalog, Nikola at 3 cents on the dollar, Canoo's KUKA and Manz lines, Natron's $74M battery plant, buyers by sector Companies cancelled, closed or downsized [$34.8 billion of US clean-energy projects in 2025](https://e2.org/reports/clean-economy-works-2025-year-end-analysis/), 61 of them, against $12.3 billion of new announcements, and EV and battery plants were more than $21 billion each of the loss. That is E2's count, and it is the wrong headline for a buyer, because a cancelled plant is not a pile of machines: most of the $34.8 billion was never poured or bolted down. The machines that did reach the secondary market came from a shorter list: Nikola, Canoo and Arrival, Lion Electric, Natron, Fisker, Lordstown, Proterra, iM3NY, Northvolt. We read the sale orders, the auction catalogs and the state clawback letters for each. This debrief covers the cancellation calendar month by month, what a closed EV plant sells as (510 lots, two of them machine tools), the court record on price (3% to 37% of reference value), who bought and in which sector, the clawbacks, and, behind the wall, the 2026 lots worth watching and how to bid on them. Fifty-five project rows from E2's [September](https://e2.org/wp-content/uploads/2025/10/E2-Clean-Economy-Works-September-2025-Analysis-Memo.pdf), [October](https://e2.org/wp-content/uploads/2025/11/E2-Clean-Economy-Works-October-Analysis-Memo-vfinal.pdf), [November](https://e2.org/wp-content/uploads/2025/12/E2-Clean-Economy-Works-November-Analysis-Memo-v1.pdf) and [year-end](https://e2.org/wp-content/uploads/2026/02/E2-Clean-Economy-Works-December-Year-End-Analysis-Memo-vfinal.pdf) memos sum to $34.3 billion, 98.7% of E2's $34.8 billion; February, June, December and October hold two-thirds of it, and EV and battery/storage rows are $29.3 billion. ## The Cancellation Calendar E2 counts cancellations, closures and downsizes since August 2022 and excludes "temporary delays and ownership transfers that do not reduce production capacity," so the two biggest schedule changes of the year are not in the $34.8 billion. Intel told its Ohio employees on [February 28, 2025](https://newsroom.intel.com/corporate/ohio-one-construction-timeline-update) that Mod 1 at New Albany would finish construction in 2030 and begin operations between 2030 and 2031, with Mod 2 finished in 2031 and operating in 2032, on a project Spectrum News puts at [$28 billion](https://spectrumnews1.com/oh/columbus/news/2025/02/28/intel-delays-ohio-s-chip-plant-to-2030--2031). TSMC's second Arizona fab moved the other way: slated for 2028 as recently as 2025, it is now [tooling up in mid 2026 for 3 nm production in 2027](https://www.trendforce.com/news/2025/12/18/news-tsmc-reportedly-accelerates-arizona-2nd-fab-eyes-3q26-tool-install-2027-3nm-production/), as reported by TrendForce from Nikkei. Neither puts a machine on the market. A delay parks the purchase order; a cancellation strands the asset. The EV rows are where the assets are. KORE Power [cancelled its $1.2 billion Buckeye, Arizona cell plant](https://www.canarymedia.com/articles/energy-storage/kore-power-has-a-new-plan-after-canceling-1-2b-battery-plant-in-arizona) in February and said it would sell the site and retrofit an existing factory instead, "around a year and a half" against a greenfield build. Freyr [scrapped a $2.6 billion Georgia plant](https://electrek.co/2025/02/19/freyr-rebrands-t1-energy/) the same month. Stellantis [dropped the Belvidere, Illinois battery plant](https://www.wardsauto.com/news/archive-auto-stellantis-uaw-belvidere-assembly-battery-plant/749830/), $3.2 billion on E2's ledger. GM said on [June 10, 2025](https://news.gm.com/home.detail.html/Pages/news/us/en/2025/jun/0611-plants.html) that Orion Assembly would build gas full-size SUVs and light-duty pickups from early 2027, a $4.3 billion EV downsize in E2's row. Michigan served Gotion a notice of default in September and by [October 24](https://www.freep.com/story/money/business/michigan/2025/10/24/michigan-economic-development-corporation-gotion-electric-vehicle/86874815007/) the $2.4 billion Big Rapids plant was dead. December brought [SK On's $2.8 billion Tennessee row and Ford's $1.5 billion Ohio row](https://e2.org/wp-content/uploads/2026/02/E2-Clean-Economy-Works-December-Year-End-Analysis-Memo-vfinal.pdf) as Ford and SK On [dissolved BlueOval SK](https://koreajoongangdaily.joins.com/news/2025-12-11/business/industry/SK-On-Ford-agree-to-dissolve-11B-battery-joint-venture-across-3-US-factories/2475633) and Ford's [December 15 8-K](https://www.sec.gov/Archives/edgar/data/37996/000003799625000238/exhibit99todecember15202.htm) replaced the Ohio Assembly EV van with a gas and hybrid van from 2029 and booked about $19.5 billion of EV-related special items. **Of the $29.3 billion of EV and battery rows, the only ones that put equipment up for public sale in 2025 were the bankruptcies and closures of companies that had already installed lines, and those were a few hundred million dollars of book value, not tens of billions.** GM, Ford, Stellantis and SK On keep their tooling or never ordered it; KORE, Freyr and Gotion cancelled buildings. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) reads the same event from the other side: cancelled orders for coaters, stackers and formation racks shorten the queue for the buyers still in it, which is why the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) shows automation OEM backlogs flattening while aerospace and power orders in the [USMTO record half-year](https://open-factory.com/articles/machine-tool-record-fewer-units/) rose. ## What a Closed EV Plant Sells As Lion Electric opened its Joliet, Illinois bus plant in 2023 and its US subsidiary filed [Chapter 15 in December 2024](https://en.wikipedia.org/wiki/Lion_Electric_Company). On May 29, 2025 Workingman Capital, with Hilco Commercial Industrial, ran a [timed online auction](https://workingmancapital.com/auction/lion-electric-chicago-facility/) of everything inside the 906,517 square foot building: 18% buyer's premium, payment by wire the next day, small items out by June 6, cranes, racking and paint booths out by June 13, three approved riggers named in the catalog. The [catalog](https://workingmancapital.com/wp-content/uploads/2026/07/Buyers-Guide-Web-3.pdf) runs to 510 lots and we categorized every one. Of 510 lots, 102 are automated guided vehicles and transfer shuttles, 93 are benches, carts and ladders, 72 are cranes and hoists, 48 are office and IT, 36 are test and EV-service equipment (two 40,000 lb Mohawk lifts, a dynamometer, a leak analyzer), 17 are plant air (two 2022 Atlas Copco GA160 compressors, one showing 342 hours) and 13 are paint booths and Graco metering systems; two lots are machine tools. **A vehicle plant liquidates as material handling, finishing and plant air; a battery plant liquidates as process equipment; neither is a machine-tool sale, which is why the [Used CNC Auction Index](https://open-factory.com/articles/used-cnc-auction-index/) will not see the EV bust in its lots.** The machine tools were in the engineering shops. Canoo's Torrance, California sale on [February 6, 2025](https://www.biditup.com/auctions/canoo-assets-of-leading-ev-automotive-manufacturing-facility) (BidItUp, 20% premium) carried a 2020 DMG Mori DMU 75 monoBLOCK five-axis, two more DMG Mori mills, a 2021 DMS Yellowstone gantry router the auctioneer priced at over $675,000 new, a GOM scanner and a 333 hp dynamometer: a prototype shop's list. The battery plants sold as lines. Lion's Mirabel, Quebec pack plant went to auction on [July 8, 2025](https://chargedevs.com/newswire/auction-announced-for-lion-electrics-mirabel-ev-battery-facility-with-1-7-gwh-production-line/) as a complete 2023 robotic NMC pack line rated 1.7 GWh (Fanuc robots, 21 Asterion wire bonders, five NHR 9300 test systems), offered whole first and then by lot. Natron Energy's sodium-ion plants in Santa Clara and Holland, Michigan, [about $74 million of assets by Tiger Group's count](https://soldtiger.com/sales/natron-energy-battery-equipment-auction/), carried a Frontier slot-die coater, a 2023 Hanwha cell assembly line at 10 cells a minute and a formation line with a Fanuc R-2000iC; Santa Clara sold in bulk before the November 13 close and the Holland terms barred any buyer not domiciled in the United States, citing CFIUS. Canoo's production robots stayed in Oklahoma City, where AssetBuilt's June 2026 sale of the [Canoo and Arrival assets](https://assetbuilt.com/past-auctions) spent a day on Fanuc ARCMate welding cells, a day on "dozens" of [2020 to 2022 KUKA KR 120, KR 210 and KR 280 arms](https://assetbuilt.com/auctions/former-assets-of-oklahoma-city-premiere-ev-manufacturer-day-3), and a day on Dürr adhesive systems, with a [Manz battery line](https://www.autoconnectedcar.com/2026/05/assets-of-canoo-arrival-4-sale-for-liquidation-on-5-18-ev-spac-speculation-over/) as the centerpiece. A new six-axis arm lists around the $61,000 that [A Robot Costs $61k](https://open-factory.com/articles/robot-61k-cell-150k/) works from. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) has no used-robot cell yet. ## The Court Record: Three Cents on the Dollar Auction houses never publish what a lot sold for. Bankruptcy courts do, and the 2023 to 2025 record is consistent. | Asset | Buyer | Price | Reference | Date | Source | |---|---|---|---|---|---| | Nikola Coolidge plant, Phoenix HQ, 884,000 sq ft | Lucid | $30M cash and non-cash | none published | Apr 11, 2025 | [ABC15 / PBJ](https://www.abc15.com/news/business/nikolas-coolidge-factory-sale-approved-by-bankruptcy-judge) | | Nikola trucks, parts inventory, all IP | Hyroad Energy | $3.85M | $114M estate valuation; trucks alone $51.74M | Aug 6, 2025 | [Transport Topics](https://www.ttnews.com/articles/hyroad-nikola-trucks-ip) | | Canoo, substantially all assets | WHS Energy Solutions (Aquila) | $4M cash plus release of $11M+ secured debt | $145M stated assets | Apr 9, 2025 | [TechCrunch](https://techcrunch.com/2025/04/10/canoo-ceo-can-buy-bankrupt-ev-startups-assets-judge-rules/) | | iM3NY Endicott cell plant, all assets | Musashi Auto Parts Michigan | $10M | $140M announced cost; $142M debt | Apr 4, 2025 | [ElevenFlo](https://elevenflo.com/blog/im3ny-bankruptcy) | | Fisker, 3,231 Ocean SUVs | American Lease | up to $46.25M, $2,500 to $16,500 each | $39,000 to $70,000 MSRP | Jul 16, 2024 | [TechCrunch](https://techcrunch.com/2024/07/16/fisker-bankruptcy-fleet-sale-order-approved/) | | Proterra Powered, Burlingame and Greer SC | Volvo Group | $210M before inventory adjustment | none published | Feb 1, 2024 | [Volvo](https://www.volvogroup.com/en/news-and-media/news/2024/feb/volvo-completes-acquisition-of-battery-business-from-proterra-inc--and-proterra-operating-company.html) | | Proterra Transit, Greenville SC | Phoenix Motorcars | $10M | none published | Nov 2023 | [Canary Media](https://www.canarymedia.com/articles/electric-vehicles/bankrupt-ev-bus-maker-proterra-finds-buyers-for-its-business-units) | | Lordstown Motors operating assets | LAS Capital (Burns) | $10M | 175 parties, 9 proposals, 1 qualified bid | Oct 18, 2023 | [Repairer Driven News](https://www.repairerdrivennews.com/2023/10/18/bankruptcy-court-approves-lordstown-sale-for-10m/) | | Lordstown plant, 6.2M sq ft (Foxconn) | SoftBank | $375M | Foxconn paid $230M in 2022 | Aug 2025 | [Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/softbank-acquires-foxconns-ohio-facility-to-build-stargate-ai-servers-usd375-million-deal-says-foxconn-will-continue-to-operate-the-plant) | | Meyer Burger Goodyear AZ module and Colorado cell tools | Waaree Solar Americas; Babacomari Solar North | $29M total | none published | Sep 2025 | [pv magazine USA](https://pv-magazine-usa.com/2025/09/18/meyer-burger-gives-up-on-selling-itself-as-a-whole-sells-u-s-solar-production-tools/) | | Northvolt Ett, Labs, Drei, Cuberg, Dwa, all IP | Lyten | undisclosed | none published | Aug 7, 2025 | [Northvolt](https://northvolt.com/articles/lyten-to-acquire-all-remaining-northvolt-assets-in-sweden-and-germany/) | *Source: court sale orders and buyer releases as reported; full table with status, sale type, auction house and lot notes at [data/ev-plant-liquidations.csv](data/ev-plant-liquidations.csv)* Canoo cleared at 2.8% of its own stated assets, Nikola's trucks and inventory at 3.4% of the estate's valuation, iM3NY at 7.1% of the plant's announced cost, Fisker's finished cars at 36.7% of base MSRP, and the Lordstown building at 163% of what Foxconn paid for it three years earlier. **Finished goods with a title hold value; purpose-built equipment inside a bankrupt estate does not, and the Canoo trustee's filing says why: a "glut of EV related assets" at "fire-sale prices," and no money for the "rents, security costs, and insurance necessary to maintain the integrity of the assets."** The same logic produced one qualified bid from nine proposals in Lordstown, [one "actionable offer"](https://www.ttnews.com/articles/hyroad-nikola-trucks-ip) for Nikola's trucks after Gordon Brothers and Hilco Streambank canvassed the market, and a judge in Canoo who approved the insider bid because "no one else" bid. In the [used electrical market](https://open-factory.com/articles/used-refurbished-market/) a 2011 2,000 kW genset asks 30% of new, and a used Haas VF-2 asks [59% to 89% of list](https://open-factory.com/articles/haas-vf2-true-cost/): general-purpose iron has a hundred buyers, an EV skateboard welding cell has three. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) puts the hundred-buyer price next to the quote. The court record is the price when there is one buyer. ## Who Bought Twenty-five transactions, ten sectors: seven with no disclosed buyer (the Lion, Natron and Canoo auctions, Natron's bulk sale, Nikola's private-treaty inventory), three battery makers (Lyten, SK On, Ford converting Glendale), two EV makers (Lucid, and Canoo buying Arrival's gear), two owners retooling for gas (GM, Ford), two truck and bus OEMs (Volvo, Phoenix), two fleet operators (American Lease, Hyroad), two founders buying back (Burns, Aquila), two solar buyers (Waaree, Babacomari), two data-center owners (SoftBank at Lordstown, Blackstone's QTS building a [GBP 10 billion campus on the Britishvolt site](https://www.silicon.co.uk/cloud/datacenter/blackstone-qts-data-centre-northumberland-627813)) and one auto supplier (Musashi). **The aerospace and defense buyers the trade press expected are not on any sale order, catalog or release we read. The only defense link is Lyten, which sells lithium-sulfur cells into "drone and defense markets," and AssetBuilt listing "aerospace applications" as a target use for Canoo's KUKA arms.** That squares with the machinery data. Aerospace ordered [the most machinery on record in the first half of 2026](https://open-factory.com/articles/machine-tool-record-fewer-units/), value about a third above the prior half, and it bought new five-axis machines from the Japanese builders, not used AGVs from a bus plant. The buyers who did show up wanted the building (SoftBank paid $375 million for 6.2 million square feet and the power behind it, [Manufacturing Dive](https://www.manufacturingdive.com/news/foxconn-softbank-joint-data-center-equipment-factory-apple-ai-lordstown-ohio/757995/) reports), the cells (LG Energy Solution paid [about half of $2.14 billion](https://www.koreaherald.com/article/10483092) for GM's share of the 98%-built Lansing plant), or the fleet. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) prices what SoftBank avoided by buying a powered shell, and the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) shows why a plant with an existing interconnection is worth more in 2025 than the truck factory it was in 2022. ## The Clawbacks The states that paid for the plants want the money back, and the record shows how little was ever out the door. Kentucky lent Ford and SK On [$250 million up front](https://www.lpm.org/news/2026-02-04/kentucky-lawmakers-seek-answers-on-250m-loan-after-ford-ev-battery-plant-closure) as a forgivable loan for Glendale; with the plant closed, Ford has taken on the whole loan, the first $10 million repayment falls in March 2027, and Ford has offered about $2 billion and 2,100 jobs to convert building one to battery energy storage while it markets building two. Michigan's Strategic Fund approved [$715 million for Gotion](https://www.freep.com/story/money/business/michigan/2025/10/24/michigan-economic-development-corporation-gotion-electric-vehicle/86874815007/) in 2022, about $540 million of it tax breaks; MEDC says none of the $125 million Critical Industry Program grant was disbursed, while FOX 17 reports [$23.6 million of a $175 million package](https://www.fox17online.com/news/local-news/michigan/mecosta/michigan-moves-to-pull-public-funding-for-gotion-battery-plant-near-big-rapids) had been handed over and the state wants it back. Oklahoma approved about $100 million for Canoo and paid [$1 million](https://www.readfrontier.org/stories/after-canoo-shuttered-factories-a-bill-seeks-to-keep-other-ev-companies-from-getting-cash-from-a-state-incentive-program/), repayable within 30 days if 80% of the jobs went within 18 months. Our Next Energy has [repaid $3.3 million of a $15 million Michigan loan](https://www.michigancapitolconfidential.com/news/our-next-energy-has-repaid-3-3m-out-of-15m-taxpayer-loan) against a $200 million grant that was never drawn. **Across Kentucky, Michigan and Oklahoma, about $290 million of roughly $1.3 billion in approved incentives was actually disbursed to the four projects above, and $250 million of that is one loan Ford now owes.** For a buyer the clawback letters matter for one reason: a state that has served a default notice holds a repayment claim against the site, and the clean-title rule in [How to Write an RFQ for Capital Equipment](https://open-factory.com/articles/rfq-capital-equipment/) applies to a building as much as to a machine. ## What to Do Monday Register with the five houses that ran these sales (Workingman Capital with Hilco, BidItUp, Tiger Group, AssetBuilt through BidSpotter, Gordon Brothers) and ask for the PDF catalog: the Lion catalog carried lot locations, rigging contacts and removal dates the listing page did not. Price every lot against a new quote before inspection day, using the [Quote Check](https://open-factory.com/tools/quote-check/) band for compressors, switchgear and chillers and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) for anything you would otherwise import: a used KUKA in Oklahoma owes no duty, a new one from Augsburg owes the 15% Annex III rate in [Tariffs, September 2026](https://open-factory.com/articles/tariff-schedule-2026/). Budget the premium (18% at Lion, 20% at BidItUp and AssetBuilt) and the rigger on top of the hammer, and put the removal deadline into the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) as a fixed date. Behind the paywall: the six 2026 sales we would attend, including AssetBuilt's two battery-equipment auctions in October 2026 and the **$14 million** of never-installed Nikola refueling gear nobody bought, plus the bid sheet for used robots and battery lines (inspection, rigging, warranty transfer, controller licences, the CFIUS clause), priced against the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). --- # When a Megapack Beats a Static UPS URL: https://open-factory.com/articles/bess-as-ups/ Date: 2026-01-08 Electrical / TCO Lithium BESS vs double-conversion UPS for a 20 MW hall, $/kW, footprint, 58.4% duty on Chinese cells, Powin's Chapter 11, fire code, who is buying it A 24 MW, 48 MWh block of twelve Tesla Megapack 2 XL units costs about $14.4 million installed at January 2026 prices, $10.1 million after the 30% storage credit, and the N+1 static UPS it would sit next to in a 20 MW hall costs about $17.5 million. A grid battery is on the one-line diagram of every AI hall in Texas for capex, not for resilience or the grid. Tesla's own datacenter pitch, [published November 10, 2025](https://digitalassets.tesla.com/tesla-contents/image/upload/megapack-resources-megapack-for-data-centers-powering-growth-of-ai-infrastructure.pdf), grades the UPS "Maybe" for load smoothing with "~1 year" of life under that duty, and grades its own 2-hour battery "Yes" on every row. The [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) holds every input below with its source. This piece covers what a UPS actually does for the money, the published $/kWh a battery block costs, the 48.4% a Chinese cell paid on January 8 (not the 58.4% in the headlines), the fire-code footprint, lead times for both, the Texas and Virginia bills, and, behind the wall, which configuration wins in each market and who is shipping one. Five rows, one hall: the UPS lands at $10 to $26 million with $17.5 million in the middle, the battery block at $10.5 to $20.5 million gross and $7.4 to $14.3 million after the 30% credit, and the configuration people are actually building, the block plus a slimmed UPS, at $15 to $35 million because it buys both. ## The Job a UPS Actually Does **A datacenter UPS is a 5-minute bridge that runs 8,760 hours a year, and the bridge is the cheap part.** Its job is to carry the IT load from the instant the utility drops until the gensets take it, which is [5 to 10 minutes of battery against generators that reach load in about 60 seconds](https://newsletter.semianalysis.com/p/datacenter-anatomy-part-1-electrical). To do that it double-converts every kilowatt-hour the hall consumes, AC to DC to AC, at [up to 97.5% efficiency on Schneider's Galaxy VXL](https://www.se.com/ww/en/about-us/newsroom/news/press-releases/schneider-electric-announces-galaxy-vxl-ups-the-industry's-most-compact-high-density-power-protection-system-for-ai-data-center-and-large-scale-electrical-workloads-674d295fa5833dcbf5002c43/) and [97.1% on Vertiv's Trinergy](https://www.vertiv.com/en-us/about/news-and-insights/news-releases/vertiv-unveils-next-generation-ups-to-support-high-capacity-high-availability-ai-power-demands-in-room-and-prefabricated-deployments-in-all-world-regions/), or 99% in the eco modes both vendors sell and few operators turn on. At 20 MW, 2.5% is 500 kW of heat, 4,380 MWh a year, before the battery has done anything. The hall we model is 20 MW of critical IT load with N+1 UPS: 20 Galaxy VXL frames of 1,250 kW, 25 MW installed, in five 5 MW parallel groups, with lithium strings for 5 minutes at full load, 1.67 MWh usable and about 2 MWh nameplate. The frames are [1.2 m2 each](https://www.se.com/ww/en/about-us/newsroom/news/press-releases/schneider-electric-announces-galaxy-vxl-ups-the-industry's-most-compact-high-density-power-protection-system-for-ai-data-center-and-large-scale-electrical-workloads-674d295fa5833dcbf5002c43/), 24 m2 for the lot, and the batteries and clearances behind them are the part nobody puts in a press release. Every input is in [data/assumptions.csv](data/assumptions.csv) and in the [Open Factory Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). AI training changed the job. Meta's engineers wrote a dummy-load flag into PyTorch because tens of thousands of GPUs stopping together produced [swings "on the order of tens of megawatts"](https://newsletter.semianalysis.com/p/ai-training-load-fluctuations-at-gigawatt-scale-risk-of-power-grid-blackout); Google measured a 15x difference in load variability between a cloud hall and an AI hall, 1.5 MW against 15 MW; Musk described 10 to 20 MW shifts several times a second at Memphis. Tesla's deck says a Megapack in parallel with the load cuts 70% or more of that variability with grid-forming controls and that the UPS, asked to do the same, is "cycling limited" to about a year. The UPS was sized for one outage a year, not 10 Hz. The grid sees the same machine from the other side. NERC's incident review of a July 10, 2024 fault on a Dominion 230 kV line found [about 1,500 MW of load, "exclusively data center-type load," dropped in under a minute](https://www.nerc.com/pa/rrm/ea/Documents/Incident_Review_Large_Load_Loss.pdf) as UPS systems transferred to battery and most of it stayed off for hours; Tesla's answer is a Megapack that charges through the sag so the site mimics a load that rode through. Microsoft, OpenAI and NVIDIA engineers arrived at storage from the compute side in August 2025: their power-stabilization paper puts training swings at tens to hundreds of megawatts concentrated at 0.2 to 3 Hz and concludes that ["rack-level energy storage emerges as the best option," with facility-scale BESS for the largest deployments](https://arxiv.org/abs/2508.14318). ## What a Megapack Block Costs **Published installed prices for a utility-scale battery run $73 to $427 per kWh, and the US number that matters is $219.** BloombergNEF's 2025 system survey puts the global turnkey average at [$117/kWh, with 2-hour systems at $124 and 4-hour at $110, China at $73, Europe at $177 and the United States at $219](https://www.energy-storage.news/battery-storage-system-prices-continue-to-fall-sharply-bnef-and-ember-reports-find/), down from $236 in the US a year earlier. Lazard's June 2025 LCOS assumptions for a 100 MW, 200 MWh 2-hour system are [$113 to $244/kWh for the DC block, $26 to $70/kW for the inverter and $29 to $122/kWh for EPC](https://www.lazard.com/media/eijnqja3/lazards-lcoeplus-june-2025.pdf), which sums to $155 to $401/kWh. NREL's June 2025 bottom-up estimate for 2024 is [$334/kWh for 4 hours and, by its own energy-plus-power fit of $241/kWh and $372/kW, $427/kWh for 2 hours](https://www.osti.gov/servlets/purl/2583471). Eleven benchmarks: China at $73, the US at $219, Lazard's 2-hour range at $155 to $401, and two transactions, Tesla's 2021 order page at $328/kWh for ten units and Stanwell's July 2025 award at $402/kWh, with everything else between. The two transactions are the honest anchors. When Tesla opened online ordering in July 2021 the page showed [$1,235,890 for one 3 MWh Megapack and $9,999,290 for ten, $328/kWh, falling toward $280 at 100 units](https://electrek.co/2021/07/26/tesla-reveals-megapack-prices/), with delivery the next year in California and 2023 elsewhere. Queensland's state-owned Stanwell paid [AU$747 million, US$482 million, for 324 Megapack 2 XL units, 300 MW and 1,200 MWh](https://www.energy-storage.news/tesla-megapack-xl-units-arrive-for-stanwells-1200mwh-bess-in-queensland-australia/), $402/kWh with grid connection, in a public award. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) has no US BESS or UPS awards yet. These are the rows it should have. Our base case is $300/kWh installed, US, 2-hour, with $219 low and $427 high. The unit is Tesla's current 2 XL: [2 MW and 4 MWh per unit, 200 MW and 400 MWh per acre](https://digitalassets.tesla.com/tesla-contents/image/upload/megapack-resources-megapack-for-data-centers-powering-growth-of-ai-infrastructure.pdf), so twelve units give 24 MW and 48 MWh, N+2 on power against a 20 MW hall, for $14.4 million. The successor is bigger: Megapack 3 is [5 MWh per 28-foot unit, four of them plus a transformer and switchgear make a 20 MWh Megablock, 91% round trip at medium voltage, deliveries from the second half of 2026 out of a 50 GWh-a-year Houston plant](https://www.energy-storage.news/tesla-launches-integrated-20mwh-megapack-bess-solution/). Then the credit. The One Big Beautiful Bill Act kept energy storage out of the wind and solar termination, so a behind-the-meter battery still takes the [Section 48E credit at 30% with prevailing wage, 40% with the domestic-content adder](https://www.govinfo.gov/content/pkg/PLAW-119publ21/html/PLAW-119publ21.htm), and the statute's definition of storage, property that "receives, stores, and delivers energy" of 5 kWh or more, has no grid-participation test. A static UPS has never claimed it. On $14.4 million that is $4.3 million, which is the entire gap in the first chart. The catch is Section 7701(a)(52): storage that begins construction after December 31, 2025 loses the credit if it includes material assistance from a prohibited foreign entity, and the [material assistance cost ratio for storage is 55% in 2026, 60% in 2027, 65% in 2028, 70% in 2029 and 75% after](https://www.nixonpeabody.com/insights/alerts/2025/07/24/renewable-energy-after-the-big-beautiful-bill). A Chinese-cell Megapack is a $4.3 million tax question before it is a tariff question. The same statute is why [the 30% fuel-cell ITC is the last federal subsidy for on-site power](https://open-factory.com/articles/fuel-cell-itc-last-subsidy/). ## What the UPS Costs, From Four Federal Awards **There is no public list price for a megawatt-class UPS, so we priced it from federal awards, and the range is $374 to $2,825 per kW.** USAspending shows the Navy paying [$10,594,213 on September 24, 2025 to replace five MGE EPS8000 units at Gulfport with five 750 kW parallel modules, new battery breaker cabinets and a 4,000 A main switchgear modification](https://www.usaspending.gov/award/CONT_AWD_N6945025F1412_9700_N6945024D0018_9700), $2,825/kW installed with the switchgear. The others are smaller and cleaner: [a 300 kVA Eaton 9395 delivered and installed with cabling for $315,559 in November 2021](https://www.usaspending.gov/award/CONT_AWD_15F06722C0004292_1549_-NONE-_-NONE-), $1,052/kVA; [a 300 kVA 9395 with two battery cabinets for $112,200 in March 2023](https://www.usaspending.gov/award/CONT_AWD_15F06723C0001471_1549_-NONE-_-NONE-), $374/kVA; and [two 225 kVA units, an 80 kVA unit and batteries installed for $204,840 in July 2022](https://www.usaspending.gov/award/CONT_AWD_6973GH22F01276_6920_6973GH22D00010_6920), $386/kVA. Four awards, one buyer type, no volume discount: a quote, not a distribution, and the piece says so. We use $700/kW installed for modules, strings and installation as the base, $400 low and $1,050 high, which puts the 25 MW N+1 system at $17.5 million with a $10.0 to $26.2 million range. The Navy retrofit is excluded from the range because it bought switchgear too. It stands as the warning about what a UPS replacement costs once the room exists. The [Quote Check](https://open-factory.com/tools/quote-check/) tool will price a UPS quote against these cells as soon as the Bid-Tab Book has eight of them. The strings are priced at $340/kWh, the midpoint of Lazard's C&I lithium range, which is a floor for a 12C UPS string, and replaced once at year ten. ## The Chinese Cell at 48.4%, Not 58.4% **On January 8, 2026 a Chinese lithium-ion storage cell entered at 48.4% ad valorem, four instruments stacked, and the 58.4% in every trade headline was already ten points stale.** The stack, instrument by instrument: column 1 on HTS 8507.60.00 is [3.4%](https://hts.usitc.gov/reststop/search?keyword=8507.60); USTR's four-year review raised non-EV lithium-ion batteries under 8507.60.0020 [from 7.5% to 25% on January 1, 2026, heading 9903.91.06](https://www.federalregister.gov/documents/2024/09/18/2024-21217/notice-of-modification-chinas-acts-policies-and-practices-related-to-technology-transfer); the IEEPA fentanyl duty under EO 14195 went to 20% in March 2025 and then [Executive Order 14357 cut it to 10% effective 12:01 a.m. on November 10, 2025, heading 9903.01.24](https://www.federalregister.gov/documents/2025/11/07/2025-19825/modifying-duties-addressing-the-synthetic-opioid-supply-chain-in-the-peoples-republic-of-china); and [Executive Order 14358 kept the reciprocal duty at 10% under 9903.01.25 by suspending the other 24 points to November 10, 2026](https://www.federalregister.gov/documents/2025/11/07/2025-19826/modifying-reciprocal-tariff-rates-consistent-with-the-economic-and-trade-arrangement-between-the). CBP's [CSMS #66749380 of November 7, 2025](https://content.govdelivery.com/bulletins/gd/USDHSCBP-3fa83c4) confirms both IEEPA duties still stack on top of Section 301. Three-point-four plus 25 plus 10 plus 10 is 48.4. Seven origins: China at 48.4% now and 58.4% as published, Vietnam at 23.4%, Malaysia at 22.4%, Korea and Japan at 15.0% all-in, the United States at zero. The 58.4% figure was [Energy-Storage.News' own arithmetic on May 13, 2025](https://www.energy-storage.news/us-and-china-put-90-day-pause-on-tariffs-effective-duty-on-bess-falls-to-c-60/), 3.4 plus 25 plus 20 plus 10 for 2026 and 40.9% for 2025, correct on the day and repeated through the autumn; Chosun Biz was [still printing 58.4% on September 1, 2025](https://battery-tech.net/battery-markets-news/u-s-raises-section-301-duties-on-chinese-ess-batteries/), along with the fact that matters more, that China supplies more than 80% of the US ESS battery market. Korea is the interesting column: KORUS makes column 1 free, and the [December 4, 2025 notice implementing the US-Korea deal caps column 1 plus reciprocal at 15%](https://www.federalregister.gov/documents/2025/12/04/2025-21940/), so a Korean cell pays exactly 15 (heading 9903.02.80). Japan is [15% all-in under EO 14345](https://www.federalregister.gov/documents/2025/09/09/2025-17389/implementing-the-united-states-japan-agreement). Vietnam and Malaysia carry [20% and 19% from EO 14326's Annex I](https://www.federalregister.gov/documents/2025/08/06/2025-15010/further-modifying-the-reciprocal-tariff-rates) on top of the 3.4. The full HTS-by-origin logic for a UPS, a lineup and a Megapack is in [Tariff Bill of Materials](https://open-factory.com/articles/tariff-bom-electrical/); the calculator holds the cells. The duty in dollars is smaller than the debate: 48.4% of a [$70/kWh stationary pack, BNEF's 2025 average](https://www.energy-storage.news/li-ion-battery-pack-prices-fell-8-since-last-year-despite-metals-prices-rising-bloombergnef-says/), is $34/kWh, $1.6 million on the 48 MWh block; a Korean cell at [$87/kWh](https://battery-tech.net/battery-markets-news/u-s-raises-section-301-duties-on-chinese-ess-batteries/) and 15% pays $13; a US cell pays nothing. The credit the Chinese path forfeits is $90/kWh at our base price, $4.3 million on the block. That arithmetic is why LG Energy Solution signed a [US$4.3 billion agreement in August 2025 to supply LFP storage cells from Michigan to a US customer it would not name](https://www.energy-storage.news/lg-energy-solutions-us4-3-billion-agreement-with-tesla-and-lfp-manufacturing-with-gm/), and why the vendor question in the paid section is a cell-origin question. It is also why Powin is gone. Powin, the Oregon integrator with [more than 11 GWh deployed and 6 GWh under construction](https://www.energy-storage.news/powin-files-for-chapter-11-bankruptcy-protection-and-spins-off-project-services-business/), filed Chapter 11 in New Jersey on June 9 and 10, 2025 (No. 25-16137) citing uncertainty over "levels of US import tariffs" and "the future of tax credit incentives," with [more than $300 million of debt months after a $200 million KKR facility](https://techcrunch.com/2025/06/12/battery-manufacturer-powin-files-for-bankruptcy-months-after-landing-200m-loan/) and its cells sourced from China. FlexGen bought the IP, software and spares for [$36 million, closing August 19, 2025](https://www.veritaglobal.net/powin/document/2516137251015000000000006), and the court granted the debtors' motion to reject every customer energy-supply and long-term service agreement on September 9, 2025 (Docket 843). Eleven gigawatt-hours of batteries lost their warranty in a docket entry. Any BESS-as-UPS purchase order needs a parent guarantee or a warranty bond from a counterparty that will exist in year eight, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) now scores integrators on whether they publish cell origin. ## Fire Code Is Where the Footprint Goes **A lithium UPS room and a Megapack yard answer to the same chapter of the fire code, and the yard has the easier answer.** Under IFC Section 1207, adopted by the 2022 California Fire Code and mirrored in NFPA 855, lithium-ion storage inside a shared building is capped at [600 kWh per fire area, in groups of no more than 50 kWh set 3 feet apart, unless UL 9540A large-scale fire testing supports more](https://up.codes/viewer/california/ca-fire-code-2022/chapter/12/energy-systems), the system must be listed to UL 9540, and outdoor units sit 10 feet from lot lines, public ways and buildings. Our 2 MWh of UPS strings is three fire areas' worth of the indoor cap unless the vendor's 9540A data buys relief from the AHJ, which is exactly what Schneider, Vertiv and Eaton sell their tested cabinets on. Outdoors, 48 MWh in twelve enclosures is a fence and a 10-foot line. UL 9540A is [the test method, not the listing: cell, module, unit and installation levels, with the results used for separation distances and indoor limits, and it is "explicitly cited in NFPA 855 for large-scale fire testing"](https://www.ul.com/services/ul-9540a-test-method). The 2026 edition of NFPA 855, [released in September 2025](https://www.energy-storage.news/national-fire-protection-association-releases-nfpa-855-ess-safety-standard-2026-edition/), makes large-scale fire testing, whole units burned with suppression disabled, a requirement, and adds emergency response plans, explosion control and thermal-runaway propagation protection. Every vendor on the paid list has 9540A data. Not every vendor has a large-scale burn, and the AHJ in Loudoun County will ask for it in 2027 whether the state has adopted the 2026 edition or not. The reason is Moss Landing: Vistra's 300 MW, 1,200 MWh indoor NMC plant burned on January 16, 2025, and [Orange County voted a six-month moratorium on new BESS twelve days later](https://www.energy-storage.news/orange-county-california-officials-enact-emergency-moratorium-on-new-bess-in-wake-of-fire-at-moss-landing/). An outdoor LFP yard with a burn report is a different permit conversation from a lithium room inside the hall. Footprint favors the yard by an order of magnitude once you count the room. At Tesla's [200 MW and 400 MWh per acre](https://digitalassets.tesla.com/tesla-contents/image/upload/megapack-resources-megapack-for-data-centers-powering-growth-of-ai-infrastructure.pdf), 48 MWh is 0.12 acre, about 5,200 square feet of gravel before setbacks, and Tesla's smoothing rule of thumb is 0.5 acre per 100 MW of DC load, 0.1 acre for our hall. The UPS frames are 258 square feet, but they live in conditioned gray space with battery rooms, switchgear and clearances around them, which is the space the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) prices at Turner and Townsend's [$9.8 per watt in Phoenix and Columbus and $13.3 in Silicon Valley](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends), with electrical at 54% of an air-cooled build. The [footprint and code table](data/footprint-and-code.csv) has every row with its section number. ## Lead Times: Same Window, Different Queue **A Megapack ships in 40 to 65 weeks and a 2,000 kVA modular UPS in 40 to 60, so the battery does not buy schedule. It buys a different queue.** The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) publishes contractor slot data of 30 to 48 weeks for a 500 to 1,500 kVA static UPS, 40 to 60 weeks for modular units above 2,000 kVA, 60 to 80 weeks or more for DRUPS, 52 to 72 weeks for 5/15 kV metal-enclosed switchgear and 50 to 66 weeks for a 1,500 to 2,000 kW diesel genset, against a pre-pandemic norm near 24 weeks. Tesla's windows are the only public BESS series: the [2021 order page promised next-year delivery in California and 2023 elsewhere](https://electrek.co/2021/07/26/tesla-reveals-megapack-prices/), and a Megapack 3 ordered at the September 2025 launch [delivers in the second half of 2026](https://www.energy-storage.news/tesla-launches-integrated-20mwh-megapack-bess-solution/), roughly 40 to 65 weeks. Seven classes: static UPS at 30 to 48 weeks, the Megapack at 40 to 65, modular UPS at 40 to 60, gensets at 50 to 66, switchgear at 52 to 72, DRUPS at 60 to 80 plus. The difference is who else is in the line. The UPS queue is the same Schneider, Vertiv and Eaton factories whose [switchgear now costs 50% more than in 2021](https://open-factory.com/articles/switchgear-50-percent/), and the genset queue is [sold out through 2028](https://open-factory.com/articles/gensets-sold-out-2028/). The Megapack queue is one vendor's 50 GWh Houston plant plus Lathrop and Shanghai, sold against utilities who signed 1,200 MWh at a time. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tool takes both PO dates and returns the energization month; for a hall breaking ground in Q2 2026 either path lands in Q2 2027, and the switchgear in front of both is the critical path. ## Efficiency and the Bill in Texas and Virginia **The battery's operating case is not the ancillary market in Texas. It is the capacity bill in Virginia, and it is worth $2.4 to $3.2 million a year to a 20 MW hall.** Start with the UPS. Double conversion at 97.5% on 20 MW burns 4,380 MWh a year, which at EIA's industrial rate of 6.47 cents in Texas is $283,000 and at 9.31 cents in Virginia is $408,000, every year, before service. Eco mode at 99% cuts that to 1,752 MWh and $113,000 or $163,000, and most operators leave it off because the transfer back to double conversion is where the risk sits. A Megapack in parallel with the load double-converts nothing. It loses 9% on what it cycles, [91% round trip at medium voltage](https://www.energy-storage.news/tesla-launches-integrated-20mwh-megapack-bess-solution/), which at 200 cycles a year on 48 MWh is 864 MWh, $56,000 in Texas and $80,000 in Virginia, plus O&M of [$3 to $8 per kWh-year in Lazard's model](https://www.lazard.com/media/eijnqja3/lazards-lcoeplus-june-2025.pdf), $269,000 at the midpoint. Seven bars: $3.24 million, $2.40 million and $2.43 million of avoidable capacity charge at the 2025/26 Dominion-zone, 2026/27 and 2027/28 clearing prices, $0.41 million of Texas ancillary revenue at 2025 rates, against UPS losses of $0.41 million in Virginia and $0.28 million in Texas, and $0.35 million of battery O&M and losses. The PJM math is simple and brutal: a Dominion-zone load pays capacity on its share of the five coincident peaks at the clearing price, [$444.26/MW-day in the Dominion zone for 2025/26 against $269.92 for the rest of the RTO](https://insidelines.pjm.com/pjm-capacity-auction-procures-sufficient-resources-to-meet-rto-reliability-requirement/), then [$329.17 for 2026/27](https://insidelines.pjm.com/pjm-auction-procures-134311-mw-of-generation-resources-supply-responds-to-price-signal/) and [$333.44 for 2027/28](https://insidelines.pjm.com/pjm-auction-procures-134479-mw-of-generation-resources/) at the cap for the whole footprint, so a 20 MW hall carries $2.4 to $3.2 million a year of capacity cost that a battery discharging through the five peak hours can take to near zero. The [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) does the zone math for any peak load. Two hours of storage covers a one-hour peak with margin. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) shows which Virginia tariffs let a customer's peak load contribution move and which lock it in a minimum bill. Selling the same battery into the auction as a resource pays less: PJM accredits 4-hour storage at [50% for 2026/27 and 58% for 2027/28](https://www.pjm.com/planning/resource-adequacy-planning/effective-load-carrying-capability), and a 2-hour block lower still. Behind the meter, shaving the hall's own tag is worth 100 cents on the dollar. Texas is the market everyone assumes pays the battery, and the data says it did, past tense. Modo Energy put 2023 ERCOT battery revenue near [$200,000 per MW with 85% of it from ancillary services](https://www.energy-storage.news/ercot-revenues-for-battery-storage-more-concentrated-volatile-but-down-on-last-year/); by 2024 [nearly 60% came from energy arbitrage](https://www.ess-news.com/2025/05/15/is-texas-battery-landscape-really-saturated/) as sub-dollar ancillary offers passed 6 GW; and Enverus projected ancillary revenue [falling from $149/kW in 2023 to about $17/kW in 2025, a 90% decline, with 11 GW installed by mid-2025](https://www.ess-news.com/2025/11/25/battery-energy-storage-revenues-for-ancillary-services-fall-nearly-90-in-ercot/) against ancillary markets that are [4 to 5 GW each](https://www.energy-storage.news/market-saturation-intelligent-optimisation-smarter-development-the-future-of-the-ercot-market/). At $17/kW our 24 MW block earns $0.41 million a year before charging cost, which pays the O&M and nothing else. ERCOT's market monitor counted [9,505 MW of batteries at the end of 2024](https://www.potomaceconomics.com/wp-content/uploads/2025/06/2024-State-of-the-Market-Report.pdf) and said total revenue fell that year on the growing surplus; every hall that adds a block adds to the surplus. [Lazard's ERCOT value snapshot](https://www.lazard.com/media/eijnqja3/lazards-lcoeplus-june-2025.pdf), priced on 2024 hourly data, still shows a 20.2% subsidized IRR for a solar-paired 4-hour battery, but that is a generation asset with a grid interconnection, not a hall's ride-through block. We have zero ancillary revenue for Texas in the base case and show the $17/kW case beside it in the CSV. What Texas does have is a 6.47 cent rate that makes the UPS losses cheaper, an interconnection queue the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) prices market by market, and SB 6, signed June 20, 2025, plus [ERCOT's NPRR1234 large-load interconnection standard approved May 15, 2025](https://www.ercot.com/mktrules/issues/NPRR1234), both of which draw the line at 75 MW and which the [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) follows for their curtailment and backup-disclosure terms. A 20 MW hall sits under that threshold. A campus of five does not, and a battery that can shed 20 MW on command is the cheapest way to sign the curtailment clause. ## What to Do Monday The battery is not a UPS yet, and nobody shipping one pretends otherwise: Tesla's own deck [assumes the Megapack is installed in tandem with existing generators and UPSs](https://newsletter.semianalysis.com/p/ai-training-load-fluctuations-at-gigawatt-scale-risk-of-power-grid-blackout), and its 208 units at Colossus sit alongside gas turbines and the hall's own protection. Vertiv's DynaFlex claims a [2 millisecond output response](https://www.tahawultech.com/news/vertiv-introduces-vertiv-dynaflex-battery-energy-storage-system/), which is the number that has to be proven on your bus before the static switch comes out. So the purchase order to write this quarter is the third row of the first chart: the 2-hour block for smoothing, peak shaving and the credit, with the UPS cut from N+1 at 5 minutes to N at 2 minutes and run in eco mode, and the gensets kept. Five clauses. Cell origin and material-assistance cost ratio warranted in writing, with the 55% 2026 threshold and the credit's dollar value as liquidated damages if the vendor is wrong. HTS classification and the four-instrument stack stated on the quote, checked against the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/), with the buyer holding the refund if any instrument falls. UL 9540A unit-level data plus a large-scale fire test report delivered with the submittal, because the 2026 NFPA 855 will be the AHJ's checklist before your CO. A warranty bond or parent guarantee sized to the cell replacement, given what happened to Powin's 11 GWh on September 9. And a delivery slot with liquidated damages at the Monitor's published window, because [switchgear at 52 to 72 weeks](https://open-factory.com/models/lead-time-monitor/), not the battery, is your critical path. Behind the wall: the ten-year table for both markets, where the block pays itself back **$10.5 million** net in Virginia and stays a $13.3 million cost in Texas, and the twelve named deployments, products and the bankruptcy that define who you can actually buy from. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) holds the workbook. --- # The OEM Disclosure Scorecard: Who Tells Buyers Anything URL: https://open-factory.com/articles/oem-disclosure-scorecard/ Date: 2025-12-23 Electrical / Ratings Eighteen electrical, power and cooling OEMs scored on whether they publish lead times, list prices, backlog in dollars, capacity in units and delivery performance, source beside every cell, replies printed verbatim Twelve of ninety cells. That is how many times one of the eighteen largest electrical, power and cooling OEMs selling into US factories and datacenters publishes, in a form a buyer can open without an account, one of five things a buyer needs: a dated lead time by product, a list price, a backlog in dollars, a production capacity in units, or a delivery record. The other 78 cells are 16 Partials and 62 Noes. GE Vernova is the only company that scores 2 of 5. Ten score 1. Seven score 0, and one of those, Modine, has no Partial to soften it. This is version 0.1 of the [Open Factory OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/), and this piece is the scorecard: the rule for every cell, the table, the five criteria one at a time with the document behind each cell, who started and who stopped disclosing since 2024, and behind the wall, what each Partial or No costs a buyer in practice and the exact disclosure that moves each OEM up one point. Eighteen rows, five columns, 12 Yes cells: one for a lead-time schedule, none for list prices, ten for a dollar backlog, one for capacity in units, none for delivery performance. ## What the Scorecard Measures **The scorecard measures disclosure, not product, not honesty and not intent, and every cell is a fact a reader can check by opening the document beside it.** A vendor can score zero and ship excellent switchgear on time. A vendor can score two and be late on every transformer. The five criteria are the five things a sourcing manager asks for on the first call and rarely gets in writing, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/), the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) exist because the answers are mostly No. The rule for each cell is fixed and printed with the data. Yes means the disclosure exists in public, is dated, and is at the product or segment level the criterion asks for. Partial means something exists but fails one of those tests: it is undated, it covers one product line, it appears only once a year or in one quarter, or it is quantified in a different measure, such as a percent instead of units or a group total instead of a segment. No means we found nothing in the documents we opened. A No is a statement about our search, not a proof of absence, which is why every company receives its row and its sources ten business days before publication and may reply in 200 words. Replies are printed verbatim, here and on the model page. The score is a count of Yes cells. Partials are counted next to it and not weighted, because any weight we chose would be an opinion and this is a compilation. The 90 cells with evidence, document, URL and date are in [data/scorecard.csv](data/scorecard.csv). The eighteen are Eaton, Schneider Electric, ABB, Siemens Energy, Siemens Smart Infrastructure, GE Vernova, Hitachi Energy, Powell, Vertiv, Generac, Caterpillar, Cummins, Trane, Carrier, Johnson Controls, nVent, Modine and Rockwell. Sixteen of them are in the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/); Siemens Smart Infrastructure and Rockwell replace Hubbell and Atkore because the lead-time and price columns are about the gear a plant electrician orders, and those two make it. ## The Table | OEM | Lead-time schedule | List prices | Backlog in $ | Capacity in units | Delivery performance | Yes | Partial | |---|---|---|---|---|---|---|---| | GE Vernova | Partial | No | Yes | Yes | No | 2 | 1 | | Caterpillar | No | No | Yes | Partial | No | 1 | 1 | | Powell Industries | No | No | Yes | Partial | No | 1 | 1 | | Schneider Electric | Yes | Partial | No | No | No | 1 | 1 | | Siemens (Smart Infrastructure) | Partial | No | Yes | No | No | 1 | 1 | | Siemens Energy | No | No | Yes | Partial | No | 1 | 1 | | Trane Technologies | No | No | Yes | Partial | No | 1 | 1 | | ABB | No | No | Yes | No | No | 1 | 0 | | Eaton | No | No | Yes | No | No | 1 | 0 | | Hitachi Energy | No | No | Yes | No | No | 1 | 0 | | Johnson Controls | No | No | Yes | No | No | 1 | 0 | | Generac | No | Partial | Partial | Partial | No | 0 | 3 | | Vertiv | No | No | Partial | Partial | No | 0 | 2 | | Carrier | No | No | No | Partial | No | 0 | 1 | | Cummins | No | No | No | Partial | No | 0 | 1 | | Rockwell Automation | No | No | Partial | No | No | 0 | 1 | | nVent | No | No | Partial | No | No | 0 | 1 | | Modine | No | No | No | No | No | 0 | 0 | Data: [data/01-scorecard-table.csv](data/01-scorecard-table.csv) and the cell-level [data/scorecard.csv](data/scorecard.csv) (90 rows, one URL per row). GE Vernova reaches 2 Yes plus 1 Partial; six OEMs sit at 1 plus 1; four at 1 plus 0; Generac has 3 Partials and no Yes; Modine has nothing in either column. [CSV: data/03-criterion-counts.csv](data/03-criterion-counts.csv) By criterion: lead time 1 Yes, 2 Partial, 15 No; list prices 0, 2, 16; backlog 10, 4, 4; capacity 1, 8, 9; delivery 0, 0, 18. The distribution says something about what disclosure is for. The ten Yes cells in the backlog column exist because investors ask for backlog, and the eight OEMs with a Partial in the capacity column exist because investors ask whether capex is landing. Nobody with a purchase order to place is asking on an earnings call, so the two columns a buyer would rank first, a dated lead time and a delivery record, hold one Yes between them. **The scorecard is, in effect, a map of who the OEM thinks it is reporting to, and in 89 of 90 cells the answer is not the buyer.** ## Lead Times: One Guide, One Letter, One PDF From 2018 Schneider Electric is the only OEM in the set with a Yes in the first column. Its Lead Times Guide is a PDF published on or about the second of every month. The September 2026 edition, published September 2, has [124 product rows, 104 of them with a made-to-order lead time](https://www.se.com/us/en/download/document/Monthly_Lead/), counted from release to manufacturing, with LV switchgear at 50 weeks and a disclaimer that the figures are "for informational purposes only and should not be relied upon for actual orders". The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) parses every edition, and [the launch piece](https://open-factory.com/articles/lead-time-monitor-launch/) walks through its thirteen families. It is the only OEM-published schedule in the United States for the gear this scorecard covers, and it has been for the whole period we have checked. [CSV: data/04-lead-time-publishers.csv](data/04-lead-time-publishers.csv) Schneider's newest figure is 0 months old, Siemens' is 52 months old, GE Vernova's is 98 months old, Rockwell's page has no figure to date, and fourteen OEMs publish nothing. Siemens Smart Infrastructure gets a Partial, and the reason is a date. The CompAS "Lead Time Adjustments" page that the trade press and the VAWN index cite for [LV switchgear at 54 weeks, switchboards at 32 to 41 and busduct at 27](https://sandy.compas.siemens-info.com/lead-time-adjustments_4824.html) has no date, and the letter behind it is [dated April 25, 2022](https://resources.compas.siemens-info.com/HomePage/Data/Documents/Leadtime_adjustments_external_4_25.pdf), addressed to distributors, and written for "all orders not released by May 9th 2022". A product-level figure that is four years old and undated on the page it is served from is a Partial by our rule. Siemens can move to Yes by putting a month on the page and refreshing it, which is the cheapest point available to anyone in this table. GE Vernova's Partial is stranger. A site search on gevernova.com for "lead time" returns one dated, product-level document: ["GE Lentronics Lead Times and Expediting Orders", July 2018](https://www.gevernova.com/grid-solutions/sites/default/files/2025-01/ge%20lentronics-%20leadtimes%20and%20expediting%20orders_july%202018.pdf), for grid communications multiplexers. It is exactly the form the criterion asks for, applied to the one product family in the GE Vernova portfolio that nobody in this readership is waiting on. For gas turbines the company discloses slot reservation agreements in gigawatts, which is a queue, not a schedule, and for transformers and switchgear nothing. We score what exists, so the cell is Partial, and it is the clearest illustration in the table of how narrowly a disclosure can satisfy a rule. Rockwell is the No that reads as a Yes on a search engine. Its supply-chain page says some product lines are "returning to pre-pandemic lead times", others "remain constrained", lists actions taken, and tells buyers to [work with your distributor](https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html) for "any necessary lead time information". A page about lead times with no lead time on it is a No, and the [ControlLogix piece](https://open-factory.com/articles/controllogix-36-weeks-to-stock/) is what that costs a panel shop. The remaining fourteen publish nothing we could find on their sites. The right-of-reply letters invite each to point to a page, and the cell will change if one exists. What the fourteen Noes mean for a buyer is that the only current numbers for the longest lead times come from third parties: Wood Mackenzie's survey put [power transformers at 128 weeks and GSUs at 143 in Q2 2025](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/), and a general contractor's slot data put [MV switchgear at 52 to 104 weeks by voltage class in June 2026](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) against roughly 24 before 2020. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) uses those figures because the OEMs give it nothing else. The [transformer piece](https://open-factory.com/articles/transformer-160-weeks/) and the [switchgear piece](https://open-factory.com/articles/switchgear-50-percent/) are the long versions. ## List Prices: Two Partials and a Share Price No OEM in the set publishes a list price for switchgear, a transformer, a genset above residential size, a chiller or a UPS. The whole column is sixteen Noes and two Partials. Schneider's Partial comes from the fact that se.com does print a price on some pages. The [QO120CP page](https://www.se.com/us/en/product/QO120CP/), a 20 A plug-in residential breaker sold as a consumer pack, shows 29.10 USD and "Normally stocked in distribution facility". The PowerPact HDL36100, a 100 A three-pole industrial breaker on the same site, shows no price, and no switchgear, switchboard or UPS page does either. Generac's Partial is the same shape at a different scale: its residential standby generators carry ["Starting at" prices from $3,769 to $6,769](https://www.generac.com/residential-products/standby-generators/) for 10 to 26 kW units, and its C&I and data center gensets have none. Both are real prices, dated by the page, for the smallest things each company sells. Rockwell deserves a note because a reader who opens a product page will see a dollar figure. The [1756-L83E page](https://www.rockwellautomation.com/en-us/products/details.1756-L83E.html) shows "$426.08" in the header, and it is the ROK share price, not the controller's price, which needs a distributor login. Vertiv's how-to-buy page routes to partners; ABB's and Siemens' prices sit behind eShop and Industry Mall logins; Caterpillar and Cummins quote through dealers. For contrast, in the adjacent machine-tool set, Haas publishes [a full price list](https://www.haascnc.com/shop/category/pricelist.html) with a VF-2 at $70,995, and the [Haas TCO piece](https://open-factory.com/articles/haas-vf2-true-cost/) is only possible because of it. Nothing like it exists for a 15 kV switchgear section, which is why the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is built from public procurement awards instead, and why the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) launches as a contribution form. ## Backlog: Ten Yes, and Where the Dollar Figure Hides **Backlog is the one column where disclosure is the norm: ten Yes, four Partial, four No, and the interesting part is which document the number is in.** Eight OEMs print a segment or company backlog in dollars every quarter in the release or the analyst deck: Powell ([$2.4 billion](https://www.sec.gov/Archives/edgar/data/80420/000008042026000103/ex991-powlq3xfy2026earning.htm) at June 30, 2026), Trane ([$12.1 billion](https://www.sec.gov/Archives/edgar/data/1466258/000162828026050743/exhibit991-q22026earningsr.htm)), Johnson Controls ([$21.0 billion](https://www.sec.gov/Archives/edgar/data/833444/000083344426000083/q3ex991xq3fy26earningsrele.htm) for Solutions and Services), GE Vernova Electrification ([$40.6 billion](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm)), ABB Electrification ([$13,676 million](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf)), Siemens Energy Grid Technologies ([EUR 51 billion](https://assets.siemens-energy.com/dam/d0147174-31a9-4a78-b062-b49d00428fc4/earnings-release-q3-fy2026-en-pdf_Original%20file.pdf)), Hitachi Energy ([10.3 trillion yen](https://www.hitachi.com/content/dam/hitachi/global/en/press/files/2026/07/260729/2026_1Qpre.pdf)) and Siemens Smart Infrastructure, whose Q3 FY2026 analyst slide reads ["Record backlog of EUR 23.7bn"](https://assets.new.siemens.com/siemens/assets/api/uuid:d4a50737-d5e4-4447-b81d-56f7f974db90/2026-q3-presentation-en.pdf) with a book-to-bill of 1.25. The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) ranks these each quarter. [CSV: data/06-backlog-location.csv](data/06-backlog-location.csv) Eight print quarterly, two print only in the 10-Q, two only in the 10-K, two for one product line or one quarter, and four print no dollar figure in any document we opened. The two 10-Q cases are the correction this scorecard makes to our own earlier work. Eaton's earnings release prints percentages only, [Electrical backlog up 43%](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm), and the [Book-to-Bill League Table](https://open-factory.com/articles/book-to-bill-league/) recorded Eaton as a non-discloser. Eaton's 10-Q for the same quarter, in the revenue note, says ["total backlog at June 30, 2026 was approximately $24.1 billion"](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000030/etn-20260630.htm), with 71% targeted for delivery within twelve months, and its 10-K gives Electrical Americas backlog of [$13,246 million at December 31, 2025](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000007/etn-20251231.htm). Eaton scores Yes. Caterpillar is the same pattern: its release did not mention backlog until October 2025, but every 10-Q has, and the June 30, 2026 filing shows [$72.1 billion, with $29.2 billion beyond twelve months](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm). A buyer who reads releases and not filings would miss both, and the [genset piece](https://open-factory.com/articles/gensets-sold-out-2028/) leans on the Caterpillar figure for a reason. The Partials are Partials for four different reasons. Rockwell prints a backlog once a year: [$2,878 million at September 30, 2025](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm) in the 10-K, by segment, and nothing in any quarterly release or 10-Q. Vertiv printed [$15.0 billion at December 31, 2025](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm) in its 10-K and in its February release, and then nothing: the Q1 and Q2 2026 releases and 10-Qs have no orders or backlog figure, which is the subject of [the Vertiv piece](https://open-factory.com/articles/vertiv-backlog-estimate/). nVent printed [$2.6 billion](https://www.sec.gov/Archives/edgar/data/1720635/000162828026029098/q12026nvtpressrelease.htm) in its Q1 2026 release, dropped the figure in Q2, and leaves a reader to find [$2,349.9 million by segment in the 10-K](https://www.sec.gov/Archives/edgar/data/1720635/000162828026008608/nvt-20251231.htm) and $2.5 billion of remaining performance obligations in the 10-Q. Generac gives a dollar figure for one product line only, [approximately $1.6 billion of data center backlog](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm) as of July 29, 2026, and no company or C&I figure anywhere. The four Noes have said so in writing. Cummins' 10-K Backlog section explains that open orders "are historically subject to month-to-month releases and are subject to cancellation on reasonable notice without cancellation charges and therefore are not considered firm", and offers instead ["strong demand for our data center products extending out six to eight quarters"](https://www.sec.gov/Archives/edgar/data/26172/000002617226000009/cmi-20251231.htm); its CEO told the February call ["we do not quantify the size of our back order"](https://www.fool.com/earnings/call-transcripts/2026/02/05/cummins-cmi-q4-2025-earnings-call-transcript/). Carrier prints orders growth, [about 40% in total and more than 300% for data centers](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000030/a99-q22026earningsexhibit.htm), and no backlog in release, 10-Q or 10-K. Modine says ["significant order book"](https://www.sec.gov/Archives/edgar/data/67347/000110465926088230/mod-20260729xex99d1.htm), and its 10-K Backlog of Orders section describes how segments manage production without giving a number. Schneider's H1 2026 release says ["a record backlog"](https://www.se.com/ww/en/assets/pdf/release-hy-results-2026) with no figure and its Q1 2026 revenues release does not use the word. One flag on that last cell: a third-party index cites a Schneider Energy Management backlog of EUR 21.3 billion "as of FY2025" that we could not find in any Schneider document we opened, and we could not locate the full-year 2025 results release on se.com; Schneider's reply letter asks for the document, and the cell moves to Partial if it exists. ## Who Started and Who Stopped Since 2024 [CSV: data/05-backlog-disclosure-timeline.csv](data/05-backlog-disclosure-timeline.csv) Vertiv printed a dollar backlog in five of the ten quarterly releases from Q1 2024 to Q2 2026, GE Vernova's Electrification segment in four, nVent in one, Generac in one, and Caterpillar's release in four while its 10-Q printed one in all ten. Vertiv's row is the reason the criterion exists. Its Q1 2024 release headlined a ["record high $6.3 billion backlog"](https://www.sec.gov/Archives/edgar/data/1674101/000162828024017513/q12024exhibit991vrt042424.htm). The Q2, Q3 and Q4 2024 releases printed book-to-bill ratios and no backlog dollar. The figure returned for four quarters, [$7.9 billion](https://www.sec.gov/Archives/edgar/data/1674101/000162828025018915/q12025exhibit991vrt042325.htm), $8.5 billion, [$9.5 billion](https://www.sec.gov/Archives/edgar/data/1674101/000167410125000020/q32025exhibit991vrt10222025.htm) and $15.0 billion with a book-to-bill of about 2.9, and then the April 22, 2026 release carried [no orders or backlog figure at all](https://www.sec.gov/Archives/edgar/data/1674101/000162828026026379/q12026exhibit991vrt04222026.htm), and neither has anything since. What a buyer has instead is the balance sheet: deferred revenue of [$3,633.7 million at June 30, 2026 against $1,814.7 million six months earlier](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050323/q22026exhibit991vrt07292026.htm), which is customer deposits, which is a queue with money in it. GE Vernova went the other way. Its Q2 2024 release said Electrification backlog ["grew 35% year-over-year"](https://www.sec.gov/Archives/edgar/data/1996810/000199681024000065/gev2q2024earningsrelease.htm) and gave no dollar; the segment dollar figure first appeared in the October 2025 release as ["approximately $26 billion"](https://www.sec.gov/Archives/edgar/data/1996810/000199681025000159/gevpressrelease3q25.htm) and has printed every quarter since, $30.5 billion, $38.6 billion, $40.6 billion. nVent printed once and stopped; Generac started in July 2026 with a product-line figure. **Four of the eighteen rows changed their release-level backlog practice in ten quarters, and three of the four changes were toward less or narrower disclosure at the moment their backlogs were growing fastest.** The [slot economy piece](https://open-factory.com/articles/slot-economy/) is about what buyers do when the queue is invisible. ## Capacity: One Number in Gigawatts, Eight in Percent The capacity criterion asks for the one number that would let a buyer estimate when a queue shortens: output per year in units or MVA. **One OEM gives it, eight give a percent or a multiple, and nine give dollars and jobs.** GE Vernova is the Yes, and it is for gas turbines, not for the transformers and switchgear its Electrification segment sells. The Q2 2026 release says the company is ["on track to deliver 20 GW of annual gas turbine output in the third quarter of 2026, with 24 GW in 2028"](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm), with actions underway to produce 30 GW in 2030, against gas equipment backlog and slot reservations that went from 100 to 116 GW in the quarter. A buyer can divide one by the other. Nobody else in the table lets a buyer do that. Siemens Energy nearly does. The Charlotte large power transformer plant will make 24 units a year at the start and 57 at full capacity, and repair 12 rising to 24, but those figures appear in [Manufacturing Dive](https://www.manufacturingdive.com/news/siemens-energy-large-power-transformer-facility-charlotte-north-carolina/708676/) and [Carolina Public Press](https://carolinapublicpress.org/63153/transformer-shortage-us-drives-siemens-nc-jobs-expansion/) quoting Siemens Energy, and not in the [February 2024 release](https://www.siemens-energy.com/us/en/home/press-releases/siemens_energy_addresses_shortage_US_powertransformers_invests_new_factory.html) or the [February 2026 $1 billion release](https://www.siemens-energy.com/global/en/home/press-releases/siemens-energy-is-investing--1-billion-and-creating-highly-skill.html), which give dollars, jobs and "first transformers set to be manufactured in early 2026". Units told to a reporter are a Partial; units in the release would be a Yes, and Siemens Energy already has the numbers. The other seven Partials are the earnings-call form of the same information. Caterpillar will ["triple its large reciprocating engine capacity from 2024 levels"](https://www.manufacturingdive.com/news/caterpillar-triple-power-generation-capacity-raises-2030-targets-q1-2026-earnings/819078/), per its CEO on the Q1 2026 call as reported by Manufacturing Dive. Cummins ["completed the doubling of our capacity of the 95-liter engine and genset"](https://www.fool.com/earnings/call-transcripts/2026/02/05/cummins-cmi-q4-2025-earnings-call-transcript/) in 2025. Trane has ["expanded our applied capacity 4x over the last 3 years"](https://www.fool.com/earnings/call-transcripts/2026/07/30/trane-technologies-tt-q2-2026-earnings-call-transcript/). Carrier raised Charlotte output ["by about 50%"](https://www.facilitiesdive.com/news/carrier-unveils-data-center-strategy-growth-hvac-sales-q4-quantum-leap/739923/) and can "more than double" North American output with a new plant, per its CEO as reported by Facilities Dive. Vertiv expects to ["double chiller production capacity"](https://www.prnewswire.com/news-releases/vertiv-expands-global-manufacturing-capacity-for-ai-ready-data-center-cooling-solutions-302830933.html) in Italy by the end of 2026. Generac put its data center genset capacity at more than $1 billion of annual output by Q4 2026 on [its Q1 2026 call](https://www.fool.com/earnings/call-transcripts/2026/04/29/generac-gnrc-q1-2026-earnings-call-transcript/). Powell's 10-K gives [335,000 square feet of added laydown area, a 62% increase](https://www.sec.gov/Archives/edgar/data/80420/000008042025000152/powl-20250930.htm). Each is a real quantity and none has a base a buyer can see, so none converts to units of anything. The [chiller piece](https://open-factory.com/articles/chiller-60-week-problem/) works through what "4x" and "double" mean against a $12.1 billion backlog. The nine Noes are announcements in dollars and headcount. Hitachi Energy's South Boston plant is [$457 million and 825 jobs](https://www.hitachienergy.com/news-and-events/press-releases/2026/06/hitachi-energy-breaks-ground-on-the-nation-s-largest-facility-for-the-production-of-large-power-transformers-in-south-boston-virginia) and "the nation's largest facility" for large power transformers, with no unit count. Eaton's Jonesville plant is [$340 million and 700 jobs](https://www.utilitydive.com/news/eaton-transformer-production-shortage-investment/739932/), as reported by Utility Dive. Schneider's US plan is [$700 million, eight sites and 1,000 jobs](https://www.se.com/us/en/about-us/newsroom/news/press-releases/schneider-electric-plans-to-invest-over-700-million-in-the-u-s-supporting-energy-ai-sectors-and-job-growth-67bdeb3ee4475a5955011b6a/) to "increase production of switchgear". ABB's is [$200 million for medium voltage in Europe](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf). Siemens' Fort Worth plant is [$190 million, 500,000 square feet and 800 jobs](https://news.siemens.com/fi-fi/siemens-opens-190m-ft-worth-mfg-hub-ai-infrastructure/). Johnson Controls, nVent, Modine and Rockwell describe expansion without a quantity. The [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool and the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) include the Wood Mackenzie deficit estimates because the OEMs' own announcements cannot be summed into MVA. ## Delivery Performance: Eighteen Noes No OEM in the set publishes on-time delivery, shipped-versus-quoted lead time, or any measure of whether the date on the order acknowledgement was the date the truck arrived. A full-text search of every SEC filing since January 2024 for "on-time delivery" across the thirteen US filers in the set returns a metric from none of them. GE Vernova's fifteen hits are the accounting note that some contracts carry ["liquidated damages related to milestones established for on-time delivery"](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000015/gev-20251231.htm). Johnson Controls' two are a risk factor that failure on ["on-time delivery could damage our reputation"](https://www.sec.gov/Archives/edgar/data/833444/000083344425000097/jci-20250930.htm). Eaton's 2023 annual report promises to ["improve on-time delivery to our customers"](https://www.sec.gov/Archives/edgar/data/1551182/000130817924000171/letn2023_ars.pdf) without saying from what to what. Rockwell's page has a heading, "Delivery Date Reliability", and no number under it. The non-US filers' releases and presentations do not use the phrase. This is the column that version 1 of the scorecard fills from the buyer's side. The distributor and EPC panel behind the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) will report quoted versus delivered weeks by OEM and class, printed only from cells that meet the methodology's size rules, and every OEM sees its number before it runs. Until then the column is an honest row of Noes, and the [liquidated damages piece](https://open-factory.com/articles/liquidated-damages/) is how buyers price the silence today. ## What to Do Monday Three actions follow from the table, and none needs a subscription. First, read the 10-Q, not the release. Two of the ten dollar backlogs in this table, Eaton's $24.1 billion and Caterpillar's $72.1 billion, are in the revenue note of the quarterly filing and nowhere in the press release, and the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) now pulls from both. If a vendor's release prints a percent, the filing may print the dollar. Second, write the five criteria into the RFQ. Ask, in the request, for the vendor's current published lead time for the product family with the document date, the list price the quote discounts from, and its on-time delivery rate over the last four quarters for that family. A vendor that will not answer in writing has told you its cell, and the [RFQ template piece](https://open-factory.com/articles/rfq-capital-equipment/) has the clause. Run the quote through [Quote Check](https://open-factory.com/tools/quote-check/) against the Bid-Tab bands and the Monitor's published weeks before you sign. Third, treat Schneider's guide as the reference lead time for LV gear regardless of whose gear you are buying, because it is the only dated one. When an Eaton or Siemens or ABB distributor quotes 40 weeks on a switchboard, [Schneider's September 2026 guide](https://www.se.com/us/en/download/document/Monthly_Lead/) at 18 to 34 weeks by frame is the number to hold up, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) will have the October edition on the first business day of the month. Behind the paywall: eighteen entries, one per OEM, on what each Partial and No costs a buyer at the point of purchase, in weeks, deposits and change orders, and the single disclosure that moves each OEM up one point at the next filing season. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) counts **12 Yes cells of 90** today. The paid section names the twenty cheapest cells to flip. --- # Rockwell's 250 Basis Points: How Tariffs Show Up in Your PLC Quote URL: https://open-factory.com/articles/rockwell-250bps/ Date: 2025-12-19 Procurement / Take 100 bps tariff pass-through, distributor discount structure, Siemens and Schneider list behaviour, what to ask the distributor before renewing the frame agreement "For the full year, we still expect about 250 basis points of price realization, with about one hundred basis points from tariff-related pricing and about 150 basis points from underlying price." That is Rockwell Automation's CFO, Christian Rothe, in the [prepared remarks](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2026/rok-prepared-remarks-q3-fy26.pdf) for the August 4, 2026 call. **The tariff is 100 basis points of your Rockwell bill in fiscal 2026, it sits on list price by product group code, and Rockwell's letters call it non-negotiable.** The other 150 basis points is ordinary price, and the September 6, 2026 list action is priced to memory and printed circuit boards, not customs. This piece reads the fiscal 2026 calls and letters, the Siemens and Schneider equivalents, the duty a PLC cabinet pays by origin, and what to ask before the frame agreement renews. Rockwell's 10-K filings put price at about 1 point of sales in [fiscal 2021](https://www.sec.gov/Archives/edgar/data/1024478/000102447821000116/rok-20210930.htm), 3.3 in [fiscal 2022](https://www.sec.gov/Archives/edgar/data/1024478/000102447822000110/rok-20220930.htm), 5.5 in [fiscal 2023](https://www.sec.gov/Archives/edgar/data/1024478/000102447823000112/rok-20230930.htm), 2 in [fiscal 2024](https://www.sec.gov/Archives/edgar/data/1024478/000102447824000114/rok-20240930.htm) and 3 in [fiscal 2025](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm), which compounds to 18.5% by the end of fiscal 2026 at the 250 basis point guide. The BLS index for switchgear, switchboard and industrial controls ([WPU1175](https://download.bls.gov/pub/time.series/wp/wp.item)) averaged 401.5 in the ten months to July 2026 against 234.8 in fiscal 2020, up 71%, because it carries the gear in [Why Your Switchgear Costs 50% More Than in 2021](https://open-factory.com/articles/switchgear-50-percent/). A PLC buyer has had an easier five years than a switchgear buyer. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) tracks both. ## How the 250 Basis Points Were Built The guide moved three times. On November 6, 2025, Rothe said "[We expect a couple of points of price for fiscal 2026, 1% on underlying price and 1% from tariff price](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2025/ROK-Prepared-Remarks-Q4-FY25.pdf)," after a quarter in which "about 4 points of our organic growth came from price, with about 1 point of that coming from tariff-based pricing." On February 5, 2026 it was still "[two points of total price](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2026/ROK-Prepared-Remarks-Q1-FY26.pdf)." On May 5, 2026 it became "[250 basis points of total price for fiscal 2026, with 150 basis points coming from underlying price and 100 basis points from tariff-based price. This is an increase of 50 basis points from our prior outlook all from underlying price](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2026/rok-pepared-remarks-q2-fy26.pdf)," and the reason given was memory, "a double-digit million-dollar headwind in the back half." The 10-Q series runs from 1 point of price in the quarter to December 2021 to 7 points in the [quarter to December 2022](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001024478&type=10-Q), back to 1 point in the [quarter to December 2024](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000010/rok-20241231.htm), then 3 points for four quarters as tariffs arrived, and 1 point in the [quarter to June 2026](https://www.sec.gov/Archives/edgar/data/1024478/000102447826000030/rok-20260630.htm), when "core price/cost was unfavorable in the quarter, reflecting rising costs and the timing of price increases." The answer was a list increase "late in Q3 that will be realized in Q4," the September 6, 2026 action. **The tariff points are designed to make Rockwell whole, not richer.** The 10-K says "[tariff costs are expected to be neutral to EPS in fiscal 2026](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm)," the May 2025 remarks said "[the objective of tariff-based price changes is the recovery of the incremental cost, and not sales growth](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2025/ROK-Q2-FY25-Prepared-Remarks.pdf)", and Rothe told the [August 2026 call](https://www.investing.com/news/transcripts/earnings-call-transcript-rockwell-automation-beats-q3-2026-estimates-shares-fall-93CH-4834705) that "tariff-based pricing is really there to create EPS neutrality around tariff-based cost. It's not really all that incremental." In aggregate you pay Rockwell's customs bill and nothing more; on your bill of materials, an adder set by price group code has no reason to match the duty on the box in front of you, and the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) is where to check the gap. ## Eleven Letters in Nineteen Months Rockwell posts customer letters under legal notices and distributors post the letters to distributors, so the actions are on the record even when the percentages are not. The [February 3, 2025 letter](https://www.wexoe.dk/wp-content/uploads/2025/02/2025-Feb-3-STD-PRD-Letter-to-Distributors-of-Rockwell-Automation.pdf) set, effective February 1, "a 1.5% List Price adjustment to a subset of the Rockwell Automation portfolio" for the 10% China tariff, "non-negotiable," plus a consolidation of the March, June and September quarterly adjustments into one action worth "approximately a 3% additional increase." The affected price group codes include 71E, Advanced Processors and I/O, where a ControlLogix 5580 lives. *[Data: rockwell-price-increases.csv](data/rockwell-price-increases.csv)* The table holds 14 actions from May 2024 to September 2026, 9 of them Rockwell's; Rockwell published a percentage on 6, Siemens on none. The [March 4, 2025 letter](https://craigcor.com/wp-content/uploads/2025/03/2025-Mar-4-STD-PRD-Letter-to-Distributors-of-Rockwell-Automation.pdf) is the most specific document any of the three has issued: "the number of total PGCs impacted is 228, while 617 PGCs are not impacted by these tariffs at all. The incremental China tariff will result in a 1.5% increase on 166 of the 228 PGCs, the Mexico tariff will result in a 3.0% increase on 176 of the PGCs, and the Canada tariff will result in a 0.75% increase on 21 of the PGCs." The [March 7 letter](https://craigcor.com/wp-content/uploads/2025/03/2025-3-7-STD-PRD-Letter-to-Distributors-of-Rockwell-Automation.pdf) "fully retracts the Canada and Mexico tariff-based increases" after the USMCA pause, while "the China tariff-based increases will need to remain in place." Distributor HESCO's [June 9, 2025 summary](https://blog.hesconet.com/price-and-tariff-adjustment-summary-updated-6/9/25) adds a tiered 1.75% to 8.25% increase on May 18 and 2.5% on 45 legacy groups on July 5, a distributor's account that the CSV marks as such. The 2026 letters change subject. The [March 5, 2026 customer letter](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/legal-notices/2026-APA-Customer-Letter-03.05.26.pdf) adds to the April 5 annual adjustment "an additional update to reflect market pricing impacts" from post-Supreme Court tariff changes and from AI data centers driving "memory, semiconductor components, and critical metals." The [August 6, 2026 letter](https://wexoeindustry.dk/wp-content/uploads/2026/08/2026-08-06-std-prd-letter-to-distributors-of-rockwell-automation.pdf) sets the September 6 action on the Section 301 rates of "10% and 12.5%" that replaced Section 122 on July 24, plus "extraordinary cost surges in memory, semiconductive electronic components, and printed circuit boards," with percentages given only on "a Distributor Operations Special Action Call." Neither letter prints a number, the behaviour the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades. ## The Duty Versus the Adder A PLC cabinet enters under HTS 8537.10.91, and the [tariff schedule as of September 9, 2026](https://open-factory.com/articles/tariff-schedule-2026/) puts the duty at 0% from Mexico or Canada under USMCA, 9.8% from Korea, 10% from the EU and Taiwan, 12.5% from Japan, 12.7% from India, 15.2% from Vietnam and 40.2% from China, where the 2018 List 3 rate of 25% stacks on July 2026's 12.5% Section 301 action and the 2.7% column 1 rate. Controls have no Section 232 metal tariff, unlike the transformers in [Tariff Bill of Materials](https://open-factory.com/articles/tariff-bom-electrical/). The China-linked adders total 3.0% of list on the affected price groups against a duty that was 20% when they were set and is 40.2% on a finished cabinet today, because the adder is on China content inside a product built elsewhere. The Mexico adder was 3.0% against a 25% rate and lasted three days. Ask country of origin by catalog number before accepting a tariff line: on a USMCA-qualifying assembly the duty is zero, and Rockwell's [terms of sale](https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/terms-and-conditions-of-sale.html) let the price move anyway: "Seller may at any time prior to shipment, upon notice to Buyer, adjust the prices for Products to address any increase in any additional or new tariff, duty, or similar tax affecting the Products." [Quote Check](https://open-factory.com/tools/quote-check/) asks for the origin field for this reason. ## Siemens, Schneider and Emerson Did It Differently Siemens Digital Industries spent fiscal 2025 clearing the channel: in the [Q2 FY2025 release](https://assets.new.siemens.com/siemens/assets/api/uuid:d84dc895-95ef-40a2-864c-1a083f785026/HQCOPR202505127163EN.pdf) "destocking of elevated stock levels at customers approached completion toward the end of Q2 FY 2025" in China, and in [Q3 FY2025](https://assets.new.siemens.com/siemens/assets/api/uuid:d1a425b2-b59d-401b-aebf-9c622e812f4d/HQCOPR202508017210EN.pdf) revenue was EUR 4,421 million, down 10%, at a 14.5% margin. Then [Q4 FY2025](https://assets.new.siemens.com/siemens/assets/api/uuid:3948cdd4-35e0-4c1d-8412-9aed4097b3d0/HQCOPR202511117277EN.pdf) orders up 29%, a fiscal 2026 margin guide of 15% to 19% [raised in May 2026](https://assets.new.siemens.com/siemens/assets/api/uuid:d4072f41-f21d-4287-af75-0cb72c194262/HQCOPR202605117394EN.pdf) to 17% to 19%, and [18.7% in Q3 FY2026](https://assets.new.siemens.com/siemens/assets/api/uuid:bdf8b45d-34dc-4911-ad72-5aa2a03d1d40/HQCOPR202608037433EN.pdf). Siemens Industry's [April 1, 2025 notification](https://sandy.compas.siemens-info.com/afile/4963/109/1/) put "updated list pricing" into effect May 1, 2025 for all channels, capped pre-increase orders at "a 12-month average customer run rate," and gave percentages to channel partners only. Siemens never separated tariff from underlying price in a filing. The margin recovery from 14.5% to 18.7% in four quarters is the number to hold at the table. Schneider Electric publishes a price bridge. In the [H1 2026 release](https://www.se.com/ww/en/assets/pdf/release-hy-results-2026) gross pricing on products was plus EUR 280 million, raw materials minus EUR 330 million, net tariffs minus EUR 104 million and net price minus EUR 154 million, with North America up 23% organic in the second quarter. Its letters are plainer: a "[tailored 3% average price increase](https://www.se.com/us/en/download/document/240315_Price_Increase/)" from May 5, 2024, a Canadian "[average range of 3 to 5% effective May 1, 2025](https://nedco.a.bigcontent.io/v1/static/Schneider%20EN)" applied through multipliers, and a May 1, 2025 US action that distributor Power Solutions [described](https://www.power-solutions.com/2025/04/24/tariff-related-price-increases-up-to-10-coming-may-1st/) as 10% across all products. Emerson disclosed "[Price contributed 2.5 pts](https://ir.emerson.com/_assets/_473b7e069a5f80d7422fe77ae982a0bc/emerson/db/938/10115/presentation/Emerson+2025+Q4+and+Full+Year+Earnings+Presentation.pdf)" in the September 2025 quarter, guided the same for fiscal 2026, then printed [3 points in the June 2026 quarter](https://ir.emerson.com/_assets/_473b7e069a5f80d7422fe77ae982a0bc/emerson/db/938/10242/presentation/Emerson+2026+Q3+Earnings+Presentation.pdf) and $82 million of IEEPA refunds. The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) includes all four vendors side by side. ## The Backlog Is Gone, So the Price Is Live Rockwell's backlog went from $1,557 million in September 2020 to $5,197 million in September 2022, a peak of [$5.6 billion at March 31, 2023](https://www.manufacturingdive.com/news/rockwell-automation-order-backlog-lead-times-supply-chain-improve-earnings-call/649647/), then $4,109 million, $3,091 million and [$2,878 million in September 2025](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm), with Intelligent Devices down to $704 million from $2,086 million. A backlog is a price hedge: the 7 points booked in the December 2022 quarter were charged to orders placed months earlier, and Rockwell's [January 2025 remarks](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2025/ROK-Prepared-Remarks-Q1-FY25.pdf) said it would "reprice our backlog to reflect our new price lists" for tariffs. With the product backlog "[largely cleared](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2024/ROK%20Prepared%20Remarks%20Q2%20FY24%20Final.pdf)" since May 2024, most ControlLogix orders ship at the list in force on the ship date; the [ControlLogix lead-time cycle](https://open-factory.com/articles/controllogix-36-weeks-to-stock/) is the same story from the delivery side, and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) reads a falling backlog as the first sign that price discipline will loosen. ## Where the Discount Actually Sits Rockwell sells "[approximately 65 percent](https://www.sec.gov/Archives/edgar/data/1024478/000102447825000116/rok-20250930.htm)" of global sales through independent distributors, the two largest about 20% of sales, and the quote you receive is the distributor's. Two mechanics in the same 10-K separate the adder from your discount. The adder is on list, and list is Rockwell's: the letters call it "non-negotiable" so it can be removed "with the same speed" if the tariff goes. Your price is not the distributor's margin but Rockwell's rebate: "Our primary incentive program provides distributors with cash rebates or account credits based on agreed amounts that vary depending on the customer to whom our distributor ultimately sells the product." **The negotiable number in a Rockwell quote is the end-user rebate Rockwell pays your distributor, and a tariff adder on list flows through the multiplier at full strength unless that rebate is re-based.** The [Grainger confession](https://open-factory.com/articles/grainger-confession/) is the MRO version of the same structure; the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands come from awards, the only public place list-minus-multiplier shows up. Rockwell's [cancellation and returns policy of March 4, 2026](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/legal-notices/Global-Order-Cancellation-Product-Returns-and-Delivery-Date-Changes-Policy-2026.pdf) keeps cancellation charges at 18% for stock and 25% for non-stock, return charges at 20% and 50%, "will not allow changes to a product's delivery date," and "reserves the right to impose purchasing limits on all orders." Rockwell "[cancelled some orders](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/company/about-us/ir/2025/ROK-Prepared-Remarks-Q3-FY25.pdf)" it judged were pre-buys, Siemens capped them at run rate, and the [RFQ for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) now puts price validity and tariff adjustment on the front page. ## What to Do Monday Mark every price group code on your bill of materials that appears on the February 1 and March 4, 2025 lists; 71E is ControlLogix processors and I/O. Ask for the April 5 and September 6, 2026 percentages by PGC, which were given on distributor calls only. Run country of origin by catalog number through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/): a Monterrey-built panel pays 0%, so a 3.0% list adder on it is a conversation, not a customs cost. Raise the IEEPA refund: the [June 2026 10-Q](https://www.sec.gov/Archives/edgar/data/1024478/000102447826000030/rok-20260630.htm) says Rockwell has filed with CBP "for a large portion of the IEEPA tariffs paid" and "expects to implement a refund process for qualified customers"; [Who Gets the IEEPA Refund](https://open-factory.com/articles/ieepa-refund-fight/) covers who qualifies. Put the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) date next to the price-validity date on the PO, and check the [Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) for the next Section 301 review before fixing a term past twelve months. Behind the paywall: the nine questions to put to the distributor before the frame agreement renews, in the order that gets answers, and the one clause that turns Rockwell's **100 basis points** of tariff price into a line you can audit against the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). --- # What Hyperscaler Nuclear PPAs Tell a Factory Negotiating a 15-Year Contract URL: https://open-factory.com/articles/nuclear-ppas-price-signal/ Date: 2025-12-18 Power / Forecast Crane at $1,916 per kW and ~$112 per MWh, Talen and Amazon $18B over 17 years, Meta's Clinton at ~$70, Vistra's 1,200 MW to an unnamed buyer, 5.8 GW contracted against 90 GW of demand, the industrial rate by state, the eight utilities offering long-term terms Microsoft is paying about $112 per MWh for 835 MW from a reactor that shut in 2019, and it signed for 20 years. That is the Jefferies estimate for the Crane restart, published five days after the deal, and Bloomberg Intelligence put it at $100 ([Bloomberg, September 25, 2024](https://www.energyconnects.com/news/renewables/2024/september/microsoft-to-pay-hefty-price-for-three-mile-island-clean-power/)). In the same month Pennsylvania's average industrial customer paid $91.20 per MWh delivered, and PJM's Western Hub cleared $58.51 for the year ([EIA-861M](https://www.eia.gov/electricity/data/eia861m/); [EIA STEO, December 2025](https://www.eia.gov/outlooks/steo/archives/dec25.pdf)). The largest, best-rated power buyers in the country chose to pay a 70 to 90% premium to wholesale for firm, carbon-free, 20-year power. **That number, not your utility's tariff, is now the reference price for any long-term contract, and the [Open Factory Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) treats it that way.** We cover the eight hyperscaler nuclear agreements and what each discloses, the arithmetic that turns Talen's $18 billion into $82 to $94 per MWh, Crane's cost floor against its price, where the deals sit against industrial rates in 14 states and against bridge power, how much of the announced datacenter load they actually cover by 2032, and, behind the wall, what a factory negotiating 15 years should ask for and which utilities will sign. The stack reaches 2,347 MW in 2027 when Crane, Clinton and Comanche Peak start, 5,771 MW in 2032 when Susquehanna and Comanche Peak hit full quantity, and 6,691 MW in 2035 with every SMR agreement counted at our assumed years. ## Eight Agreements, 6.8 GW, One Notional Price Between September 2024 and October 2025 five hyperscalers signed for nuclear output from six plants, and only one of the six sellers has put a dollar figure next to its contract. Constellation's [September 20, 2024 release](https://www.sec.gov/Archives/edgar/data/1868275/000186827524000058/ceg-202409208kexh991.htm) calls the Microsoft deal "its largest-ever power purchase agreement", 20 years, "approximately 835 megawatts", and says nothing about price. Talen's [June 11, 2025 8-K](https://www.sec.gov/Archives/edgar/data/1622536/000162828025030559/a20250611pressreleasebusin.htm) gives 1,920 MW through 2042 and its investor deck adds "~$18bn notional revenue under 17-year contract" with "price escalators through 2042" ([Talen deck, slide 3](https://www.sec.gov/Archives/edgar/data/1622536/000162828025030559/a20250611talenbusinessup.htm)). Vistra's [September 29, 2025 8-K](https://www.sec.gov/Archives/edgar/data/1692819/000119312525221774/d43667d8k.htm) sells 1,200 MW from Comanche Peak for 20 years, with options for 20 more, to "a large, investment grade company" it will not name. Eight rows sum to 6,811 MW at full quantity, of which 5,691 MW is signed PPAs on existing or restarted reactors and the rest is Kairos's 500 MW fleet, Amazon's SMR development agreements, and Meta's open request for 1 to 4 GW. | Buyer / seller | Plant | MW | Term | Start | Price | Implied $/MWh | |---|---|---|---|---|---|---| | Microsoft / Constellation | Crane (TMI-1), PA restart | 835 | 20 yr | 2027 | n/d | 98 to 115 (analysts) | | Amazon / Talen | Susquehanna, PA | 1,920 | to 2042 + options | ramp to 2032 | ~$18bn notional | 82 to 94 (ours) | | Meta / Constellation | Clinton, IL | 1,121 | 20 yr | Jun 2027 | n/d | ~70 (Jefferies) | | Undisclosed / Vistra | Comanche Peak, TX | 1,200 | 20 yr + options | Q4 2027 | n/d | n/d | | Google, CIPCO / NextEra | Duane Arnold, IA restart | 615 | 25 yr | by Q1 2029 | n/d | n/d | | Google / Kairos | SMR fleet | 500 | PPAs under MPDA | 2030 to 2035 | n/d | n/d | | Amazon / X-energy, Dominion | Cascade WA, North Anna VA | 620+ | development | early 2030s | n/d | n/d | | Meta / RFP | 1 to 4 GW new nuclear | 1,000+ | open | early 2030s | n/a | n/a | *Full rows with structure, announcement dates and sources: [data/deal-book.csv](data/deal-book.csv). n/d = not disclosed.* The remaining rows are development deals rather than supply contracts. Google and Kairos signed a master plant development agreement on October 14, 2024 for 500 MW, first unit in 2030, complete by 2035, with PPAs "for energy, ancillary services, and environmental attributes" to follow ([Kairos Power](https://www.kairospower.com/updates/google-and-kairos-power-partner-to-deploy-500-mw-of-clean-electricity-generation)). Amazon's October 2024 package is 320 MW with Energy Northwest, expandable to 960 MW, "beginning in the early 2030s", plus "at least 300 megawatts" of SMR capacity at Dominion's North Anna site ([Amazon](https://www.aboutamazon.com/news/sustainability/amazon-nuclear-small-modular-reactor-net-carbon-zero)). Meta issued a request for proposals for 1 to 4 GW of new nuclear on December 3, 2024, delivery "starting in the early 2030s", and as of this writing has announced no award ([Meta](https://sustainability.atmeta.com/blog/2024/12/03/accelerating-the-next-wave-of-nuclear-to-power-ai-innovation/)). None of these carries a price, and none delivers a megawatt before 2030, which is why the [Datacenter Electrical Roadmap 2026 to 2032](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) includes them as options, not supply. ## The Susquehanna Arithmetic: $82 to $94 per MWh Talen's deal is the only one with a revenue number, so it is the only one where the implied price is arithmetic rather than an analyst's guess. The contract ramps from a 300 MW behind-the-meter arrangement today to 840 to 1,200 MW in 2029 and 1,680 to 1,920 MW by 2032, then runs flat through 2042 ([Utility Dive, June 11, 2025](https://www.utilitydive.com/news/talen-amazon-aws-susquehanna-nuclear-data-centert/750440/)). We fill the 2026 to 2028 gap with 300, 420 and 660 MW, the 120 and 240 MW steps Talen's deck and the original 2024 agreement describe ([POWER, March 4, 2024](https://www.powermag.com/aws-acquiring-data-center-campus-powered-by-nuclear-energy/)), and run every unit at Constellation's 2024 fleet capacity factor of 94.6% ([Constellation Q4 2024 release](https://www.sec.gov/Archives/edgar/data/1868275/000186827525000020/ceg-20250218991.htm)). At the midpoint the contract delivers 25,020 MW-years, or 206 TWh, over 17 years. The low ramp gives 191 TWh and the high ramp 221 TWh. **Divide $18 billion by those volumes and Amazon is paying a flat-equivalent $87 per MWh at the midpoint, $82 on the high ramp and $94 on the low one.** Every input and the division is in [data/assumptions.csv](data/assumptions.csv) and [data/implied-price-susquehanna.csv](data/implied-price-susquehanna.csv). Two caveats cut in opposite directions. The deck says "notional", which in Talen's usage means the revenue at full contract quantity with escalators, so the year-one price is lower: a 2% annual escalator over 17 years puts the starting price about 15% below the flat equivalent, near $74. And the contract covers "carbon-free energy and capacity", so a factory comparing it to an energy-only price should add PJM capacity, which for the 2027/28 delivery year cleared at $333.44 per MW-day, about $16 per MWh for a load running at 85% ([PJM, December 17, 2025](https://www.pjm.com/-/media/DotCom/about-pjm/newsroom/2025-releases/20251217-pjm-auction-procures-134479-mw-of-generation-resources.pdf); [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/)). The structure matters as much as the price. FERC rejected the original behind-the-meter expansion from 300 to 480 MW on November 1, 2024 by a 2-1 vote, with Commissioners Christie and See finding PJM "has not met its burden" and Chairman Phillips dissenting that "access to reliable electricity is the lifeblood of those data centers" ([Utility Dive, November 4, 2024](https://www.utilitydive.com/news/ferc-interconnection-isa-talen-amazon-data-center-susquehanna-exelon/731841/)). The rewritten deal puts Susquehanna's power onto the PJM grid, makes Talen the retail generation supplier and PPL Electric the wires company, needs no FERC approval, and eliminated Amazon's option to cap at 480 MW ([Talen deck, slide 4](https://www.sec.gov/Archives/edgar/data/1622536/000162828025030559/a20250611talenbusinessup.htm)). A merchant generator acting as a licensed retail supplier under a state's competitive-supply law is the template a Pennsylvania or Texas factory can actually copy. The [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool models both routes. The 2024 version of the same deal shows how far the terms moved in fifteen months. Amazon paid $650 million for Talen's Cumulus campus on March 4, 2024, a 960 MW site next to the 2.5 GW station, with a 10-year PPA, "two 10-year options to extend the agreement", power ramping "in 120-MW increments over several years", and "a one-time option to cap commitments to 480 MW" ([POWER, March 4, 2024](https://www.powermag.com/aws-acquiring-data-center-campus-powered-by-nuclear-energy/)). The 2025 version doubled the volume, removed the cap, stretched the firm term to 17 years and attached a notional revenue figure. Every change ran in the seller's favour, and the buyer agreed because the alternative, waiting for a grid connection in PJM, is measured in years in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). ## Crane: $1,916 per kW and a $64 Premium Crane is the cleanest cost case because the capital number is public. Constellation's restart costs $1.6 billion, of which the Department of Energy is lending $1 billion at 0.375% over Treasuries, closed November 18, 2025, and the unit is now scheduled for 2027 rather than the original 2028 ([Utility Dive, November 19, 2025](https://www.utilitydive.com/news/doe-loan-constellation-crane-nuclear-restart/805923/); [Constellation Q2 2025 release](https://www.sec.gov/Archives/edgar/data/1868275/000186827525000070/ceg-20250807991.htm)). Over 835 MW that is $1,916 per kW, against $2,240 per kW for a new gas reciprocating plant in EIA's cost table and $2,670 per kW for a 2026-priced aeroderivative plant in our [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). Capital recovery at 8% over the 20-year term is $23.70 per MWh at a 94% capacity factor, plant operating cost using Talen's published Susquehanna all-in figure of under $24 as a proxy brings the floor to about $48, and the Jefferies estimate of $112 leaves a $64 premium. The premium is not a mistake by Microsoft. Jefferies' Paul Zimbardo wrote that the deal "should provide Constellation with about $785 million in annual revenue by 2030", which over 6.87 TWh a year is $114 per MWh and checks against his $112 ([Bloomberg via Energy Connects, September 25, 2024](https://www.energyconnects.com/news/renewables/2024/september/microsoft-to-pay-hefty-price-for-three-mile-island-clean-power/)). The same note put regional wind and solar PPAs near $60. **Microsoft paid roughly double the renewable price for three things a factory also wants: firmness, a 20-year term, and a PJM location where the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) shows large power transformers at 128 to 144 weeks and new gas turbines sold into 2029.** The premium is the market value of not waiting. ## Clinton at $70 and the Undisclosed Half Meta's Clinton deal is the cheapest of the three priced contracts because it buys something narrower. Constellation's [August 7, 2025 release](https://www.sec.gov/Archives/edgar/data/1868275/000186827525000070/ceg-20250807991.htm) describes a 20-year PPA "for the full output of the Clinton Clean Energy Center" beginning June 2027, with a 30 MW uprate taking the plant from 1,092 to 1,121 MW by 2029. It is a virtual PPA for clean energy attributes, not physical delivery, so it needed no regulator's signature. Jefferies valued it "around $70/MWh, or a roughly $20/MWh premium to Illinois' energy-plus-capacity market compensation from 2028 and as much as $15/MWh higher than regional wind power purchase agreements" ([Utility Dive, June 12, 2025](https://www.utilitydive.com/news/meta-constellation-ppa-could-be-first-of-many-deals-for-existing-reactors/750567/)). The same article records Morgan Stanley and Jefferies putting Crane at $98 to $115. Clinton also shows who the seller was negotiating against. The plant ran for ten years on Illinois zero-emission credits that "saved Clinton from premature closure in 2017 after years of financial losses", and Meta's contract picks up when the ratepayer subsidy ends ([Utility Dive, June 12, 2025](https://www.utilitydive.com/news/meta-constellation-ppa-could-be-first-of-many-deals-for-existing-reactors/750567/)). A buyer arriving as a subsidy expires gets a $70 price. A buyer asking a seller to spend $1.6 billion gets $112. The gap between them is the clearest single number here for a factory: the price of long-term nuclear depends on whether you are replacing someone else's money or supplying it. The other half of the contracted megawatts has no price at all. Vistra's Comanche Peak PPA begins delivery in the fourth quarter of 2027 and ramps to 1,200 MW by 2032 ([Utility Dive, September 29, 2025](https://www.utilitydive.com/news/vistra-corp-natural-gas-permian-basin-power-plant/761358/)). NextEra's 25-year Duane Arnold PPA with Google restarts a 615 MW plant in Iowa "no later than the first quarter of 2029 and perhaps as early as the fourth quarter of 2028", and the Central Iowa Power Cooperative "will purchase 50 megawatts of the plant's output on terms and conditions consistent with the Google PPA" ([NextEra Q3 2025 release](https://www.sec.gov/Archives/edgar/data/753308/000075330825000055/neeq32025exhibit99.htm)). That last clause is the most useful sentence a smaller buyer will find in any of these filings: a 50 MW co-op got the hyperscaler's terms because it was already at the table as a part-owner. We record both as "not disclosed" in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) and do not guess. ## The Price Ladder: Nuclear, Retail, Bridge Power Put the three priced deals on one axis with what a factory actually pays and the picture is uncomfortable for anyone hoping a long contract means a low price. Industrial retail averages for September 2025 from EIA's monthly sales file run $67.50 per MWh in Texas, $87.50 in Ohio, $91.20 in Pennsylvania, $90.20 nationally and $107.50 in Illinois ([EIA-861M](https://www.eia.gov/electricity/data/eia861m/); rows in [data/industrial-rates-sep-2025.csv](data/industrial-rates-sep-2025.csv)). Our five-year bridge-power model for a 100 MW site lands at $96.00 per MWh for solid oxide fuel cells, $104.20 for gas reciprocating engines and $112.00 for aeroderivative turbines at $3.78 gas ([Recips vs Turbines vs Fuel Cells](https://open-factory.com/articles/turbines-recips-fuel-cells/); [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/)). Crane at $98 to $115 brackets Illinois retail; Susquehanna at $82 to $94 straddles Pennsylvania and Ohio retail; Clinton at $70 sits just above Texas retail and $11 above PJM West's 2025 average of $58.51. Three adjustments keep the comparison honest. The retail rates are delivered, and include distribution, transmission and capacity. The PPA figures are energy, or energy and capacity in Talen's case, at the plant. A Pennsylvania factory paying $91.20 delivered is paying roughly $60 for energy and $30 for everything else, so Susquehanna's $87 is not a 5% premium to its bill, it is a 45% premium to the energy line. The bridge-power figures are all-in at the site but exclude the grid connection the site still needs for backup and for the day the bridge ends ([Behind the Meter, In Front of the Queue](https://open-factory.com/articles/behind-meter-90gw/)). And the wholesale reference is moving: EIA's December outlook has PJM West at $40.75 in 2024, $58.51 in 2025 and $63.69 in 2026, a 56% rise in two years ([EIA STEO, December 2025, Table 7a](https://www.eia.gov/outlooks/steo/archives/dec25.pdf)). The decomposition is the part most factory bills hide. On a Pennsylvania industrial invoice near $91 per MWh, roughly $60 is energy at a price that follows PJM West, about $16 is capacity at the 2027/28 clearing price of $333.44 per MW-day for a load running 85% of the hours, and the balance is distribution, transmission and riders ([PJM, December 17, 2025](https://www.pjm.com/-/media/DotCom/about-pjm/newsroom/2025-releases/20251217-pjm-auction-procures-134479-mw-of-generation-resources.pdf); split is our estimate in [data/assumptions.csv](data/assumptions.csv)). Only the first two lines are what a nuclear PPA replaces, and only the first is what a retail supplier's fixed-price offer replaces. A factory that compares a $75 energy-plus-capacity quote to a $91 delivered bill and concludes it is saving 17% has compared two different products. The [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) exists to separate them before the comparison is made. Twelve jurisdictions span $60.60 in Louisiana to $107.50 in Illinois, a 1.8x spread, before New Jersey's $139.40 and California's $254.70 are even on the chart. ## PJM States Rose 16 to 25% in a Year The year-over-year move is why the 15-year question is being asked at all. From September 2024 to September 2025, Ohio's average industrial rate went from 7.01 to 8.75 cents per kWh, up 24.8%; Pennsylvania from 7.87 to 9.12, up 15.9%; Illinois from 9.36 to 10.75, up 14.9%; Virginia from 8.96 to 9.69, up 8.1% ([EIA-861M](https://www.eia.gov/electricity/data/eia861m/)). Texas rose 10.7% to 6.75 cents and Georgia 4.8% to 7.47. The states with the largest increases are the states whose capacity price went from $28.92 to $269.92 per MW-day in one auction, a mechanism we walk through bill by bill in [A Factory Owner's Guide to the PJM Capacity Charge](https://open-factory.com/articles/pjm-capacity-factory-bill/). The mechanism is not a mystery. PJM's capacity price for the rest of the region went from $28.92 per MW-day for 2024/25 to $269.92 for 2025/26, then to the $329.17 cap for 2026/27 and $333.44 for 2027/28, and the December 17, 2025 auction that set the last of those cleared 6,623 MW short of its reliability requirement with about 5,100 MW of the year's forecast growth attributed to data centers ([PJM, December 17, 2025](https://www.pjm.com/-/media/DotCom/about-pjm/newsroom/2025-releases/20251217-pjm-auction-procures-134479-mw-of-generation-resources.pdf)). At 85% load factor $333.44 per MW-day is $16 per MWh; at a 50% load factor it is $27. An industrial customer with a poor load factor in Ohio or Pennsylvania has absorbed most of that in the twelve months on the chart, and the [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) shows what the next two delivery years add. Ohio gained 1.74 cents, Pennsylvania 1.25, Illinois 1.39 and Virginia 0.73 in twelve months, while North Carolina fell 0.19 and Georgia gained 0.34. **A PJM factory that signed a five-year fixed price in September 2024 at that month's rate is 16 to 25% ahead today, and the hyperscalers locked their number for 17 to 25 years.** That is the asymmetry a factory should carry into any negotiation: the counterparties that know the most about where power prices are going have all chosen very long fixed-price contracts at levels above today's retail. The [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) records the term each utility now demands of large loads, 12 years at AEP Ohio and Indiana Michigan Power, 14 at Dominion, up to 15 in Georgia. The utilities are asking for the same thing the hyperscalers asked for, only with the price floating. ## 5.8 GW Against 90: The Forecast Now the demand side of the first chart. Lawrence Berkeley National Laboratory's December 2024 report puts US data center consumption at 176 TWh in 2023, 4.4% of the country's electricity, and at 325 to 580 TWh in 2028, 6.7% to 12.0% ([LBNL, p. 5](https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf)). As average load that is 37 to 66 GW. Grid Strategies' November 18, 2025 report attributes "about 90 GW, or roughly 55%" of a 166 GW five-year peak-load growth forecast to data centers by 2030, a forecast that was 38 GW in the 2023 edition ([Latitude Media, November 18, 2025](https://www.latitudemedia.com/news/one-us-load-growth-estimate-quadrupled-in-just-two-years/)). BloombergNEF's December 1 forecast reaches 106 GW of data center demand by 2035, 36% above its April number ([Utility Dive, December 3, 2025](https://www.utilitydive.com/news/us-data-center-power-demand-could-reach-106-gw-by-2035-bloombergnef/806972/)). Contracted hyperscaler nuclear is 2.8 GW in 2028 against 37 to 66 GW of demand, 4.6 GW in 2030 against 90 GW of datacenter peak growth, and 6.7 GW in 2035 against 106 GW. **Our forecast: by 2032, hyperscalers will have 5.8 GW of nuclear under contract, 5.6 GW of it from reactors that were already built, and it will cover between 6% and 9% of the datacenter load every published forecast expects by then.** The other 91 to 94% comes from the grid as it stands, from the gas turbines and engines whose slots are sold through 2029 ([The $60 Million Deposit](https://open-factory.com/articles/slot-economy/); [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/)), and from fuel cells that still carry a 30% investment tax credit ([The 30% Fuel-Cell ITC](https://open-factory.com/articles/fuel-cell-itc-last-subsidy/)). Nuclear's contribution before 2035 is capped by a simple count: the US has two shut plants with credible restarts (Crane and Duane Arnold, 1,450 MW between them, both now contracted), one more restarting for two co-ops with $2.8 billion of federal loans and grants (Palisades, [Utility Dive, October 1, 2024](https://www.utilitydive.com/news/doe-usda-announce-nearly-3b-for-palisades-restart/728538/)), and a merchant fleet whose uncontracted output is being sold plant by plant. Every reactor with a hyperscaler contract in 2032 has already been named. Three consequences follow for a factory. First, the nuclear PPA is not a template a 50 MW buyer can sign. The smallest counterparty in the eight rows is the 50 MW co-op that already owned part of the plant. Second, the price it sets is a ceiling that utilities will use, not a floor a factory can demand: a regulated utility that can sell firm power to a hyperscaler at $87 to $112 through a special contract has no reason to write a factory a 15-year fixed price at $70. Third, the megawatts not covered by nuclear are the megawatts the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) exist to track, because they will be built from transformers, switchgear, turbines and engines, and every one of those queues is where a factory's order sits behind a datacenter's. The one firm, carbon-free source that is being contracted at sub-gigawatt scale is geothermal, and it is following the nuclear price logic rather than undercutting it. Fervo's Cape Station in Utah plans 100 MW of firm power online in 2026 and 500 MW in total, and the company raised $462 million on December 10, 2025 with Google in the round, bringing its total to $1.5 billion since 2017 ([Latitude Media, December 10, 2025](https://www.latitudemedia.com/news/whats-behind-fervos-half-billion-dollar-series-e/)). Its first offtake structure with a utility, NV Energy's Clean Transition Tariff approved in May 2025, delivers 115 MW to Google at a premium that Google alone pays ([Latitude Media, May 15, 2025](https://www.latitudemedia.com/news/the-clean-transition-tariff-won-approval-in-nevada-whats-next-for-fervo/)). No Cape Station price is public. A 100 MW first phase for a single buyer's premium tariff is the same pattern as the reactor deals one order of magnitude down, and it is the pattern we expect the [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) to record in Utah, Nevada and Texas before any SMR delivers a megawatt-hour. ## What to Do Monday Price your energy line, not your bill. Take twelve months of invoices, strip distribution and transmission, and add the capacity charge the [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) computes from your peak-load contribution. The result for a Pennsylvania plant is roughly $75 per MWh for energy plus capacity, which is the number to hold against Susquehanna's $87 and Clinton's $70. Then ask your utility or retail supplier for three quotes on the same day: three-year fixed, seven-year fixed, and the longest fixed term they will write. The spread between them is the utility's own forecast, and it is the one document in the negotiation you will not have to pay for. If the long quote comes back above $95 per MWh delivered in a PJM state, price the alternative before signing: a 100 MW bridge plant in our model runs $96 to $112 per MWh for five years and lands 20 to 30 months after the purchase order ([Genset TCO](https://open-factory.com/tools/genset-tco/); [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/)), and a factory that can shed 20% of load at PJM's five peak hours cuts its capacity line by the same fraction. Behind the paywall: the eight utilities and two merchant generators that have written long-term large-load contracts as of December 2025, with the threshold, term, minimum bill and whether the price is fixed for each, and the arithmetic showing that at 50 MW the gap between Meta's price and Microsoft's is **$15.6 million a year**, from the [Open Factory Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). --- # Datacenter Construction Is Eating Your Electricians URL: https://open-factory.com/articles/electricians-wages-by-metro/ Date: 2025-12-01 Procurement / Deep dive Electrician and pipefitter wages in 10 metros, Loudoun contractor payrolls +138%, the 90th percentile at $121,570, 6,350 MW under construction, travel-card and per diem terms, the clauses that protect a plant project Loudoun County's electrical contractors had 5,615 people on payroll in 2019. In the fourth quarter of 2025 they had 13,352, and by March 2026 14,022, a 138% increase in six years. That is what a few gigawatts of datacenter construction does to one trade in one place, and the same curve is now running in Franklin and Licking counties in Ohio, in Maricopa, in Washoe, in Fulton. The median electrician did not get rich: the BLS median hourly wage for the trade rose 12.5% nationally between May 2019 and May 2025 while CPI rose 25.6%. The money went to the counties with campuses, to travelers on per diem, and to the contractors that can field a named crew. If you are wiring a plant expansion in 2026 you are bidding against that market whether you know it or not. This piece covers the BLS wage tables for electricians and pipefitters in ten datacenter metros, the county payroll data that shows where the crews went, the apprenticeship pipeline that cannot refill them before 2028, the travel-card and per diem terms that set the real price, and, behind the wall, the three contract clauses that protect a plant project and the three metros where a 2026 job can still be staffed. The median electrician in Reno went from $26.48 to $35.00 an hour (+32.2%), in Columbus from $23.93 to $31.11 (+30.0%), in Portland from $39.30 to $50.53 (+28.6%) and in Phoenix from $23.23 to $29.43 (+26.7%), while Salt Lake City moved from $28.15 to $30.11 (+7.0%) and Atlanta from $25.39 to $28.20 (+11.1%). ## Ten Metros, One Wage Table We pulled the [BLS Occupational Employment and Wage Statistics](https://www.bls.gov/oes/) rows for electricians (SOC 47-2111) and for plumbers, pipefitters and steamfitters (SOC 47-2152) in the ten metros that matter for datacenter power: Washington (Loudoun), Columbus, Dallas-Fort Worth, Phoenix, Atlanta, Chicago, Memphis, Salt Lake City, Reno and Portland-Hillsboro. May 2025 came from the BLS API; May 2019 came from the archived metro tables. The sheet lives as a labor tab in the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/), next to the study fees and security deposits for the same 15 markets, because a buyer needs both numbers before a site decision. **The national median electrician earned $30.38 an hour in May 2025, up from $27.01 in May 2019, a 12.5% raise against a 25.6% rise in CPI.** Construction average hourly earnings for all employees rose 28.6% over the same window ([FRED CES2000000003](https://fred.stlouisfed.org/graph/fredgraph.csv?id=CES2000000003), $30.69 to $39.47), so the electrician median lagged its own industry by 16 points. The OEWS median is a three-year rolling sample that includes residential and service electricians, and it is a poor instrument for a market being pulled by 300 MW campuses. It is still the number your estimator's labor rate is anchored to, which is the problem. Employment tells the other half. Phoenix went from 11,130 to 17,240 electricians (+54.9%), Salt Lake City from 4,090 to 5,710 (+39.6%), Columbus from 5,610 to 7,210 (+28.5%), Atlanta from 11,080 to 13,430 (+21.2%), Dallas from 17,350 to 20,930 (+20.6%) and Washington from 13,830 to 16,580 (+19.9%). Chicago fell from 19,390 to 17,720 (a loss of 1,670) and Portland from 7,510 to 7,340. National electrician employment rose 10.0%, from 688,620 to 757,220. Dallas has the most electricians of the ten at 20,930, then Chicago at 17,720 and Phoenix at 17,240; Reno has 1,490 and Memphis 2,770, which is why a single 200 MW build in either metro clears the local book. | Metro | Electricians May 2019 | May 2025 | Median $/hr 2019 | 2025 | Change | Median annual 2025 | 90th pct annual 2025 | |---|---|---|---|---|---|---|---| | United States | 688,620 | 757,220 | $27.01 | $30.38 | +12.5% | $63,190 | $108,510 | | Washington (Loudoun) | 13,830 | 16,580 | $29.53 | $36.50 | +23.6% | $75,930 | $121,570 | | Columbus | 5,610 | 7,210 | $23.93 | $31.11 | +30.0% | $64,700 | $94,410 | | Dallas-Fort Worth | 17,350 | 20,930 | $23.75 | $28.37 | +19.5% | $59,010 | $81,660 | | Phoenix | 11,130 | 17,240 | $23.23 | $29.43 | +26.7% | $61,210 | $86,800 | | Atlanta | 11,080 | 13,430 | $25.39 | $28.20 | +11.1% | $58,650 | $84,000 | | Chicago | 19,390 | 17,720 | $40.99 | $49.21 | +20.1% | $102,350 | $124,070 | | Memphis | 2,450 | 2,770 | $23.80 | $29.27 | +23.0% | $60,890 | $84,890 | | Salt Lake City | 4,090 | 5,710 | $28.15 | $30.11 | +7.0% | $62,620 | $88,640 | | Reno | 1,310 | 1,490 | $26.48 | $35.00 | +32.2% | $72,800 | $102,250 | | Portland (Hillsboro) | 7,510 | 7,340 | $39.30 | $50.53 | +28.6% | $105,090 | $134,380 | *Source: BLS OEWS May 2019 (archived metro tables) and May 2025 (BLS API). [CSV: oes-wages-by-metro-2019-2025.csv](data/oes-wages-by-metro-2019-2025.csv)* Two things in the table are not obvious from the chart. Chicago and Portland, the two metros with a median above $49, are the two that lost electricians, so the high-scale locals are the ones exporting travelers to the Sun Belt campuses. And the four metros with a median under $30 in 2025 (Dallas, Phoenix, Atlanta, Memphis) are the four where a datacenter GC paying $60 to $70 an hour all-in plus per diem faces the least local resistance, which is exactly where the pipeline went. Salt Lake City's $30.11 is the fifth-lowest median in the table and it moved the least, 7.0% in six years, on a 39.6% increase in headcount. That combination appears nowhere else in the ten. The one-year change is where the 2025 acceleration shows, and the OEWS metro tables do not publish it cleanly, so we use the QCEW contractor payroll instead. Average annual pay per electrical-contractor employee rose 11.9% in Ohio between 2024 and 2025, 11.7% in Nevada, 8.6% in Virginia, 8.2% in Georgia, 8.0% in Illinois, 7.9% in Utah, 7.1% in Arizona and 6.5% in Texas, against 5.3% nationally and a 2.4% rise in CPI between May 2024 and May 2025 ([FRED CPIAUCSL](https://fred.stlouisfed.org/graph/fredgraph.csv?id=CPIAUCSL)). Oregon (2.2%) and Tennessee (2.9%) were the only two of the ten states below the national rate. Ohio's number is the Intel and datacenter build in Licking County arriving in the payroll data at once: contractor employment in the state went from 35,263 in 2024 to 43,064 in Q4 2025 while pay per head rose 12%, which is what a market looks like when it is paying travelers. ## The Median Hides the Traveler Market The IBEW's own magazine, in its [June 2025 cover story on the datacenter surge](https://ibew.org/electrical_worker/the-data-center-surge-a-new-generation-of-ibew-jobs/), says the quiet part: "Between 45% and 70% of the entire budget for data center construction goes to the electrical subcontractor," and "multiple locals are facing single data center projects that require two, three, sometimes four times their current membership." Local 446 in Monroe, Louisiana has 500 members and the Meta Holly Ridge campus needs 1,000 to 2,000 electricians for five to ten years; the local's scale is just under $30 an hour and the job "will likely pay far above that." Local 26 in Washington "does close to 97% of the work" in Northern Virginia. Local 24 in Baltimore, with 2,400 members, needed 1,000 on site at the Frederick campus by August 2025 with peak "double that. Or triple." That is where the wage went. The 90th percentile electrician in the Washington metro made $121,570 in May 2025, in Chicago $124,070, in Portland $134,380, against medians of $75,930, $102,350 and $105,090. A Cheyenne business manager told [Cowboy State Daily in April 2026](https://cowboystatedaily.com/2026/04/05/us-data-centers-drive-explosion-in-pay-for-electricians-some-as-much-as-260-000/) that one datacenter crew's journeymen reached $190,000 a year including benefits and per diem because "the company had bid that they wanted to maintain the same workforce." Casper-area journeymen average about $120,000 with benefits. The viral $260,000 figure is real for a few travelers in Texas and, as the same locals say, "you become a slave to that money." The gap between median and 90th percentile is $45,640 in Washington, $29,450 in Reno and $29,290 in Portland, and it is the traveler and datacenter premium that a plant project has to match or design around. **The datacenter premium is a per diem and a named crew, not a scale increase, which is why the BLS median cannot see it.** [Microsoft's Brad Smith](https://fortune.com/2026/03/02/ai-data-centers-electrician-shortage-gen-z-training-careers/) said the company has electricians commuting from 75 miles or relocating to fill sites, and called electrical talent the number one problem slowing its US datacenter expansion. Bloomberg reported in December 2025 that some Oracle datacenters for OpenAI slipped from 2027 to 2028 in part on labor, as reported by Fortune; Oracle disputed the report and said its projects remain "on schedule and on plan." Either way the crew that did not show up at Oracle's site was on somebody else's, and in 2026 that somebody is not a plant owner paying scale. ## Contractor Payrolls Tell the Real Story The QCEW county files for electrical contractors (NAICS 23821, private) are the cleanest series we found for where the crews actually are. They count employees on contractor payrolls in the county where the work is, so a traveler from Local 613 working in Loudoun shows up in Loudoun. Loudoun went from 5,615 in 2019 to 8,546 in 2024 and 13,352 in Q4 2025, and its average weekly wage in the trade was $2,383 in Q4 2025, an annualized $123,900 against $86,846 in 2019 (+42.7%). Franklin County (Columbus) went from 4,687 to 7,774 (+65.9%); Licking County, home of the New Albany campuses, was suppressed by BLS in 2019 for having too few employers to publish, had 516 in 2024, 1,402 in Q4 2025 and 1,811 in Q1 2026. Loudoun +137.8%, Fairfield County, Ohio +119.9%, Washoe +92.7%, Ellis County, Texas +83.3%, Franklin +65.9%, Maricopa +54.0% and Fulton +53.3% are the datacenter counties; Cook County fell 3.0% and Washington County, Oregon (Hillsboro) fell 9.5% even with the Hillsboro campuses, because Portland's electricians already earned $50 an hour and the work went to travelers who are counted at home. **Maricopa County alone added 9,918 electrical-contractor employees between 2019 and Q4 2025, more than the entire electrician workforce of Columbus.** Statewide the picture is the same: Arizona contractors went from 22,199 to a 2025 average of 33,813 (+52.3%), Ohio from 29,462 to 40,745 (+38.3%), Virginia from 27,967 to 37,829 (+35.3%), Utah from 16,900 to 21,975 (+30.0%), Texas from 92,711 to 115,836 (+24.9%). Illinois grew 4.9% and Maryland 1.2%. National private electrical-contractor employment rose from 964,618 to 1,127,622 (+16.9%) and average annual pay from $66,157 to an estimated $87,516 (+32.3%), so the contractor payroll beat inflation by 7 points while the OEWS median lost 13. Texas at 115,800 is larger than Ohio, Virginia and Georgia combined; Arizona's 11,600 added jobs and Ohio's 11,300 are the two biggest absolute gains outside Texas. ## Wage Growth Against Headcount Growth Put the two BLS series side by side and the metros sort into three groups. Phoenix, Salt Lake City, Columbus, Dallas and Atlanta grew headcount 20% to 55%, which means the supply response worked and wage growth stayed in the teens or twenties. Reno and Portland grew wages 29% to 32% with headcount flat or down, which is a book that emptied. Chicago is the odd one: wages up 20.1% on the highest base in the sample and headcount down 8.6%, a metro that is exporting electricians to the campuses at travel rates. Salt Lake City is the only metro where headcount growth (39.6%) exceeded wage growth (7.0%) by more than 30 points; Reno is the mirror image at 32.2% wage growth on 13.7% headcount growth. The pipeline behind those curves is documented but not evenly. CBRE counted [6,350 MW under construction in North America's primary markets at the end of 2024](https://www.datacenterknowledge.com/data-center-construction/cbre-data-center-vacancy-hits-record-low-amid-construction-boom), more than double the 3,077 MW a year earlier, with the primary markets being Northern Virginia, Atlanta, Chicago, Phoenix, Dallas-Fort Worth and Hillsboro. Atlanta alone had [3,450 MW under construction in 2024](https://www.ajc.com/business/2026/03/atlantas-data-center-rush-hits-new-peak-but-growth-has-started-to-slow/) per CBRE, more than its operating inventory. In the first half of 2025 CBRE recorded [538.6 MW of net absorption in Northern Virginia and an 80% surge in under-construction capacity](https://www.datacenterfrontier.com/colocation/article/55312558/two-lenses-on-one-market-jll-and-cbre-show-data-centers-in-a-pinch), and JLL put H1 absorption at 647 MW in Northern Virginia and 575 MW in Dallas. Every one of those megawatts is 45% to 70% electrical work by budget. The connection side of the same markets is in our [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/): Dominion quoting up to 7 years for 100 MW connections, AEP Ohio's 85% minimum bill for 12 years, Texas SB6's $100,000 screening study fee and uniform security per MW. The comparison across all 15 markets is in [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/), and the deposit math is in the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool. A developer who has posted $2.5 million of security and waits four years for a substation will pay any per diem to hit the energization date, which is why labor is the last thing a plant buyer wins on price. ## Pipefitters Are Next The mechanical side of a datacenter is chilled water, and the pipefitter series is a year or two behind the electrician series. Pipefitter medians rose 33.9% in Phoenix ($22.89 to $30.65), 30.5% in Portland, 27.8% in Washington ($27.04 to $34.55), 23.2% in Reno and 21.8% in Columbus, while Chicago rose 9.9% and Memphis 7.3%. National pipefitter employment grew 5.2%, from 442,870 to 465,840, half the electrician rate. Pipefitters beat electricians on wage growth in Phoenix (33.9% vs 26.7%), Washington (27.8% vs 23.6%), Atlanta (19.5% vs 11.1%), Salt Lake City (19.0% vs 7.0%) and Portland (30.5% vs 28.6%), and lost in Chicago, Memphis, Columbus, Reno and Dallas. **Columbus pipefitter headcount grew 42.3% and Reno 42.7% in six years, the two largest gains of any trade in any metro in this sample.** Chicago lost 10.6% of its pipefitters and Atlanta 12.5%, which for a plant with a process chiller replacement means the crew is coming from out of town at the terms in the table below. The chiller lead times that make the schedule worse are in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) and in [Chillers Are the 60-Week Problem](https://open-factory.com/articles/chiller-60-week-problem/). A 60-week chiller and a crew that costs 30% more than the estimate is the standard 2026 mechanical package. ## The Apprentice Pipeline Cannot Close the Gap by 2028 Registered apprenticeship is a four to five year program and its output is small against the demand curve. The Department of Labor's [national results](https://www.dol.gov/agencies/eta/apprenticeship/about/statistics/2021) show 81,552 completers across all occupations in FY2019, 82,379 in FY2020 and 96,915 in FY2021, with new entrants of 252,271, 222,243 and 241,849. Electricians were the largest single occupation with 47,483 active apprentices in the federal workload in FY2019 and 71,812 in the expanded RAPIDS count for FY2021. By FY2024, [679,105 apprentices were enrolled across all industries and 35.7% were in construction](https://constructioncitizen.com/blog/abc-government-registered-apprenticeship-programs-still-do-not-meet-construction-workforce/2503201); ABC estimated the construction programs produced fewer than 40,000 completers that year. Completers rose from 51,386 in FY2008 to 96,915 in FY2021, an average of about 3,500 a year, while new entrants roughly doubled from 132,782 to 241,849 over the same period. Against that: [ABC projected in January 2025](https://www.enr.com/articles/60213-construction-needs-over-439-000-new-workers-in-2025-says-abc) that construction needed 439,000 new workers in 2025 and 499,000 in 2026; the figures cited by [Fortune](https://fortune.com/2026/03/02/ai-data-centers-electrician-shortage-gen-z-training-careers/) put nearly 30% of union electricians between 50 and 70 and about 20,000 retirements a year. Google's April 2025 grant to the electrical training ALLIANCE aims to [train 100,000 electrical workers and 30,000 new apprentices and lift the pipeline 70% in five years](https://blog.google/company-news/outreach-and-initiatives/google-org/electrical-workers-ai-training/). A first-year apprentice entering in 2026 is a journeyman in 2030 or 2031. Nothing in the pipeline changes 2026 or 2027 supply. The [AGC and NCCER workforce survey released August 28, 2025](https://www.agc.org/news/2025/08/28/construction-workforce-shortages-are-leading-cause-project-delays-immigration-enforcement-affects) (about 1,400 firms) found 92% of firms hiring had trouble filling positions, 45% had project delays from labor shortages, seven in eight raised base pay, 57% said applicants lacked skills or licenses, and 28% were hit by immigration enforcement in the prior six months, 75% of firms in Georgia. That last number matters for Atlanta specifically and we come back to it behind the wall. ## Travel Cards, Per Diem and the Daybook The union hall's referral rules are public, and they define the terms under which a crew shows up in your county. We read the Local 26, Local 134 and Local 48 pages and the Oregon prevailing wage book so the table below is the language, not a summary of it. | Term | Where | What it says | Source | |---|---|---|---| | Travel letter age | IBEW Local 26 (DC, NoVA, MD) | Letter of Introduction dated within one year, paid dues receipt, termination slip; daybook signing 7:30 to 8:30 a.m. only; 30-day re-sign between the 1st and 25th | [Local 26 travelers page](https://ibewlocal26.org/member-resources/travelers/) | | Portability | IBEW Local 26 | A traveling contractor may bring a limited number of its own employees; each must have been on payroll at least 2 weeks before traveling | [Local 26 travelers page](https://ibewlocal26.org/member-resources/travelers/) | | Short call | IBEW Local 134 (Chicago) | A short call is 14 calendar days or less; a Book 2 call over 14 days in any local removes you from Local 134's residential book | [Local 134 referral rules 2024](https://www.lu134.org/PDF's/2024-REFERRAL-RULES.pdf) | | Journeyman package | IBEW Local 48 (Portland, Hillsboro) | $65.50 wage plus $33.24 fringe, $98.74 per hour, effective January 1, 2026 | [Local 48 wage information](https://www.ibew48.com/join/2026-wage-information/) | | Shift differential | Oregon BOLI prevailing wage, electricians, Jan 2025 | Swing plus 10% to 17%, graveyard plus 15% to 31%, high time 1.5x to 2x base above 50 ft | [Oregon PWR book](https://www.ibew48.com/wp-content/uploads/2025/09/2025-oregon-pwr-book.pdf) | | Zone pay | Oregon BOLI prevailing wage, Jan 2025 | Plus $1.25 per hour at 30 to 40 miles from the reference city, plus $10.00 beyond 100 miles (carpenter schedule) | [Oregon PWR book](https://www.ibew48.com/wp-content/uploads/2025/09/2025-oregon-pwr-book.pdf) | | Per diem crew | IBEW Local 415 (Cheyenne) | One datacenter crew's journeymen at about $190,000 a year including benefits and per diem; contractor bid to keep the same crew | [Cowboy State Daily](https://cowboystatedaily.com/2026/04/05/us-data-centers-drive-explosion-in-pay-for-electricians-some-as-much-as-260-000/) | | Commute radius | Microsoft | Electricians commuting from 75 miles or relocating temporarily | [Fortune](https://fortune.com/2026/03/02/ai-data-centers-electrician-shortage-gen-z-training-careers/) | | Scale vs job pay | IBEW Local 446, Meta Holly Ridge | Scale just under $30 per hour, job pays far above; 500 members vs 1,000 to 2,000 needed | [IBEW Electrical Worker](https://ibew.org/electrical_worker/the-data-center-surge-a-new-generation-of-ibew-jobs/) | | Remote-site wage signal | Storey vs Washoe County, NV | Contractor average weekly wage $2,909 in Storey (TRIC) vs $2,420 in Washoe, Q4 2025 | [BLS QCEW](https://data.bls.gov/cew/data/api/2025/4/area/32029.csv) | *[CSV: travel-card-per-diem-terms.csv](data/travel-card-per-diem-terms.csv)* The Storey County row is the cleanest per diem signal in public data: contractor payroll per employee 20% higher one county east of Reno, where the Tahoe Reno Industrial Center campuses are and where nobody lives. Every other row is a rule a plant owner can read before bidding, and none of them appears in a typical electrical subcontract exhibit. **A journeyman in Portland costs the contractor $98.74 an hour before markup, overtime or per diem, and a graveyard shift under Oregon prevailing wage adds 15% to 31% on top.** When a datacenter GC compresses a schedule it buys second and third shifts at those differentials and pays zone or per diem for anyone outside the local's radius. A plant project that budgets straight time at the OEWS median of $36.50 in Washington or $31.11 in Columbus is 40% to 60% under the real 2026 cost of a crew that shows up on the date in the schedule. The same escalation logic for material is in the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) and in [Copper at $6.77](https://open-factory.com/articles/copper-tariff-panel/); labor needs its own clause, and we give the language below. ## What a 300 MW Campus Does to the County Licking County, Ohio is the cleanest natural experiment we have. BLS suppressed its electrical-contractor employment in 2019 because the handful of firms there could be identified; by 2024 the county had 516 electrical-contractor employees, by Q4 2025 1,402, by Q1 2026 1,811, and its neighbor Franklin County went from 4,687 to 8,399 over the same window. Fairfield County, the other neighbor, went from 402 to 811. Every one of those people was hired away from somewhere: a school district electrical bid in Lancaster, a plant expansion in Newark, a residential service company in Columbus. AEP Ohio had paused new datacenter requests with 30 GW in queue in March 2023, the tariff that followed is in the [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/), and the labor market did not wait for the tariff. CPI rose 25.6% and construction hourly earnings 28.6% between May 2019 and May 2025. The national electrician median rose 12.5%, national contractor pay per employee 32.3%, and Loudoun contractor pay per employee 42.7%. **The plant owner's problem is not the median rate; it is that the marginal electrician in a datacenter county is priced at the datacenter's per diem, and a 2 MW service upgrade competes for the same 14 people.** The utility side of that upgrade, the pad-mount wait, the AHJ timeline and the CIAC, is in [The 2 MW Plant Service Upgrade Nobody Budgets Right](https://open-factory.com/articles/2mw-plant-service-upgrade/); the gear is in [Transformer Lead Times Hit 160 Weeks](https://open-factory.com/articles/transformer-160-weeks/) and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). Labor is the third queue and the only one without a published price, which is why we put the BLS rows in the model. ## Where the Estimator's Rate Comes From I sat through six months of diligence on an electrical contractor in 2024 and read every labor line in its bids. The rate on the estimate was the same number for eighteen months: the owner's sense of what a journeyman cost, not a series. That is normal. A nonunion contractor's estimator carries a blended rate per crew and updates it when someone quits for more money. A union contractor carries the local's package and adds a markup. Neither rate contains the per diem, the shift differential or the retention bonus that the datacenter GC in the next county is paying this quarter, because those are project costs, not wage rates, and they land in the change order. So the estimate you receive for a 2026 plant job in Columbus will carry a journeyman rate near the Local 683 package or the open-shop equivalent of $31 to $38 an hour base, and the crew that shows up will have been offered $10 to $20 an hour more plus $100 to $150 a day to go to New Albany instead. The contractor covers the gap by staffing your job with apprentices at a higher ratio, by stretching the schedule, or by asking for a change order in month four. The 212 public change orders we read for [What 212 Public Change Orders Show About Electrical Contractors](https://open-factory.com/articles/change-orders-212-invoices/) are that story told 212 times. The fix is not a lower rate. It is a rate that is indexed, a crew that is named, and a schedule whose float you own, which is the paid section. ## What to Do Monday Ask three questions before you sign the electrical subcontract. First, what is the labor rate assumption in the bid, by classification, and is it indexed. If it is a flat number with no index it is the estimator's guess at scale and it will be wrong by the per diem. Second, who is the foreman and how many of the journeymen are on the contractor's payroll today, in this local's jurisdiction. The Local 26 portability rule says a traveling contractor can bring people who have been on payroll two weeks, so "our crew" can mean anyone hired last month. Third, what does the schedule assume about shifts, and is the shift differential priced or hidden in a contingency. The Oregon book puts graveyard at 15% to 31% and every local has an equivalent. Then re-run the schedule. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) takes your PO dates and gear classes and returns the energization month from the Monitor's published lead times. If the switchgear lands in week 60 and the crew is only committed for weeks 30 to 50, you have bought a demobilization and a second per diem. Price the gear with [Quote Check](https://open-factory.com/tools/quote-check/) against the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) so the material line is not hiding a labor problem, and write the RFQ per [How to Write an RFQ for Capital Equipment in 2026](https://open-factory.com/articles/rfq-capital-equipment/). Change orders on electrical work are where the labor overrun lands. The public record on that is in [What 212 Public Change Orders Show About Electrical Contractors](https://open-factory.com/articles/change-orders-212-invoices/), and the delay math is in [Liquidated Damages at 0.5% a Week](https://open-factory.com/articles/liquidated-damages/). Behind the paywall: the three clauses we would put in every 2026 plant electrical subcontract (labor escalation indexed to a public BLS series with a shared band, a named-crew clause with substitution rules and a book letter, and a schedule-float clause that says who owns the float and what triggers per diem), and the three metros where a plant can still staff a 2026 project, starting with the one where the median electrician wage rose **7.0%** in six years while headcount grew 39.6%. The wage rows sit in the labor tab of the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/). --- # Why Your Switchgear Costs 50% More Than in 2021 URL: https://open-factory.com/articles/switchgear-50-percent/ Date: 2025-11-30 Electrical / Deep dive Switchgear PPI +78% and +15% YoY at peak, copper $6.14/lb and a 50% tariff, labor 27% of the factory, Eaton margin 20% to 30%, Powell backlog $2.4B, lead times 21 to 104 weeks by class, the decomposition behind the wall The BLS producer price index for switchgear and switchboard manufacturing stood at 394.8 in July 2026, [78% above January 2021](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCU335313335313). Wood Mackenzie's survey puts medium-voltage switchgear at +50% since 2021 and circuit breakers at +47%, [as reported by POWER Magazine](https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/). Either number is the largest increase for any category of electrical gear except transformers, and the buyer signing the PO gets one line for it: "base price." This piece takes the line apart. We weight every input in a switchgear factory with the Bureau of Economic Analysis cost structure for the industry, run each input's own price series from January 2021 to July 2026, and see how much of the 78 points the inputs explain. **Copper, steel, fabricated metal, bought-in components, services and labor together account for 36 of the 78 points. The other 43 points are margin.** Eaton's Electrical Americas segment went from a 20.6% operating margin in 2021 to 29.9% in 2025, and Powell Industries' gross margin went from 15.9% to 29.0% over the same years, which is what 43 points of margin looks like in a 10-K. What follows: the price series under the quote, the cost stack of a switchgear section, copper and its tariff, steel and the enclosure, labor, the decomposition, Eaton and Powell as the proof, published lead times by voltage class, and what to write into the next RFQ. Rebased to January 2021 = 100, the switchgear industry index reached 178 in July 2026, the narrower "switchgear except relays and ducts" commodity index reached 212, low-voltage panelboards reached 158, and power and distribution transformers reached 187. ## The Price Series Under the Quote There is no public transaction price for a 15 kV, 2,000 A metal-clad lineup. What exists is the BLS producer price index, a survey of what manufacturers say they received, published monthly about two weeks after month end. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes four of these series for switchgear and one for transformers, and the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is our attempt to put $/section under the index from public awards. Until that book has enough MV switchgear rows, the index is the number. The level first: [PCU335313335313](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCU335313335313) went from 221.5 in January 2021 to 394.8 in July 2026, +78.2%, and the commodity index for switchgear excluding relays and ducts, [WPU11752201A](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU11752201A), went from 123.6 to 262.6, +112.4%. The shape of the rise: the industry index rose 10.1% in 2021, 19.6% in 2022, 6.6% in 2023 and 6.2% in 2024, then re-accelerated. The recent path: year-over-year inflation was +4.6% in March 2025, +10.7% by July 2025, peaked at +15.4% in February 2026 and +15.2% in March 2026, and was still +8.5% in July 2026. Year over year, the industry index ran +4.6% in March 2025, +10.7% in July 2025, +15.4% in February 2026 and +8.5% in July 2026. The switchgear-only commodity index peaked at +22.3% in March 2026 and panelboards were +2.8% in July 2026. **The two biggest single-month jumps in the series since 2022 are April 2025 (+4.2%) and April 2026 (+3.2%), and both follow price-increase letters that OEMs issue at the start of the second quarter.** Eaton told analysts on May 5, 2026 that it had "announced and implemented April 1 price increases in Electrical Americas," per the [Q1 2026 call transcript](https://www.fool.com/earnings/call-transcripts/2026/05/05/eaton-etn-q1-2026-earnings-call-transcript/). The PPI recorded the industry's version of that letter three weeks later. Wood Mackenzie's +50% for MV switchgear and +47% for breakers come from its September 2025 report with American Clean Power, which is gated. We cite it [as reported by POWER Magazine](https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/) and do not headline it as ours. Its 2021 base and survey method differ from the BLS, so 50% and 78% are two views of the same move, not a disagreement. For a buyer the useful conversion is this: a lineup quoted at $1.00 million in January 2021 would be quoted at roughly $1.5 to $1.8 million today on either basis, before the lead-time premium. ## What a Switchgear Section Is Made Of The decomposition needs weights, and the only public cost structure for the industry is the Bureau of Economic Analysis input-output table. The [2017 benchmark Use table](https://apps.bea.gov/industry/xls/io-annual/IOUse_Before_Redefinitions_PRO_2017_Detail.xlsx) breaks NAICS 335313, switchgear and switchboard apparatus manufacturing, into 405 input commodities plus labor, taxes and operating surplus, for an industry with $12.1 billion of output that year. It is dated, but the physics of a switchgear section have not changed since 2017: steel enclosure, copper bus, breakers and relays, and people. Compensation of employees is $3.3 billion or 27.3% of output, purchased intermediate inputs $6.6 billion or 54.9%, gross operating surplus $2.0 billion or 16.7%, and taxes 1.1%. Inside the intermediates, copper and other nonferrous smelting and refining is $761 million (6.3%), iron and steel mills $646 million (5.3%), ornamental and architectural metal products, which is where enclosures sit, $530 million (4.4%), and forgings, stampings, screw products, machine shops, plate work and roll forming another $787 million (6.5%). Relays and industrial controls bought from NAICS 335314 are $429 million and switchgear bought from other switchgear makers $233 million, together 5.5%. Wholesale margins on all of that are $872 million, 7.2%. **Copper, the input every buyer asks about, is 6.3% of a switchgear factory's output on the BEA basis, and all nonferrous metal together is 8.7%.** Powell Industries lists its raw materials in the [FY2025 10-K](https://www.sec.gov/Archives/edgar/data/80420/000008042025000152/powl-20250930.htm) as "steel, copper and aluminum, as well as various engineered electrical components," in that order. Grain-oriented electrical steel, the input that gates transformers and the subject of our [Butler Works piece](https://open-factory.com/articles/butler-works-goes/), enters switchgear only through instrument transformers and is not a line in this table. ## Copper: Doubled, Then Tariffed The IMF global copper price averaged [$2.76 per pound in December 2019, $3.62 in January 2021 and $6.14 in July 2026](https://fred.stlouisfed.org/graph/fredgraph.csv?id=PCOPPUSDM), a 70% rise over the decomposition window. COMEX touched [$6.77 per pound on August 7, 2026](https://investingnews.com/daily/resource-investing/base-metals-investing/copper-investing/highest-price-for-copper/). The US producer price for copper wire and cable, [WPU10260314](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU10260314), rose 68.9% over the same window and was up 17.9% year over year in July 2026. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) converts the weekly COMEX print into $/ft by conductor size and carries the escalation-clause calculator we return to at the end. Copper was $3.62/lb when the 2021 base was set, $4.37 in June 2021, $3.80 in December 2022, $4.04 in December 2024, and $6.14 in July 2026. It is up 69.9% since January 2021 and 38.6% in the twelve months to July 2026, so most of the rise landed after the tariff. The tariff is the reason for the last leg. [Proclamation 10962 of July 30, 2025](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states) (90 FR 37727) imposed a 50% Section 232 duty on "semi-finished copper products and intensive copper derivative products" entered on or after 12:01 a.m. on August 1, 2025, applied "only to the copper content" of the article. [Proclamation 11021 of April 2, 2026](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states) (91 FR 18201) then moved the metals duties to "the full customs value of the imported product, regardless of metal content" from April 6, 2026, at 50% for articles that are essentially metal, 25% for metal-heavy derivatives, and a temporarily reduced 15% for the Annex III class of "metal-intensive industrial equipment and electrical grid equipment" through December 31, 2027. [Proclamation 11032 of June 1, 2026](https://www.federalregister.gov/documents/2026/06/04/2026-11314/further-adjusting-the-tariff-regimes-for-imports-of-aluminum-steel-and-copper-into-the-united-states) (91 FR 34085) restated that structure and amended the annexes from June 8, 2026. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) includes each of these instruments with its HTS lines and a checked-on date. The general rule for an imported MV lineup as of August 2026 is the 15% Annex III rate, and for imported bus bar or breaker sub-assemblies it is 50% or 25%. Refined copper itself is not yet tariffed. The Commerce Secretary's June 2025 report recommended a phased universal tariff of 15% on refined copper from January 1, 2027 and 30% from January 1, 2028, which the proclamation left for a later decision; by August 2026 the COMEX premium over LME had become, in CNBC's words, ["a gauge of U.S. tariff expectations"](https://www.cnbc.com/2026/08/14/copper-trump-tariffs-metal-commodities-trade-war.html), with the December 2026 spread near $1,000 a tonne. Our [copper tariff take](https://open-factory.com/articles/copper-tariff-panel/) walks a 200-amp panel through the same arithmetic. For switchgear the arithmetic is short: **an 8.7% copper weight times a 70% price rise is 6.0 index points, or one thirteenth of the 78-point increase.** ## Steel and the Enclosure Steel is the other input every buyer names, and its path is the opposite shape from copper's. The PPI for steel mill products, [WPU1017](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU1017), went from 208.0 in January 2021 to a peak of 449.7 in December 2021, fell to 266.5 by December 2024, and climbed back to 374.2 in July 2026, +79.9% over the full window and +22.5% year over year. Cold-rolled sheet and strip, the enclosure steel, [WPU101707](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU101707), followed the same arc to +78.6%. Fabricated metal products, [WPU107](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU107), which is the closest proxy for purchased enclosures, stampings and machined parts, rose 66.6%. Weighted, steel from mills at 5.3% of output contributes 4.3 points and fabricated metal at 10.9% contributes 7.2 points, so the whole steel chain is 11.5 of the 78 points. **Steel round-tripped: a switchgear OEM that raised prices on the 2021 steel spike and did not cut them when steel fell 41% from December 2021 to December 2024 kept that money for three years before the 2026 tariffs made the price right again.** Steel and aluminum carry the same 50% Section 232 rate as copper under Proclamation 11021, with the same 15% Annex III carve-out for grid equipment through 2027. Grain-oriented electrical steel gets mentioned in every switchgear conversation and belongs in almost none of them. GOES prices [doubled from 2020 to 2024](https://www.breakthroughjournal.org/p/america-makes-the-wrong-steel-for) and Cleveland-Cliffs' Butler Works is the only US mill, with a 920 mm coil-width ceiling against a market that wants 932 mm and up. That is a transformer story, and our [transformer lead-time piece](https://open-factory.com/articles/transformer-160-weeks/) and the [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool include it. In a switchgear lineup, GOES appears in the current and potential transformers, a rounding error against the bus and the breakers. ## Labor: The Biggest Line, the Smallest Mover Labor is 27.3% of switchgear output on the BEA basis, more than copper, steel and enclosures combined, and it moved least. The employment cost index for manufacturing wages and salaries, [ECIMANWAG](https://fred.stlouisfed.org/graph/fredgraph.csv?id=ECIMANWAG), rose 23.4% from Q1 2021 to Q2 2026. Manufacturing average hourly earnings, [CES3000000008](https://fred.stlouisfed.org/graph/fredgraph.csv?id=CES3000000008), went from $23.23 to $30.33, +30.6%, and we use the higher figure in the sensitivity case. On the base case, labor contributes 6.4 points, the same as copper. The cost that OEMs describe on calls is not the wage. It is the fixed cost of adding capacity ahead of volume. Eaton's CEO told analysts in May 2026 that "every time you add fixed cost, labor, depreciation of new CapEx and start-up expenses ahead of volume, it creates this temporary margin headwind," per the [Q1 2026 transcript](https://www.fool.com/earnings/call-transcripts/2026/05/05/eaton-etn-q1-2026-earnings-call-transcript/). Schneider's [$700 million US program](https://www.se.com/us/en/about-us/newsroom/news/press-releases/schneider-electric-plans-to-invest-over-700-million-in-the-u-s-supporting-energy-ai-sectors-and-job-growth-67bdeb3ee4475a5955011b6a/) puts MV in Mt. Juliet, switchgear in El Paso and breakers in Columbia, Missouri, on 2025 to 2027 timelines. Depreciation sits inside gross operating surplus in the BEA accounts, so some of what we book below as margin is new plant. Not most of it: Eaton's Electrical Americas operating margin, which is after depreciation, still rose 930 basis points. ## The Decomposition The method is one line of arithmetic per input. Take the BEA 2017 weight for each input group, multiply by that group's own price change from January 2021 to July 2026, and sum. Whatever the industry PPI rose that the inputs do not explain is booked to gross operating surplus. The assumptions are fixed 2017 weights (no substitution of aluminum for copper, though it happens), a named proxy series per group, taxes moving with output, and a wage index for labor. The [full table with every series](data/03-decomposition.csv) is in the paid section. From 100 in January 2021: copper and nonferrous metal add 6.0 points, steel 4.3, fabricated metal and enclosures 7.2, bought-in relays, controls and breakers 4.2, electronics 0.2, wholesale margins and other services 6.5, labor 6.4, taxes 0.8, for 35.6 points of input push; the residual to reach 178.2 is 42.6 points. **Input costs explain 36 of the 78 points on the base case and 40 on the high-cost case; margin expansion is 38 to 43 points either way, at least half of the price increase.** The high-cost case swaps the wage index for average hourly earnings (+30.6%) and prices bought-in parts and services at the electrical machinery PPI (+36.6%) instead of the services PPI (+27.0%); it moves the residual from 42.6 to 38.4 points, per the [sensitivity file](data/03-decomposition-sensitivity.csv). Read as shares of 2026 output, gross operating surplus goes from 16.7% of the 2017 dollar to roughly a third of the 2026 dollar. Hold that number against a vendor's letter that cites "unprecedented input cost inflation." Two caveats belong here rather than in a footnote. The BEA weights are 2017 and the industry has shifted toward larger, more engineered lineups for datacenters, which carry more bought-in protection and controls. That raises the controls weight and lowers the residual by a point or two. And the PPI is a producer's selling price, which includes the distributor-facing price but not the distributor's margin, so the buyer's invoice inflation may be higher still. Neither caveat gets the margin residual below a third of the increase. ## Eaton's Margin Is the Proof Eaton reports Electrical Americas, its US, Canadian and Latin American electrical business, as one segment, with $13.3 billion of 2025 sales. Its segment operating margin was [20.6% in 2021, 22.5% in 2022, 26.5% in 2023, 30.2% in 2024 and 29.9% in 2025](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000007/etn-20251231.htm), on sales that rose from $7.2 billion to $13.3 billion. The [FY2023 10-K](https://www.sec.gov/Archives/edgar/data/1551182/000155118224000006/etn-20231231.htm) attributes the 2023 step to "higher sales volumes and net price realization," and the same filing says the 2022 gross margin gain came from "inflationary pricing recovery." Those are the company's words for the residual. Quarter by quarter the margin was 21.1% in Q4 2020, fell to 19.2% in Q4 2021 on what Eaton called "higher commodity and logistics costs," crossed 23% by mid-2022, 26% by mid-2023, 29% by early 2024, peaked at 31.6% in Q4 2024, held 29.5 to 30.3% through 2025, dropped to 25.6% in Q1 2026 and recovered to 27.5% in Q2 2026. The 2026 dip is the most instructive part of the series because management narrated it. In May, Eaton described a "negative price cost lag based on commodity inflation beginning of the year" that "will be more than offset in the full year by pricing that we already implemented on April 1," and committed to "exit the year with margins north of 30%" and to a 32% margin by 2030, per the [Q1 2026 call](https://www.fool.com/earnings/call-transcripts/2026/05/05/eaton-etn-q1-2026-earnings-call-transcript/). In July, the CFO said the 190 basis point sequential improvement split "about 100 points from price cost and 90 points from output scale," and that second-half margins would improve 450 to 500 basis points with 300 from price-cost after "pricing actions implemented in Q2 or early August," per [Yahoo Finance's call summary](https://finance.yahoo.com/markets/stocks/articles/eaton-corp-plc-etn-q2-230128644.html). **Eaton's own bridge says price, not volume, carries 300 of the 450 to 500 basis points it expects to recover in the second half of 2026.** The demand side explains why it can. Electrical Americas backlog was [$13.2 billion at December 31, 2025, up 31%](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000007/etn-20251231.htm), with book-to-bill at 1.2; by June 30, 2026 the twelve-month rolling average of orders was up 41% and backlog up 33%, per the [Q2 2026 release](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm). The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) includes every listed electrical OEM on this basis, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) notes what Eaton does not publish: lead times, list prices, or backlog in dollars by product line. ## Powell: 16% to 31% Gross Margin, $415M to $2.4B Powell Industries is the clean case because it makes almost nothing but switchgear, e-houses and the controls that go in them, and it reports backlog in dollars every quarter. Its backlog was [$414.9 million at September 30, 2021](https://www.sec.gov/Archives/edgar/data/80420/000008042021000045/powl-20210930.htm), $592.2 million a year later, $1.3 billion from June 2023 through March 2025, $1.4 billion at June 2025, $1.6 billion at December 2025, $1.8 billion at March 2026 and [$2.4 billion at June 30, 2026](https://www.sec.gov/Archives/edgar/data/0000080420/000008042026000107/powl-20260630.htm), of which $1.3 billion converts to revenue within twelve months. Backlog fell from $464.9 million in December 2020 to $414.9 million in September 2021, tripled to $1.3 billion by June 2023, sat there for seven quarters, and then rose $1.1 billion in the four quarters to June 2026, including a $619 million sequential jump in the last one. The margin followed the backlog with a lag. Gross margin was 15.9% in fiscal 2021 (implied by the [FY2022 10-K](https://www.sec.gov/Archives/edgar/data/80420/000008042022000036/powl-20220930.htm), which reports gross profit rising $10.0 million to $85.0 million on revenue up $62.0 million to $532.6 million), 16.0% in fiscal 2022, 21% in fiscal 2023, [27% in fiscal 2024](https://www.sec.gov/Archives/edgar/data/80420/000008042024000086/powl-20240930.htm), [29% in fiscal 2025](https://www.sec.gov/Archives/edgar/data/80420/000008042025000152/powl-20250930.htm), and 30.6% in the June 2026 quarter. **Powell added 13 points of gross margin between fiscal 2021 and fiscal 2025, more than any other listed switchgear maker, on a book-to-bill that hit 3.0x in the June 2026 quarter.** Management's language on the [August 2026 call](https://www.fool.com/earnings/call-transcripts/2026/08/10/powell-industries-powl-q3-2026-earnings-call-transcript/) was "a stable pricing environment," "some opportunity for price in the market," and "moderate inflation on core commodities, copper, aluminum, steel." Moderate inflation and stable pricing at a 30.6% gross margin is the residual, said politely. The order that moved the backlog was a datacenter: a "$400 million plus" award for phase one of a behind-the-meter "power island," which Powell says increases its addressable content per site because the switchgear sits outside the building, where the architecture stays AC even as the white space moves to 800 VDC. Electric utility is 24% of the backlog and commercial and other industrial, which holds datacenters, is 40%. This is the mechanism our [slot economy piece](https://open-factory.com/articles/slot-economy/) describes from the buyer's side: one hyperscale phase can absorb a mid-sized OEM's annual output, and every other buyer prices against it. ## Lead Times by Voltage Class Price and delivery move together, and the delivery side has the same shape as the margin side: the datacenter buyer sets the marginal slot. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) compiles published lead times, with the publisher and date on every cell, and refuses to average them. Here is the switchgear page as of August 2026. | Equipment class | Published lead time, weeks | Source | Date | |---|---|---|---| | LV panelboards | 21 to 32 | Siemens CompAS schedule via VAWN index | Aug 13 2026 | | LV switchboards (MCCB incomers) | 32 to 41 | Siemens CompAS schedule via VAWN | Aug 13 2026 | | LV switchgear (power circuit breaker) | 54 | Siemens CompAS schedule via VAWN | Aug 13 2026 | | LV power circuit breaker switchboards (ACB incomers) | 84+ | Electronate estimator guide | Mar 20 2026 | | MV switchgear, survey average | 44 | Wood Mackenzie T&D survey Q2 2025 | Oct 15 2025 | | MV 5 kV / 15 kV metal-enclosed | 52 to 72 | Terrapin Construction Group slot reservations | Jun 2026 | | MV 15 kV / 27 kV metal-clad | 60 to 80 | Terrapin Construction Group | Jun 2026 | | MV 38 kV | 78 to 104 | Terrapin Construction Group | Jun 2026 | | MV, datacenter specification, peak-demand OEMs | 104 to 156 | Electronate: approaching 2 to 3 years | Mar 20 2026 | | MV switchgear, pre-pandemic reference | ~24 | Terrapin: schedules that assumed 24 weeks now need 72 | Jun 2026 | [Data: 06-lead-times-by-class.csv](data/06-lead-times-by-class.csv) The spread across rows is the point. Wood Mackenzie's survey average of [44 weeks for MV switchgear in Q2 2025](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/) is a utility-weighted number; a contractor's June 2026 slot reservations show [52 to 72 weeks for 5 and 15 kV metal-enclosed, 60 to 80 for 15 and 27 kV metal-clad, and 78 to 104 for 38 kV](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026), against about 24 weeks before 2020; an estimator's guide puts [datacenter-specification MV at two to three years](https://www.electronate.app/blog/switchgear-lead-times-2026-data-center-boom) at the busiest OEMs. On the LV side, Siemens' published CompAS schedule, [republished by VAWN on August 13, 2026](https://usevawn.com/resources/electrical-equipment-lead-times/), has LV switchgear at 54 weeks, switchboards at 32 to 41, panelboards at 21 to 32 and commodity breakers in stock. **A 38 kV lineup ordered today ships in 78 to 104 weeks. A panelboard ships in 21 to 32, and the panelboard index rose 58% against 112% for switchgear.** Census M3 unfilled orders for NAICS 3353 were $28.4 billion in July 2026 against $5.0 billion of monthly shipments, 5.7 months, versus 4.1 months in January 2021 and 2.6 months in January 2015. The peak was 6.7 months in July 2022, and shipments have grown 26% since then without denting the ratio. Every row in that table is a published number from the party named. None of it is a panel or a survey of ours, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) uses exactly these cells to turn PO dates into an energization month. When the [Quote Check](https://open-factory.com/tools/quote-check/) tool tells you a 15 kV quote at 60 weeks is inside the published range, that is the range it means. ## Why Prices Will Not Fall When Copper Does The consensus in every trade-press piece is that switchgear prices track copper and steel and will ease when they do. The decomposition says the opposite. Copper contributed 6 of 78 points. If copper fell back to its January 2021 level tomorrow, the index would give back 6 points, from 178 to 172, and no OEM with a 5.7-month backlog and a 30% margin target would pass even that through unprompted. Eaton has told investors it will reach a 32% Electrical Americas margin by 2030, up from 29.9% in 2025, and the industry's capacity additions (Schneider's eight sites through 2027, Eaton's [more than $1 billion of North American electrical investment since 2023](https://www.eaton.com/us/en-us/company/news-insights/news-releases/2025/eaton-invests-in-new-south-carolina-transformer-manufacturing.html), Powell's leased Houston expansion) are sized for the datacenter order book, not for price competition. Three things could reverse it, and none is scheduled before 2027. First, the Annex III 15% rate for grid equipment expires December 31, 2027 under Proclamation 11021, after which imported lineups face 25% or 50% unless the annex is extended. That is a floor under domestic prices, not a ceiling. Second, the recommended 15% refined copper tariff starts January 1, 2027 if adopted, which raises the copper line again. Third, Vertiv [stopped publishing its backlog](https://investors.vertiv.com/news/news-details/2026/Vertiv-Reports-Strong-Second-Quarter-2026-with-Diluted-EPS-Growth-of-53-Adjusted-Diluted-EPS-Growth-of-60-Raises-Full-Year-2026-Guidance-Across-All-Key-Metrics/default.aspx) in the second quarter of 2026, which we reconstructed in our [Vertiv estimate](https://open-factory.com/articles/vertiv-backlog-estimate/); the day Eaton or Powell stops publishing theirs is the day the book-to-bill is turning, and it has not come. **Who wins: the four OEMs with US switchgear capacity and the distributors carrying their price letters; who pays: every buyer without a frame agreement, whose lineup is priced at the datacenter's marginal slot whether or not a datacenter is anywhere near the project.** The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes switchgear at its current $/MW and lead time for the developers; the plant owner buying two sections for a 2 MW service upgrade pays the same index. ## What to Do Monday The decomposition converts into four clauses and one phone call. First, index the copper. Put a copper escalator in the PO that references COMEX on the quote date and adjusts both ways at shipment, using the conductor weights the OEM now has to know anyway: Proclamation 11021 requires importers to declare metal content by weight, and every OEM has the kilograms per section in its costing system. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) has the clause and the calculator. Our [switchgear quote anatomy](https://open-factory.com/articles/switchgear-quote-anatomy/) shows where it sits among the eleven line items. Second, ask for the steel and copper weight per section in the quote, and price the escalator only on those kilograms. On the BEA weights, metal is under 15% of the price. An escalator that floats the whole base price on a copper index hands the OEM the other 85% as a free option. Third, put a price-validity clause and a price-decrease trigger next to the escalator, because the April letters are annual and the steel round trip showed what happens without one. Fourth, run the quote through [Quote Check](https://open-factory.com/tools/quote-check/) against the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands and the Monitor's published lead time for the class before you sign. A 15 kV quote at 60 weeks is inside the range, and one at 40 weeks is a promise to verify with the factory, not the rep, as our [RFQ guide](https://open-factory.com/articles/rfq-capital-equipment/) sets out. The phone call is to the panelboard and switchboard side of the same OEM. Where the one-line permits it, LV switchboards at 32 to 41 weeks and panelboards at 21 to 32 weeks, at an index that rose 58% rather than 112%, are the cheapest schedule and price relief available in 2026, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) will show what the substitution does to energization. Liquidated damages at the OEM's margin, the subject of our [LD piece](https://open-factory.com/articles/liquidated-damages/), are the backstop for the MV sections you cannot avoid. Behind the paywall: the decomposition line by line with every series and weight, the sensitivity that moves margin between **38 and 43 points** of the 78, the OEMs ranked by how much price each has taken since 2021, and what each one's own filings say to ask for at the negotiating table. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists the switchgear series monthly. --- # The 30% Fuel-Cell ITC Is the Last Federal Subsidy for On-Site Power URL: https://open-factory.com/articles/fuel-cell-itc-last-subsidy/ Date: 2025-11-29 Power / How-to OBBBA leaves fuel cells, storage, geothermal and nuclear through 2034, wind and solar need construction by July 2026, gas CHP gets nothing, payback for a 10 MW plant On July 4, 2025, Congress cut the federal investment credit for every kind of on-site generation a factory can buy except one. Wind and solar lose it unless construction starts by July 4, 2026. Gas engines and turbines never had it. Natural-gas fuel cells, which lost theirs at the end of 2024, get it back: [Public Law 119-21 section 70513(f)](https://www.govinfo.gov/content/pkg/PLAW-119publ21/html/PLAW-119publ21.htm) writes a new section 48E(j) that pays 30% of the installed cost of "qualified fuel cell property", waives the zero-emissions test, allows no wage or apprenticeship condition, and applies to construction beginning after December 31, 2025. For a plant paying $210/MWh in California or $127/MWh in New York City, that paragraph is the difference between a 2.5-year payback and a 1.7-year one; in Ohio it is the difference between never and 16 years. This piece walks the statute, prices a 10 MW fuel-cell plant against recip CHP and the grid in three states on the actual tariffs, shows where each breaks on gas price, lists every Bloom Energy deal since November 2024, and ends with the states that clear six years and the contract Bloom will sign. Eight technologies, one row at 30% with no conditions, two at 0%, two with a deadline seven months out: that is the federal incentive map for behind-the-meter power in November 2025. ## What Section 48E(j) Actually Says The Inflation Reduction Act made the credit technology-neutral: a "qualified facility" under [48E(b)(3)(A)](https://www.law.cornell.edu/uscode/text/26/48E) needs an anticipated greenhouse gas emissions rate "not greater than zero." A gas engine cannot pass that, and neither could a Bloom Energy Server on pipeline gas, which is why Bloom's [FY2024 10-K](https://www.sec.gov/Archives/edgar/data/1664703/000162828025008747/be-20241231.htm) says "the ITC for fuel cells operating on non-zero carbon fuels expired at the end of fiscal year 2024" and that it safe-harbored equipment in 2024 to keep deploying under the old Section 48 credit through 2028. OBBBA reverses that for fuel cells alone. New 48E(j) says that for qualified fuel cell property "subsection (b)(3)(A) shall be applied without regard to clause (iii) thereof," that "the applicable percentage shall be 30 percent and such percentage shall not be increased or otherwise adjusted by any other provision of this section," and that the emissions recapture in subsection (g) "shall not apply." Section 70513(g)(4) makes it effective for "property the construction of which begins after December 31, 2025." **A factory that starts construction of a gas fuel-cell plant in January 2026 gets 30% of the installed cost back as a federal tax credit. The same plant started in December 2025 gets nothing unless it sits on 2024 safe-harbor equipment.** Bloom's [Q3 2025 10-Q](https://www.sec.gov/Archives/edgar/data/1664703/000162828025046844/be-20250930.htm) reads it the same way. Three fine points. "Not increased or otherwise adjusted" means no 10% domestic-content or energy-community adders, which storage, geothermal and nuclear can still stack. The credit still phases down with everything else under [48E(e)](https://www.law.cornell.edu/uscode/text/26/48E): 100% for construction beginning through 2033, 75% in 2034, 50% in 2035, zero from 2036. And 48E(b)(6) denies the credit to any facility begun after 2025 with "material assistance from a prohibited foreign entity," a China, Russia, Iran and North Korea test that Bloom's Taiwan-heavy supply chain does not trip but that your tax counsel will audit. Wind and solar go the other way. New 48E(e)(4) ends the credit for wind or solar "placed in service by the taxpayer after December 31, 2027," for facilities that begin construction after July 4, 2026, and [IRS Notice 2025-42](https://www.irs.gov/pub/irs-drop/n-25-42.pdf) removed the 5% safe harbor: only "physical work of a significant nature" counts. Storage is carved out by 48E(e)(4)(C), which is why the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) treats a battery as the only credit-eligible way to shave a demand charge in a PJM state. ## The Tax Stack on a 10 MW Plant The credit is not the only tax change. [Section 70301](https://www.govinfo.gov/content/pkg/PLAW-119publ21/html/PLAW-119publ21.htm) restored 100% bonus depreciation for property acquired after January 19, 2025, and 48E qualified property is 5-year MACRS property under [168(e)(3)(B)(viii)](https://www.law.cornell.edu/uscode/text/26/168), so the whole depreciable basis is deducted in year one. [IRC 50(c)(3)](https://www.law.cornell.edu/uscode/text/26/50) reduces that basis by half the credit, so a fuel-cell buyer expenses 85% of cost, not 100%. A recip CHP plant gets the same 100% expensing on 100% of cost and no credit. At $3,500/kW installed, a 10 MW fuel-cell plant costs $35.0M gross. The credit returns $10.5M and year-one expensing of the $29.75M basis is worth $6.25M at 21%, so the owner's net capital is $18.25M. The $25.0M engine plant expenses all of it for $5.25M and nets $19.75M. **After tax, the fuel-cell plant is $1.5M cheaper than the engine plant that costs $10M less on the invoice.** The $3,500/kW is the number to argue about, because Bloom does not publish one. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) brackets it: AEP's [FY2024 10-K](https://www.sec.gov/Archives/edgar/data/4904/000000490425000027/aep-20241231.htm) shows $430M of 2024 construction spend "driven by expenditures for fuel cell generation assets" against a 100 MW first order, $4,300/kW if every dollar is the Bloom order and its installation, and that is our high case; EIA's generic 10 MW fuel cell at [$7,291/kW](https://www.eia.gov/outlooks/aeo/assumptions/pdf/elec_cost_perf.pdf) (2022$) is a number no data-center buyer has paid and we do not use it; the EPA catalog's [2014 fuel-cell prices](https://www.epa.gov/sites/default/files/2015-07/documents/catalog_of_chp_technologies_section_6._technology_characterization_-_fuel_cells.pdf) of $4,600 to $10,000/kW make $3,000 a floor. Every input is in data/assumptions.csv with its source, and [Quote Check](https://open-factory.com/tools/quote-check/) flags a fuel-cell quote outside $3,000 to $4,300/kW installed. ## Fuel, Service and the Heat Rate A Bloom Energy Server 5 is a 300 kW box rated on its [datasheet](https://www.bloomenergy.com/wp-content/uploads/es5-300kw-datasheet.pdf) at 65% to 53% electrical efficiency (LHV, cumulative over life), 5,811 to 7,127 Btu/kWh HHV, NOx of 0.0017 lb/MWh, no water, and "exempt from CA Air District permitting." We use EIA's 6,469 Btu/kWh, inside that band. A gas engine runs 8,295 Btu/kWh in [EIA's AEO2025 table](https://www.eia.gov/outlooks/aeo/assumptions/pdf/elec_cost_perf.pdf), 28% more fuel, and recovers about 35% of fuel input as heat per the [EPA catalog's 9.3 MW case](https://www.epa.gov/sites/default/files/2015-07/documents/catalog_of_chp_technologies_section_2._technology_characterization_-_reciprocating_internal_combustion_engines.pdf). Service is the fuel cell's hidden line. Bloom's FY2024 service revenue was $213.5M on roughly 1.4 GW deployed, about $150/kW-year or $18/MWh at a 95% capacity factor; EIA's $34.65/kW-year is a quarter of that and we do not believe it. Stack replacements sit inside the service agreement, behind $588.1M of capped performance guarantees. The engine gets EIA's $39.57/kW-year plus $8/MWh for oil, plugs and SCR catalyst, about $13/MWh. Over October 2024 to September 2025 [Henry Hub](https://fred.stlouisfed.org/series/MHHNGSP) averaged $3.20/MMBtu while the [EIA citygate](https://www.eia.gov/dnav/ng/hist/n3050ny3m.htm) averaged $5.67 in New York, $5.74 in Ohio, $3.86 in California and $3.68 in Texas. We add $1.00/MMBtu of firm transport ($1.50 in California, where the gas also carries a cap-and-trade compliance cost) for delivered prices of $5.36, $6.67 and $6.74. At those prices the fuel cell burns $34.7/MWh of gas in California, $43.1 in New York and $43.6 in Ohio. The engine burns $44.5, $55.3 and $55.9 before its heat credit. ## Three Tariffs, One Flat Load The grid side is where most on-site power models cheat, so each state's bill is built from the tariff a 10 MW flat load saw in the fourth quarter of 2025. California is $209.5/MWh all in, of which $81.1 is demand charges; New York City is $127.4 with $33.3 of delivery demand and $11.1 of capacity; Ohio is $87.5 with $25.8 of demand and transmission. California: [PG&E Schedule B-20](https://www.pge.com/assets/rates/tariffs/Industrial_B-20_250901-251231.xlsx) at primary voltage, September to December 2025, bills $36.18/kW every month on maximum demand plus $51.93/kW on summer peak and $10.57/kW on summer part-peak: $7.1M a year on a flat 10 MW, $81/MWh, before a kilowatt-hour is bought, and TOU energy adds $128/MWh. The $209.5/MWh total sits within 3% of EIA's [California industrial average](https://www.eia.gov/electricity/data/eia861m/) of 21.61 c/kWh, our check that the load shape is not rigging the answer. New York: [Con Edison SC 9 Rate II](https://www.coned.com/en/rates-tariffs/rates/electric-rates-schedule/electric-psc-10) high-tension delivery bills $11.33 plus $22.82 per kW of maximum demand in June through September, $19.41/kW the other eight months, and $0.0079/kWh. Capacity comes through the Market Supply Charge at [$8.13/kW-month](https://www.coned.com/-/media/files/coned/documents/rates/electric/psc-10/supply/msc-cap/msccap-202511.pdf) in the November 2025 statement. Supply energy is the one number we assume: Con Ed's NYISO index price for NYC averaged 10.3 c/kWh from July to November 2025 with capacity embedded, and we take 6.5 c for a flat load plus $10/MWh of riders. The statewide EIA figure for New York, 9.52 c/kWh, is an upstate number. A Queens plant does not see it. Ohio: [AEP Ohio Schedule GS](https://www.aepohio.com/lib/docs/ratesandtariffs/Ohio/September_2026_AEP_Ohio_Tariff_Book.pdf) primary bills $8.08/kW of demand plus a $10.79/kW Basic Transmission Cost Rider, and the Generation Capacity Rider passes PJM's [$329.17/MW-day](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2026-2027/2026-2027-bra-report.pdf) 2026/27 price through at 1.607 c/kWh. We take EIA's Ohio industrial price for September 2025, 8.75 c/kWh, as the all-in. The [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/) puts the capacity piece alone near $12,000 per MW-month for a plant that peaks with the RTO, and the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool prices the alternative of asking AEP for more service. ## Standby Is the Line Bloom's Brochure Skips An on-site plant does not remove the utility. It changes what you pay it for. **In California and New York a natural-gas fuel cell pays full standby rates, and the New York exemption most buyers remember was rewritten to exclude it.** PG&E's [Schedule S](https://www.pge.com/assets/rates/tariffs/Standby_250901-251231.xlsx) charges a reservation fee of $17.17/kW-month at primary voltage on 85% of reservation capacity, $1.75M a year on 10 MW or $21/MWh of output, and bills outage energy at standby TOU rates that reach $1.3265/kWh on summer weekday afternoons; at a 5% outage rate and our hours-weighted $290/MWh that is another $1.27M. Together they cost the California plant $36/MWh, more than its gas. Con Edison's [General Rule 20.3.2](https://www.coned.com/en/rates-tariffs/rates/electric-rates-schedule/electric-psc-10) exempts "Customers With Designated Technologies" from standby rates, but the 2019 order in Case 15-E-0751 narrowed that to "fuel cells which do not utilize a fossil fuel resource in the process of generating electricity" and closed the efficient-CHP exemption to units of 1 MW or less in service by May 31, 2021. A 10 MW Bloom plant on pipeline gas in Queens lands on SC 9 Rate IV: a contract demand delivery charge of $4.31/kW-month at high tension ($9.88 low tension) plus as-used daily charges of $0.71 to $0.85/kW-day when it draws. We carry $0.77M a year, $9/MWh. Ohio is the gap in the record: AEP Ohio routes self-generators to Schedule SBS, whose sheets are not in the posted book, so we assume half of the $18.87/kW-month demand and transmission charges survive. State money is gone too. California's SGIP still lists fuel cells and pays a [$2.00/W generation incentive](https://www.cpuc.ca.gov/-/media/cpuc-website/divisions/energy-division/documents/self-generation-incentive-program/2025-sgip-handbook-v1.pdf), but section 7.4 of the 2025 handbook requires "100% renewable fuel for the lifetime of the project," and NYSERDA's 2025 fuel-cell money ($3.7M) is hydrogen only. The 30% federal credit is the last subsidy standing, and the only one that does not care what the fuel is. ## Payback by Gas Price At the state's own delivered gas price the fuel cell with the credit recovers its $18.25M in 1.7 years in California, 3.7 in New York City and 16.1 in Ohio; without the credit, 2.5, 5.6 and 24.4 years. Ohio clears six years only at $3.00/MMBtu delivered, a price its citygate has not touched since 2024. The mechanism is simple. In California the plant displaces $18.35M of PG&E bills with $2.89M of gas, $1.5M of service, $1.75M of standby reservation and $1.27M of top-up, saving $10.9M a year. In New York City it saves $4.9M against an $11.2M bill. In Ohio it saves $1.1M against $7.7M, because the fuel cell's fuel, service and standby alone run $78/MWh against an $87.5 grid. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool reproduces these cash flows for any gas curve; data/payback-by-gas-price.csv has every point. The engine is the honest surprise. At the state gas price recip CHP pays back in 1.8, 3.6 and 11.4 years, tying the fuel cell in California and beating it by a month in New York, because $2,500/kW with full expensing is cheap capital and using 60% of the recovered heat is worth $12 to $15/MWh. Take the heat away, at a plant with no year-round thermal load, and the engine's New York payback stretches to 4.6 years and its levelized cost to $102/MWh, above the fuel cell. Then add permitting: a lean-burn engine sits at 3.05 lb NOx/MWh at the NSPS floor before SCR against the fuel cell's 0.0017, and in the Bay Area or Los Angeles basin that is offsets, BACT and a year. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) lists 16 to 18 months for a fuel-cell delivery against 24 to 30 months for a new large engine, per the dealer allocation in [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/), and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) has the switchgear and transformer that both plants still need at 44 and 128 weeks. ## Fifteen-Year Cost per MWh, Three Ways Levelized over 15 years at 8%: California $209.5/MWh on PG&E, $108.9 with fuel cells, $109.6 with engines; New York City $127.4, $95.4 and $89.0; Ohio $87.5, $98.9 and $92.0. The fuel cell's levelized number decomposes cleanly: $24/MWh of net capital recovery, $18 of service, $35 to $44 of gas, and $9 to $36 of standby and top-up depending on the utility. **Without the credit the capital line is $46/MWh and the fuel cell loses to the engine in every state and to the grid in Ohio by $24/MWh.** Ohio is the case to remember because it is where AEP put the first 100 MW: as a time-to-power product sold to data centers at a premium, not as a bill-cutter at $87.5/MWh. And that route is closed to newcomers. AEP's [Q3 2025 10-Q](https://www.sec.gov/Archives/edgar/data/4904/000000490425000171/aep-20250930.htm) reports that Ohio House Bill 15 "repeals the statute that permits electric distribution utilities, including OPCo, to execute contracts to provide customer-sited renewable generation service such as fuel cell technology" after August 14, 2025, grandfathering only two PUCO-approved contracts of about 98 MW now before the Ohio Supreme Court. An Ohio factory buys its own or signs with Bloom. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) tracks the docket. Sensitivity is where to spend your skepticism. At $3,000/kW installed the paybacks are 1.4, 3.2 and 13.8 years; at $4,300/kW they are 2.0, 4.5 and 19.8 (data/sensitivity-fc-capex.csv). Doubling service to $300/kW-year adds $18/MWh. What does not move is the ordering: California always clears, Ohio never does at today's gas, and New York City clears with the credit and is marginal without it. ## Bloom's Order Book Without a Price Six named counterparties, one gigawatt of option, one 15-year factory PPA, and the only dollar figure that lets anyone infer a unit price is AEP's $430M of 2024 construction spend against a 100 MW first order. The deals are real and the disclosure is thin. AEP's [100 MW with an option to 1 GW](https://www.bloomenergy.com/news/bloom-energy-announces-gigawatt-fuel-cell-procurement-agreement-with-aep-to-power-ai-data-centers/) (November 14, 2024), Equinix [passing 100 MW](https://www.bloomenergy.com/news/bloom-energy-expands-data-center-power-agreement-with-equinix-surpassing-100mw/) across 19 sites, [Oracle](https://www.bloomenergy.com/news/oracle-and-bloom-energy-collaborate-to-deliver-power-to-data-centers-at-the-speed-of-ai/) with no megawatts stated and [Brookfield's $5 billion](https://www.bloomenergy.com/news/brookfield-and-bloom-energy-announce-5-billion-strategic-ai-infrastructure-partnership/) all came without a price. The one factory deal, Conagra's [15-year PPA](https://www.bloomenergy.com/news/conagra-brands-collaborates-with-bloom-energy-to-utilize-fuel-cell-technology-at-its-ohio-production-facilities/) for about 6 MW covering 70 to 75% of two Ohio food plants, is, tellingly, a PPA rather than a purchase. The financials say the pricing is firm. Bloom's [Q3 2025 release](https://www.sec.gov/Archives/edgar/data/1664703/000162828025046745/ex991_q32025financialresul.htm) reported revenue of $519.0M, up 57.1%, gross margin of 29.2% against 23.8% a year earlier, and its first positive GAAP operating income at $7.8M; the 10-Q credits "improved pricing" and says the factory goes from 1 GW to 2 GW a year by the end of 2026. A vendor with rising margins, a doubling factory and a new 30% subsidy that lands on its customer's tax return does not cut price in 2026. The buyer's lever is structure, not discount. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades Bloom on the $/kW question in December. ## What to Do Monday Do not start construction in December. The 30% credit attaches to property whose construction begins after December 31, 2025, and under the physical-work rules a binding contract with off-site manufacturing under way is a start. If you sign in December, write the contract so the manufacturer's work begins in January, and put the 48E(b)(6) prohibited-foreign-entity representation in the same document. Pull your last twelve bills and split them the way the bill-anatomy chart does: energy, demand and transmission, capacity, riders. If demand and transmission are under 25% of the bill and your all-in is under $100/MWh, you are Ohio, and the fuel cell does not pay unless someone else is paying for it. If you are above $120/MWh with a flat load, run the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) case with your own gas transport tariff and your utility's standby schedule, not the vendor's. If the reason you are looking at on-site power is a service upgrade you cannot get, price the wait first: the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) includes the study fees and energization years by utility, as laid out in [October's launch piece](https://open-factory.com/articles/large-load-cost-table-launch/), and a [143-week transformer](https://open-factory.com/articles/transformer-160-weeks/) is the reason a 16-month fuel cell is being bought at all. Then get the standby sheet from the utility before the quote from Bloom. It is $21/MWh at PG&E and $9/MWh at Con Ed high tension, and it is the number the brochure leaves out. Behind the paywall: the two states where a factory clears a **six-year** payback on fuel cells today, a 12-state ranking with and without the credit, and the four contract forms Bloom sells, with the terms to demand in each, from the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/). --- # Vertiv Stopped Reporting Its Backlog. Here Is Our Estimate URL: https://open-factory.com/articles/vertiv-backlog-estimate/ Date: 2025-11-11 Electrical / Take $15B and 2.9x book-to-bill in Q4 2025, deferred revenue doubling to $3.6B, guidance and peer read-across, what deposits tell you about 2027 slots Vertiv's order backlog was [$15.0 billion on December 31, 2025](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm), up 109% in a year, on a fourth-quarter book-to-bill of ~2.9x. That was the last time the company printed the number. The [April 22](https://www.sec.gov/Archives/edgar/data/1674101/000162828026026379/q12026exhibit991vrt04222026.htm) and [July 29, 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050323/q22026exhibit991vrt07292026.htm) releases have no backlog, no orders and no book-to-bill. What the filings still include is the deposit line: current deferred revenue went from $1.81 billion to [$3.63 billion in six months](https://data.sec.gov/api/xbrl/companyconcept/CIK0001674101/us-gaap/ContractWithCustomerLiabilityCurrent.json). **Our estimate for the backlog Vertiv is no longer publishing is $16 to $20 billion at June 30, 2026, central case ~$17.4 billion**, and the [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) will list the range until the FY2026 10-K prints the real figure in February 2027. This piece covers what Vertiv said when it pulled the metric, what deferred revenue does and does not measure, the seven quarters where both numbers exist, what Eaton, ABB, GE Vernova, Schneider, nVent and Modine disclosed for the same period, the three cases and the ceiling, and, behind the wall, the quarter-by-quarter table through December 2026 and what it does to 2027 delivery slots. [Data: backlog-vs-deferred-revenue.csv](data/backlog-vs-deferred-revenue.csv) Backlog ran $7.2, $7.9, $8.5, $9.5 and $15.0 billion over the five quarters to December 2025, then stops; deferred revenue ran $1.06 to $1.81 billion over the same five quarters, then added $0.65 billion and $1.17 billion in the two quarters with no backlog to compare it to. ## The Disclosure That Ended in February For nine straight quarters Vertiv's release included a book-to-bill and, in most of them, a dollar backlog: [1.3x and $5.5 billion in Q4 2023](https://www.sec.gov/Archives/edgar/data/1674101/000162828024005765/exhibit991vrt02212024.htm), [1.5x and $6.3 billion in Q1 2024](https://www.sec.gov/Archives/edgar/data/1674101/000162828024017513/q12024exhibit991vrt042424.htm), [~1.2x and $8.5 billion in Q2 2025](https://www.sec.gov/Archives/edgar/data/1674101/000167410125000006/q22025exhibit991vrt7302025.htm), [~1.4x and $9.5 billion in Q3 2025](https://www.sec.gov/Archives/edgar/data/1674101/000167410125000020/q32025exhibit991vrt10222025.htm). The Q4 2025 release of February 11, 2026 printed organic orders up ~252% year over year, ~117% sequentially, and the $15.0 billion. Then, on the same call, CEO Giordano Albertazzi said Vertiv would [stop reporting quarterly orders, order forecasts or backlog with earnings](https://finance.yahoo.com/news/vertiv-q4-earnings-call-highlights-044246559.html), on the grounds that lumpiness in large orders creates "excessive volatility," while continuing to disclose the year-end backlog in the 10-K. Two months later the CEO described first-quarter EMEA bookings as strong [even as the company does not disclose orders](https://finance.yahoo.com/markets/stocks/articles/vertiv-q1-earnings-call-highlights-114915151.html), which is why the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) now scores Vertiv's backlog cell as annual rather than quarterly. The 10-K still defines the thing we are estimating. Backlog is "product and services for which a customer purchase order or purchase commitment has been received and which has not yet been delivered," and "the majority of the combined backlog as of December 31, 2025 is considered firm and is expected to be [shipped within the next 12 to 18 months](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm)." The same document attributes the doubling to "customer orders being placed in advance of our ability to fulfill them." On the February call the CEO added that [backlog consists of binding purchase orders](https://finance.yahoo.com/news/vertiv-q4-earnings-call-highlights-044246559.html), and the CFO said the large fourth-quarter orders "came with larger advanced payments." That sentence is the method: the deposits stay on the balance sheet every quarter, and they are the only series that moves with the orders. Vertiv is not alone. nVent printed a record [$2.6 billion backlog on May 1, 2026](https://www.sec.gov/Archives/edgar/data/1720635/000162828026029098/q12026nvtpressrelease.htm) and its [July 31 release](https://www.sec.gov/Archives/edgar/data/1720635/000162828026051203/q22026nvtpressrelease.htm) has no backlog figure at all; Schneider Electric's [H1 2026 release](https://www.se.com/ww/en/assets/pdf/release-hy-results-2026) says "record backlog" and gives no number. The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) counts eight electrical and cooling OEMs that print a ratio or a backlog for the quarter to June 2026 and ten that do not. The [league table article](https://open-factory.com/articles/book-to-bill-league/) ranks them, and Vertiv is the one that moved from the first group to the second mid-cycle. ## What Deferred Revenue Measures Vertiv's revenue policy says that "payments received in advance for service arrangements or product delivery are [recorded as deferred revenue](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000008/vrt-20251231.htm) and recognized in net sales when the revenue recognition criteria are met." Current deferred revenue is therefore the stock of customer cash sitting against undelivered product plus prepaid service, and the noncurrent piece, [$122.6 million at June 30, 2026](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050609/vrt-20260630.htm), is described in the 10-Q as "primarily maintenance, extended warranty and other service contracts," so the current line is the one that tracks equipment. The 10-Q also gives the drawdown: $604.8 million of the opening balance was recognized as revenue in the second quarter and $1,277.3 million in the first half. Add the drawdown back to the change in the balance and the gross inflow of new customer cash was [$1.32 billion in Q1 2026 and $1.78 billion in Q2](data/deposit-inflows.csv), $3.10 billion in six months, against $805 million in the first nine months of 2025. The useful ratio is deferred revenue divided by backlog, which we can compute for the seven quarters where Vertiv published both. **It sat between 11.6% and 14.8% from December 2023 to December 2025, mean 13.0%, and it was 12.1% on the day the backlog disclosure stopped.** Our [slot-economy piece](https://open-factory.com/articles/slot-economy/) plotted the same deferred revenue against quarterly sales (0.30x to 0.54x for eleven quarters, then 0.63x, 0.93x, 1.11x). This piece uses backlog as the denominator because backlog is what a buyer is queued behind. [Data: deposit-intensity.csv](data/deposit-intensity.csv) The seven disclosed quarters read 11.6%, 11.7%, 14.8%, 13.9%, 14.8%, 11.9% and 12.1%. The low, mid and high cases for June 2026 require 23.3%, 20.9% and 18.4%, so every case assumes customers are now paying at least 4 points more cash per dollar of order than they did in 2024 and 2025. That assumption is why the estimate is a range and not a point. Had the deposit ratio stayed at 11.6% to 14.8%, $3.63 billion of deferred revenue would mean a backlog of $24.6 to $31.3 billion at June 30, 2026, a book-to-bill above 2.5x in both undisclosed quarters. No peer printed that, and the CEO's own reason for pulling the metric was that Q4's lumpiness was not representative. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes GE Vernova's turbine reservations at ~21% of contract value and Modine's chiller capacity deposit at 4.1%, so a Vertiv ratio drifting from 12% toward 20% as project sizes grow is the ordinary direction of travel. $25 to $31 billion is the ceiling the balance sheet permits, not the estimate. ## Peer Read-Across for the Two Blind Quarters The second constraint is what everyone else booked. In the quarter to June 30, 2026, [Eaton's Electrical businesses ran a 1.2 rolling-twelve-month book-to-bill](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm) with total Electrical backlog up 43%; [ABB Electrification booked $7.23 billion against $5.20 billion of revenue](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf), 1.39x, with backlog up 57% to $13.7 billion; [GE Vernova Electrification booked ~1.7x](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm) with equipment backlog up 69% to $40.6 billion; [Trane ran 1.23x](https://www.sec.gov/Archives/edgar/data/1466258/000162828026050743/exhibit991-q22026earningsr.htm) with backlog up 70% to $12.1 billion. Modine, the closest cooling comparable, discloses no backlog but did disclose a [$165.0 million up-front deposit](https://www.sec.gov/Archives/edgar/data/67347/000110465926088230/mod-20260729xex99d1.htm) from one datacenter customer against more than $4 billion of 2027 to 2029 chiller sales. [Data: peer-book-to-bill.csv](data/peer-book-to-bill.csv) Seven peers with a printable ratio sit between 1.20x and 2.10x, Powell alone is at [3.0x on a $934 million order quarter](https://www.sec.gov/Archives/edgar/data/80420/000008042026000103/ex991-powlq3xfy2026earning.htm), and Vertiv's own six disclosed quarters before the spike average 1.37x. **Peer backlog growth of 43% to 70% year over year, applied to Vertiv's $8.5 billion of June 2025, gives $12.2 to $14.5 billion, all of it below the $15.0 billion Vertiv had already reported in December.** Vertiv's Q4 was the outlier in the group, and the peer evidence says the book kept growing in 2026 at a rate closer to the group's 1.2x to 1.75x than to Vertiv's 2.9x. Management's language points the same way: the Q2 2026 release describes "growing pipelines," the [Q2 call](https://finance.yahoo.com/technology/ai/articles/vertiv-q2-earnings-call-highlights-140000648.html) described order momentum strengthening globally, and the CEO told CNBC after the 17% share-price drop that "we have [very strong backlog supporting that](https://www.cnbc.com/2026/07/29/vertiv-ceo-addresses-stock-plunge-disappointing-quarter.html)." None of those is a number. ## The Estimate: $16 to $20 Billion at June 30 The arithmetic is the backlog identity, prior backlog plus orders minus sales, run forward from the disclosed $15.0 billion with reported sales of $2,649.5 million in Q1 2026 and $3,274.3 million in Q2. The unknown is orders, set as a book-to-bill in three cases, each stated in [assumptions.csv](data/assumptions.csv) with its source. Low is 1.1x, the floor set by Eaton's 1.2 rolling ratio and Trane's 1.23x, with room for the EMEA order softness the CFO described. Mid is 1.4x, Vertiv's own mean across its six disclosed quarters and inside ABB Electrification's 1.39x to 1.44x. High is 1.8x, GE Vernova Electrification's H1 2026 range and well below the 2.9x that management called unrepresentative. **Those cases give $15.6 billion, $17.4 billion and $19.7 billion of backlog at June 30, 2026, or 4.8 to 6.0 quarters of current sales, against 3.0 to 3.9 quarters in every disclosed period before the fourth quarter of 2025.** All three imply a deposit ratio of 18% to 23%, above the 2024 to 2025 band and consistent with the "larger advanced payments" the CFO attached to the large orders. A flat backlog at $15.0 billion would require the ratio to have doubled to 24% in six months with no order growth, which no peer's disclosure supports. The two constraints leave this open: the balance sheet says the book grew, the peers say it did not grow at 2.9x. For a buyer the ratio to output matters more than the number. At the mid case Vertiv entered July 2026 holding 5.3 quarters of sales in binding purchase orders, up from 3.5 in September 2025. The 10-K says the majority ships within 12 to 18 months, so the queue in front of a new order is longer than anything in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) for Vertiv's product classes: the Monitor's published lead time for a static UPS of 500 to 1,500 kVA is [30 to 48 weeks and for a CDU 30 to 50 weeks](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) (Terrapin Construction Group, June 2026), and those are the quotes for a slot that exists. A rack-level power and cooling package for a 2027 hall is competing with $9 to $12 billion of already-booked 2027 deliveries, which is the paid table below. ## What to Do Monday If you need Vertiv UPS, busway, PDU or CDU for an energization date in 2027, place the purchase order or the paid capacity reservation this quarter, and get the factory slot date written into it, not the lead time. Vertiv's 10-K says the book is binding POs that ship in 12 to 18 months, so a PO signed in December 2026 is a 2028 delivery unless the slot is named. Ask for the deposit percentage in writing and compare it with the 12% to 23% band above and with the deposit line of the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/): a vendor asking for 30% is pricing scarcity, and that is a term to negotiate, not a formality. Put the equipment PO dates into the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) and see which month the hall actually energizes. Then calendar two filings. The Q3 2026 10-Q, due in late October, prints the September 30 deferred revenue; at the mid case's 21% deposit ratio, ~$3.3 billion reads low, ~$3.9 billion mid and ~$4.8 billion high ([our arithmetic](data/backlog-estimate-by-quarter.csv)). The FY2026 10-K, due in February 2027, prints the December 31 backlog, the one number Vertiv has committed to keep publishing. Our [800 VDC piece](https://open-factory.com/articles/800vdc-strands-frame-agreements/) is the other reason to fix slots rather than sign a five-year frame: downstream SKUs change in 2027. Behind the paywall: the quarter-by-quarter estimate table through December 2026 with orders, backlog, coverage and implied deposit ratio in each case, and the 2027 slot math, which shows that **at the mid case Vertiv ends 2026 with $20.6 billion of binding orders against ~$16.8 billion of 2027 output**, so 2027 is allocated before it starts. The full series is in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). --- # Transformer Lead Times Hit 143 Weeks: Who Gets Delivered First URL: https://open-factory.com/articles/transformer-160-weeks/ Date: 2025-10-25 Electrical / Deep dive Wood Mackenzie 128 and 143 weeks, 82% of large units imported, Hitachi and Siemens Energy capacity by 2028, GOES and Butler Works, $25 million slot fees, the six buyers ranked A generator step-up transformer ordered in the United States in the second quarter of 2025 was quoted at [143 weeks on average](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/), per Wood Mackenzie's T&D supply chain survey published October 15, 2025. A substation power transformer was quoted at 128 weeks and medium-voltage switchgear at 44. In 2021 the average transformer took [about 50 weeks](https://www.woodmac.com/news/opinion/supply-shortages-and-an-inflexible-market-give-rise-to-high-power-transformer-lead-times/). The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists both readings, with the Department of Energy's pre-pandemic "under a year" as the baseline. The chart reads 52 weeks before 2020, 50 in 2021, 120 in 2024, 138 in the first quarter of 2025 and 128 in the second, with the GSU point at 143 and switchgear at 44. The rows and their sources are in [lead-times-by-class-2020-2025.csv](data/lead-times-by-class-2020-2025.csv). Every number below is dated and linked, and the piece covers the survey behind the 143 weeks, the demand and price series under it, the 82% import share and the single steel mill, the six announced US expansions and the one that discloses units, the backlogs that put recovery past 2028, how factory slots are actually bought, and, behind the wall, the six buyer types ranked by who gets delivered first. ## The Numbers Behind 143 Weeks Wood Mackenzie's survey is the only public series that tracks US transformer lead times by class, and it is paywalled, so every figure here is as reported in the firm's own notes or in the federal reports that quote it. In April 2024 the firm wrote that lead times had gone [from around 50 weeks in 2021 to 120 weeks on average in 2024, with large substation and GSU units ranging from 80 to 210 weeks](https://www.woodmac.com/news/opinion/supply-shortages-and-an-inflexible-market-give-rise-to-high-power-transformer-lead-times/). The President's National Infrastructure Advisory Council repeated those numbers in June 2024 and added that [one US large power transformer plant had disclosed a five-year wait](https://www.cisa.gov/sites/default/files/2024-09/NIAC_Addressing%20the%20Critical%20Shortage%20of%20Power%20Transformers%20to%20Ensure%20Reliability%20of%20the%20U.S.%20Grid_Report_06112024_508c_pdf_0.pdf). The Department of Energy's July 2024 resilience report put it in months: [36-month lead times commonly quoted, 60 months at the maximum, against under a year before COVID-19](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf). The October 2025 note is the first to show any easing at the top of the range. **Power transformers fell 10 weeks quarter over quarter to an average of 128 weeks in Q2 2025, GSUs eased slightly to 143, and switchgear edged down to 44**, while [pad-mount single- and three-phase transformers, reclosers, and wire and cable all recorded steady increases](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/). The easing is at the large end, where renewable developers are the marginal buyer and their pipeline is shrinking. The tightening is at the small end, where every data hall, every plant expansion and every storm restoration competes for the same pad-mount. Prices did not ease. The BLS producer price index for power and distribution transformers stood at [351.2 in August 2025 against 200.3 in December 2019](https://fred.stlouisfed.org/series/WPU117409), a 75% increase, and it has not posted a December-to-December decline in any year of the run. Our [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) treats this series as the ceiling reference for any transformer quote dated 2025. Read back, the index rose 36% in 2021, 19% in 2022, 4.5% in 2023, 1.3% in 2024, and 1.8% in the eight months to August 2025. Wood Mackenzie's survey basis says the same thing by class: [unit prices up 45% for GSUs, 77% for power transformers and 78 to 95% for distribution transformers since 2019](https://www.woodmac.com/news/opinion/transformer-troubles-manufacturing-and-policy-constraints-hit-us-transformer-supply/). Its April 2024 note added that [GOES prices had nearly doubled since January 2020 and copper was up about 50%](https://www.woodmac.com/news/opinion/supply-shortages-and-an-inflexible-market-give-rise-to-high-power-transformer-lead-times/), and that suppliers kept raising prices after raw materials eased because the order book let them. ## Demand Tripled, Supply Sat Still The demand side is not a forecast. It is a count of orders already placed. Wood Mackenzie estimates that since 2019 US demand rose [116% for power transformers, 274% for generator step-up units and 41% for distribution transformers](https://www.woodmac.com/press-releases/power-transformers-and-distribution-transformers-will-face-supply-deficits-of-30-and-10-in-2025/), and that 2025 supply runs [30% short for power transformers and 10% short for distribution units](https://www.woodmac.com/press-releases/power-transformers-and-distribution-transformers-will-face-supply-deficits-of-30-and-10-in-2025/). The firm's August 13 opinion piece gives the distribution deficit as 6% and the distribution demand range as 30 to 80% by specification. We print both figures because the two Wood Mackenzie documents disagree. Behind the growth: US electricity consumption up 7% since 2020 after a 1% decline from 2010 to 2020, a [data center pipeline above 125 GW in Q1 2025 adding 226 MW a month, and solar deployments up more than 400% since 2019](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/). The bars read 274% demand and 45% price for GSUs, 116% and 77% for power transformers, 41% and up to 95% for distribution units, which means the smallest units carry the largest price increase for the smallest demand shock. The reason is the input side: distribution cores are nearly all GOES, and [GOES nearly doubled from January 2020](https://www.woodmac.com/news/opinion/supply-shortages-and-an-inflexible-market-give-rise-to-high-power-transformer-lead-times/). The Census Bureau's M3 survey shows the same queue in dollars. **Unfilled orders at US electrical equipment makers (NAICS 3353: transformers, switchgear, motors, generators) stood at $26.7 billion in July 2025 against $4.7 billion of monthly shipments, a 5.7-month backlog**, up from [$14.2 billion and 4.0 months in December 2019 and $8.7 billion and 2.6 months in January 2015](https://www.census.gov/manufacturing/m3/). The ratio peaked at 7.1 months in October 2022 and has sat between 5.5 and 6.2 months for two and a half years. The [171-day capex lead time](https://open-factory.com/articles/171-day-capex-lead-time/) that ISM reports for capital equipment overall is a third of what a transformer buyer faces. The [Lead-Time Monitor launch note](https://open-factory.com/articles/lead-time-monitor-launch/) explains how we combine M3, ISM and the OEM guides into one monthly table. The line averaged 3.0 months from 2015 through 2019, crossed 4 months in December 2019, 5.5 in January 2022, peaked at 7.1 in October 2022, and printed 5.7 in July 2025. Shipments rose from $3.3 billion a month in January 2015 to $4.7 billion in July 2025, so the industry is producing 41% more in dollars and still shipping less than one fifth of its order book each month. Our [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) shows every listed OEM booking more than it ships. The Census aggregate says the same thing for the whole sector. ## Eighty Percent Imported, Steel Included The domestic base was small before the boom. In 2019 the Commerce Department counted [137 large power transformers built in the US for US use, 18% of the total, against 617 imported units, 82%](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf), out of roughly 750 units of demand above 60 MVA. Commerce estimated total domestic capacity at about 343 LPTs a year running at about 40% utilization, which means the plants were not full. They were short of people and steel. DOE's survey of 87 component makers found [89% reporting trouble hiring, with labor averaging 36% of manufacturing cost](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf). Wood Mackenzie's 2025 estimate is that [imports supply 80% of US power transformers and 50% of distribution transformers](https://www.woodmac.com/press-releases/power-transformers-and-distribution-transformers-will-face-supply-deficits-of-30-and-10-in-2025/). The bars read 82% of LPT units imported in 2019, 80% of power transformer supply and 50% of distribution supply imported in 2025, 80% of core steel imported in 2019, and a NIAC target of 50% domestic LPT share by 2029 from [an estimated 20% today](https://www.cisa.gov/sites/default/files/2024-09/NIAC_Addressing%20the%20Critical%20Shortage%20of%20Power%20Transformers%20to%20Ensure%20Reliability%20of%20the%20U.S.%20Grid_Report_06112024_508c_pdf_0.pdf). Hitting that target would require domestic output to rise faster than a demand series that grew 116% in six years. The steel is the harder constraint. Grain-oriented electrical steel and continuously transposed copper conductor are [each about 25% of a large transformer's production cost](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf), and about 80% of GOES was imported in 2019. Cleveland-Cliffs, through its 2020 purchase of AK Steel, is the only active US GOES producer and [can meet 12 to 20% of domestic demand](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf); ATI shut its GOES line in 2016 and imports rose 155% the next year; the highest grade, domain-refined PDR, comes from one company, Nippon Steel. We laid out the width and loss numbers for that mill in [One Steel Mill Gates Every US Transformer](https://open-factory.com/articles/butler-works-goes/) on October 12. **A US transformer plant that wins a hiring battle still buys four of every five tons of core steel from Japan and Korea.** Trade policy raised the price of that steel in the same quarter the survey eased. On August 19, 2025 the Bureau of Industry and Security [added 407 HTS codes to the Section 232 steel and aluminum derivative list, effective August 18, 2025](https://www.federalregister.gov/documents/2025/08/19/2025-15819/adoption-and-procedures-of-the-section-232-steel-and-aluminum-tariff-inclusions-process), and the annex includes 8504.23.00 (liquid dielectric transformers above 10,000 kVA), 8504.33.00, 8504.34.00 and the 8504.90 parts lines that cover imported cores and laminations. The [Section 232 copper proclamation published August 5, 2025](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states) covers the winding side. Wood Mackenzie's October read is that the tariff effect [could be upwards of 20% for some products and will increase competition for domestic production slots](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/). Our [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) prices the stack by HTS line and origin, and the [tariff bill of materials](https://open-factory.com/articles/tariff-bom-electrical/) we published October 10 walks a substation package through it line by line. ## Six Expansions and One Unit Count Every large OEM has announced a US expansion since February 2024. We compiled the six US transformer announcements with a dollar figure and a first-output date into [one table](data/us-capacity-announcements.csv); Wood Mackenzie's count for all of North America since 2023 is [$1.8 billion](https://www.woodmac.com/press-releases/power-transformers-and-distribution-transformers-will-face-supply-deficits-of-30-and-10-in-2025/). | Vendor | Site | Investment | Announced | First output | Disclosed | |---|---|---|---|---|---| | Siemens Energy | Charlotte, NC | $150M | Feb 2024 | early 2026 | 24 to 57 new LPT/yr, 12 to 24 refurbished/yr | | Eaton | Jonesville, SC | $340M | Feb 2025 | 2027 | 700 jobs; no units | | Hitachi Energy | VA, MO, MS components | $250M global, >40% US | Mar 2025 | 2025 to 2027 | 100+ US jobs; no units | | Hyosung HICO | Memphis, TN | $51M | May 2025 | 2025 to 2027 | 123 jobs; no units | | Hitachi Energy | South Boston, VA plus Alamo TN, Bland VA, Mount Pleasant PA | $1B incl. $457M LPT plant | Sep 2025 | 2028, ramping | 825 jobs; no units | | GE Vernova / Prolec GE | Goldsboro, NC plus US and Mexico sites | $300M+ incl. $140M Goldsboro | Oct 2025 | 2025 to 2028 | 330 jobs; no units | *Table data: [us-capacity-announcements.csv](data/us-capacity-announcements.csv)* The table totals about $2.1 billion of US announcements across six programs, with first output spread from early 2026 to 2028, and exactly one row that states units per year. Siemens Energy's Charlotte plant, announced February 14, 2024 at $150 million, will [produce 24 new large power transformers a year initially and 57 at full capacity, plus 12 refurbished units rising to 24](https://www.publicpower.org/periodical/article/siemens-energy-expanding-nc-facility-new-refurbished-large-transformers), with [the first transformers manufactured in early 2026](https://www.siemens-energy.com/us/en/home/press-releases/siemens_energy_addresses_shortage_US_powertransformers_invests_new_factory.html). Against 750 units of 2019 demand grown 116%, roughly 1,600 units in 2025, Charlotte at full rate is ~5% of the market. That is the largest disclosed number in the table. Hitachi Energy announced on September 4, 2025 a [$1 billion US package including a $457 million large power transformer plant in South Boston, Virginia, $106 million for components in Alamo, Tennessee, $22.5 million for dry-type capacity in Bland, Virginia and $70 million for high-voltage components around Mount Pleasant, Pennsylvania](https://www.utilitydive.com/news/hitachi-unveils-1b-grid-manufacturing-investment-including-virginia-trans/759219/), 825 jobs, operational by 2028 with a ramp. That followed a [$250 million component investment announced March 10, 2025, more than 40% of it in Virginia, Missouri and Mississippi](https://www.utilitydive.com/news/hitachi-energy-commits-250-million-transformer-shortage/742010/), inside a [$9 billion global program](https://www.hitachienergy.com/us/en/news-and-events/press-releases/2025/09/hitachi-announces-historic-1-billion-usd-manufacturing-investment-to-power-america-s-energy-future-through-production-of-critical-grid-infrastructure). Eaton's [$340 million Jonesville, South Carolina three-phase plant starts production and hiring in 2027](https://www.utilitydive.com/news/eaton-transformer-production-shortage-investment/739932/). Hyosung HICO is [spending $51 million and adding 123 jobs in Memphis](https://www.actionnews5.com/2025/05/14/manufacturing-company-expand-memphis-operations-add-123-jobs/) and says it wants to be the largest LPT supplier in the country by 2027. GE Vernova agreed on October 21, 2025 to [buy the other 50% of Prolec GE for $5.275 billion](https://www.gevernova.com/news/press-releases/ge-vernova-fully-acquire-prolec-ge-joint-venture), a business with 10,000 employees, seven plants, five of them in the US, about $3 billion of 2025 revenue and $300 million of expansions under way including $140 million and 330 jobs in Goldsboro, North Carolina. None of the last five releases states a units-per-year figure, measured against the standard Siemens Energy set in its own February 2024 release. The documents linked above are the entire disclosure. Our [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) scores that gap directly: does the vendor publish capacity in units, lead times, and backlog in dollars. Dollars without units tell a buyer nothing about when the queue shortens. **$2.1 billion of announced US capacity carries one disclosed number, 81 units a year at Charlotte, against a market of roughly 1,600 large units.** ## Backlogs Say 2028 at the Earliest The order books say the queue is still growing. GE Vernova reported on October 22, 2025 that [Electrification orders rose 102% organically to $5.1 billion in Q3 2025 on revenue of $2.6 billion, and that the segment's equipment backlog reached about $26 billion, up $6.5 billion year to date](https://www.sec.gov/Archives/edgar/data/1996810/000199681025000159/gevpressrelease3q25.htm). Orders at roughly twice revenue is a book-to-bill near 2.0, our arithmetic on the two disclosed figures. The same release shows the turbine side of the queue: [Gas Power equipment backlog and slot reservation agreements grew from 55 to 62 GW, with slot reservations rising from 25 to 29 GW as 12 GW were signed and 7 GW converted to firm orders](https://www.sec.gov/Archives/edgar/data/1996810/000199681025000159/gevpressrelease3q25.htm). Every one of those gigawatts needs a generator step-up transformer, the class quoted at 143 weeks. The [Book-to-Bill League Table](https://open-factory.com/articles/book-to-bill-league/) we published September 10 puts every electrical OEM on the same page, and the [Book-to-Bill League Table model](https://open-factory.com/models/book-to-bill-table/) updates it each quarter. The pattern in October 2025 is uniform: Siemens Energy described US LPT lead times of [up to five years when it announced Charlotte](https://www.siemens-energy.com/us/en/home/press-releases/siemens_energy_addresses_shortage_US_powertransformers_invests_new_factory.html); Hitachi Energy is running the [largest capacity program in the industry at $9 billion](https://www.hitachienergy.com/us/en/news-and-events/press-releases/2025/09/hitachi-announces-historic-1-billion-usd-manufacturing-investment-to-power-america-s-energy-future-through-production-of-critical-grid-infrastructure); GE Vernova is buying its transformer partner outright. Vendors expand when the book covers the plant, and the books cover the plants through 2028. One class does clear earlier. Wood Mackenzie projects the [GSU shortfall narrowing to about 140 units by 2030 from more than 700 in 2025](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/), because the One Big Beautiful Bill Act cut the renewable pipeline that consumed most GSUs, while the new plants come online. That is a shortfall of 700 units in one class in one year, a number to hold beside Charlotte's 57. The gas turbine side is worse and later: Cummins and Caterpillar are sold out through 2028 per our [genset allocation piece](https://open-factory.com/articles/gensets-sold-out-2028/), and GE Vernova's 29 GW of paid slots run to the end of the decade. **No public series shows US transformer lead times returning to their 2019 level before 2028: the two largest announced LPT plants ship in early 2026 and 2028, demand grew 116% in six years, and the sector holds 5.7 months of orders.** Our [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) takes the Monitor's class table and returns the energization month for a set of PO dates; at 128 weeks, a power transformer ordered in the last week of October 2025 lands in the second quarter of 2028. ## How the Queue Is Actually Bought Lead time is an average over buyers who paid to move up and buyers who did not. The instruments are public because the OEMs and the federal reports describe them. | Instrument | Who | Terms disclosed | Source | |---|---|---|---| | Slot reservation agreement, turbines | GE Vernova | 29 GW of paid slots at September 30, 2025; 7 GW converted to orders in Q3 | GE Vernova 8-K, Oct 2025 | | Reservation fee, one turbine | Two Kentucky utilities | $25M to hold a turbine for a 2030 COD | Latitude Media, Sep 2025 | | Framework agreement, transformers | E.ON and Hitachi Energy | up to $700M; Hitachi reserves manufacturing capacity | Hitachi Energy, Jul 2025 | | Multi-year commitments | European utilities | 5 to 10 year purchase commitments; capacity already reserved | NIAC, Jun 2024 | | No long-term contracts | Most US utilities | 5-year capital budgets; only the first years are firm | NIAC, Jun 2024 | | Shared spare inventory | Grid Assurance | 31 utility subscribers, 23 states; no LPT delivered yet | DOE, Jul 2024 | | Refurbishment line | Siemens Energy Charlotte | 12 refurbished LPT/yr at start, 24 at full rate | APPA, Feb 2024 | *Table data: [slot-and-commitment-terms.csv](data/slot-and-commitment-terms.csv)* Seven instruments, three price points: 29 GW of slots, one $25 million fee, one $700 million framework. The turbine market shows the mechanics most clearly because GE Vernova reports them every quarter: a slot reservation agreement is a paid option on a delivery position, and in Q3 2025 the company [signed 12 GW of them and converted 7 GW to orders](https://www.sec.gov/Archives/edgar/data/1996810/000199681025000159/gevpressrelease3q25.htm). Latitude Media reported in September 2025 that all three large turbine OEMs now require a reservation fee and that [two Kentucky utilities paid GE Vernova $25 million to hold a turbine for a 2030 commercial date](https://www.latitudemedia.com/news/gas-turbine-prices-are-up-and-arent-going-down-anytime-soon/). Our companion piece, [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/), follows the same money into datacenter equipment. Transformers have the framework agreement instead. On July 28, 2025 Hitachi Energy signed a deal with E.ON [worth up to $700 million under which it will deliver a considerable part of the transformers by reserving manufacturing capacity](https://www.hitachienergy.com/news-and-events/press-releases/2025/07/hitachi-energy-and-e-on-sign-deal-worth-up-to-700-million-usd-for-critical-grid-infrastructure-to-bolster-energy-security-and-resilience-in-germany) for the German utility. NIAC heard the same from the manufacturers a year earlier: [European utilities have been making 5- to 10-year equipment purchase commitments and have already reserved significant manufacturing capacity, while most US electric utilities do not source transformers under long-term contracts](https://www.cisa.gov/sites/default/files/2024-09/NIAC_Addressing%20the%20Critical%20Shortage%20of%20Power%20Transformers%20to%20Ensure%20Reliability%20of%20the%20U.S.%20Grid_Report_06112024_508c_pdf_0.pdf), because a five-year capital budget is only firm in its first years and commissions frown on inventory. **The buyer with a 10-year commitment holds the slot. The buyer with a 5-year budget holds the lead time.** The third instrument is the spare. Grid Assurance, [owned largely by AEP, FirstEnergy and Berkshire Hathaway, has 31 utility subscribers across 23 states and keeps LPTs, breakers and bushings in storage](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf), though as of DOE's July 2024 report it had not yet delivered a replacement unit. Utilities told DOE they hold high-voltage spare LPTs equal to 116% of the units in the substations ORNL rated most crucial. WattStock's FLEX concept, a single-phase modular unit that replaced a 400 MVA three-phase LPT in Houston in about seven days, found that [two designs could stand in for 60% of North American substation LPTs and nine designs for 80% of GSUs](https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf); the program stalled for lack of funding. The fourth is refurbishment: Charlotte's 12 to 24 refurbished units a year is the only disclosed used-unit capacity in the country. ## What a Plant Buyer Does Monday The buyer at the back of this queue is the one this publication exists for: a plant adding 20 MVA of service, a colocation developer on a 50 MW site, a municipal utility replacing a substation. Their common trait, per NIAC, is that they buy [after design, without a long-term commitment, in one-off ratings](https://www.cisa.gov/sites/default/files/2024-09/NIAC_Addressing%20the%20Critical%20Shortage%20of%20Power%20Transformers%20to%20Ensure%20Reliability%20of%20the%20U.S.%20Grid_Report_06112024_508c_pdf_0.pdf). Each of those traits has a fix with a number attached. Place the transformer order before the site closes and before the single-line is finished, with a release-to-manufacture date written into the PO. At 128 weeks, a PO placed the week of October 27, 2025 delivers in the second quarter of 2028. Every month of design the order waits on moves energization a month. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) puts PO dates against the Monitor's class table and returns the critical path and the temporary-power cost of the gap. Size to a rating that more than one mill builds, and write the impedance and the cooling class into the RFQ as ranges rather than single values, so a second bidder's standard unit qualifies. The [Transformer Sizing and Cost tool](https://open-factory.com/tools/transformer-sizing-cost/) returns MVA, impedance and the published lead time by class from load and voltage; NIAC's fifth recommendation is standardization across the 80,000 designs in service, and WattStock's arithmetic (two designs for 60% of substation LPTs) shows how much of the catalogue is variation for its own sake. A standard rating can be bought from Charlotte, Goldsboro, Memphis or South Boston. A custom one is bought from whichever plant has a slot. Ask for the slot in writing and budget the fee. The turbine market prices a reservation at $25 million for a single 2030 unit. A transformer buyer should expect a deposit against a delivery position and should carry it as a line in the capital budget, refundable against the PO. Ask each bidder for its refurbished-unit availability while the RFQ is open; Charlotte's line is the only one disclosed, and the [Quote Check tool](https://open-factory.com/tools/quote-check/) shows what a compliant transformer quote must state, including the factory slot month. Price the quote against the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands and the 75% PPI move before signing, and run the origin through the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/), because a Korean or Mexican unit now carries the August 2025 Section 232 derivative duty on its steel content. Ask the GSU question even if the plant has no generator. The class that eases first is the generator step-up, because the pipeline that was buying it is shrinking: Wood Mackenzie's projected drop in the [GSU shortfall from more than 700 units in 2025 to about 140 by 2030](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/) means cancelled renewable projects will release delivery positions on 34.5 kV to 138 kV step-up units over the next two years. A plant with a 115 kV service and a substation transformer in that range should ask every bidder whether a cancelled position exists before accepting the 128-week average. Check the utility's own transformer queue before yours. The substation transformer on the utility side of the meter is on the same 128-week list, and the [Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/) records the realistic energization year each utility is quoting for a 50 MW connection in 15 markets. A plant that orders its own unit on time and waits on the utility's has bought nothing. Behind the paywall: the six buyer types ranked by queue position, with the disclosures that put each one there; why the **29 GW** of paid turbine slots already own the GSU line; the four moves, frame agreement, reservation fee, refurbished unit and spec change, with the clause each one needs; and the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) class table the ranking is built on. --- # Quote Check: What Your Switchgear Quote Should Have Said URL: https://open-factory.com/articles/quote-check-launch/ Date: 2025-10-18 Procurement / How-to A $2.25M 15 kV lineup at +52% against n=12, a 2,500 kVA pad-mount 6% under the median, a 500 kW genset above p75, the n >= 5 rule, what the tool cannot see, the negotiation script by band position Riverside Public Utilities paid [$2,254,387.50](https://riversideca.legistar1.com/riversideca/attachments/99a5fd23-cb4f-4c0e-b009-d7390ecf7818.pdf) for one 15 kV metal-clad switchgear lineup in May 2025. The median public award for a 5 to 15 kV metal-clad lineup in the [Open Factory Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is $1,481,321 across 12 awards from five utilities dated December 2016 to May 2025, so the Riverside price is 52% above the median and $63,669 under the 75th percentile. **That one sentence, with its n and its dates, is what [Quote Check](https://open-factory.com/tools/quote-check/) exists to print, and it took a board four months and a 20% contingency to get to a worse version of it.** This piece runs three real public quotes through the tool the way a reader would: a 15 kV lineup from Riverside, a 2,500 kVA pad-mount from Fayetteville, a 500 kW genset from a federal award; what the band says, what the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) says, how to read the gap, what the tool cannot see, when a quote above p75 is still the right buy, and the script for each position in the band. The Riverside lineup sits at 152 on a scale where the band median is 100 and the 75th percentile is 156; the Fayetteville pad-mount sits at 94 against a band that runs 96 to 109; the Army genset sits at 138 against a band that runs 76 to 137. One inside, one just under, one just over. The story is the number next to each: n=12, n=15, n=10. ## What Quote Check Reads The tool takes five inputs: class, size, voltage, region, and the quote if you have one. It finds the class and size band in the Price Book, prints the 25th to 75th percentile with n, the number of buyers, the award-date range and the share of rows from the largest single buyer, then places your quote against the median and prints the [Monitor's](https://open-factory.com/models/lead-time-monitor/) published lead time for the class with its publisher and date. Every input is mirrored to the URL so the scenario can be sent to the vendor as a link. Five steps, two of which can end the run: a band with fewer than five awards prints "insufficient public awards yet" and a link to the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) contribution form instead of a number, and a region with no cell falls back to the US band with a note saying so. The [Price Book's](https://open-factory.com/articles/bid-tab-price-book-launch/) September launch set both rules; nothing below relaxes them. The [Lead-Time Monitor launch](https://open-factory.com/articles/lead-time-monitor-launch/) the same month set the third rule the tool inherits: a lead time prints only with the OEM or survey that published it and the date it was published. ## Quote One: A 15 kV Lineup at +52% Riverside's [Bid SUB-876](https://riversideca.legistar1.com/riversideca/attachments/99a5fd23-cb4f-4c0e-b009-d7390ecf7818.pdf) opened on January 21, 2025 with four bids for one 15 kV metal-clad lineup at the La Colina Substation: Switchgear Power Systems of Winneconne, Wisconsin at $2,254,387.50, Avail Switchgear Systems at $2,403,320.63, Crown Technical Systems at $2,567,340.62 and Myers Power Products at $3,205,870.61, all including California sales tax, against an engineer's estimate of $2,600,000. The [board memo of May 12, 2025](https://riversideca.legistar1.com/riversideca/attachments/c543fa97-4254-4b6d-8413-8103f7cc5f04.docx) approved the low bid with a 20% contingency of $450,507 and put delivery in the fourth quarter of 2027. In Quote Check: class MV metal-clad, 15 kV, region West, quote $2,254,388. The band is 12 lineups: $838,234 at p25, $1,481,321 at the median, $2,318,056 at p75, from [Clark Public Utilities](https://www.clarkpublicutilities.com/meeting_minutes/january-21-2025/) in Washington (five awards, 42% of the band), Riverside (four), Snohomish County PUD, the City of Mesa, and [Naperville](https://naperville.legistar.com/LegislationDetail.aspx?ID=12613&GUID=F54CCBA8-A39B-4B5C-A90E-EE7D33725DCF) in Illinois. The Riverside low bid is the ninth highest of the 12, +52% on the median and 2.7% under p75. The tool prints "inside the band". Myers' high bid at $3.21M would print "above the 75th percentile" at +116% on the median, with only Naperville's $3.83M Tollway Substation lineup above it. The recent window, awards dated 2024 on, holds four lineups and prints nothing: **the tool refuses to publish a 2024-on metal-clad band because n=4, and a reader who wants one will have to wait for the fifth public award.** The lead time line is where the two models disagree, and the tool prints both sides. The [Monitor's](https://open-factory.com/models/lead-time-monitor/) published figure for MV switchgear is [44 weeks on average, Wood Mackenzie's Q2 2025 survey](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/), published October 15, 2025, down from the prior quarter. Riverside's document says fourth quarter 2027 from a May 2025 award, about 131 weeks, and the utility's own [June 2024 board presentation](https://riversideca.legistar1.com/riversideca/attachments/34f6d7f6-a219-4872-95f4-9220c7bb1edf.pdf) called switchgear "long lead-time (Average 70 weeks)". Both are true: the survey is the industry, the calendar is one 15 kV substation lineup with drawings, a control house and a utility's approval loop inside it. The line-by-line walk of what sits inside that calendar is in [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/). The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) takes the 131, not the 44, when it dates your energization. ## Quote Two: A 2,500 kVA Pad-Mount Under p25 Fayetteville Public Works Commission opened [bid PWC2324061](https://www.faypwc.com/wp-content/uploads/2024/04/PWC2324061-Bid-Tab.pdf) on March 19, 2024 for a year of three-phase pad-mounts. On stock code 1-295-786, a 2,500 kVA 12470GrdY/7200 to 480Y/277 unit, quantity four, the awarded bidder Alpha Plotter LLC of Greenville, North Carolina (Eaglerise) priced $65,530.50 a unit, $26.21/kVA, with an 18 to 20 week lead time. The [Commission approved the award on April 10, 2024](https://www.faypwc.com/wp-content/uploads/2024/05/board-minutes-20240410.pdf) at $767,188.10 across all lines, 13 bidders received of 14 solicited. In Quote Check: pad-mount transformer, 2,500 kVA, 12.47 kV, Southeast, quote $65,531. Twelve vendors priced the line, from PulseMac Solution at $64,285 with no lead time stated to Technology International at $219,000 and 10 to 15 months; WEG Transformers USA bid $148,650, 2.3 times the winner, at 48 to 50 weeks. The band the tool reaches for is pad-mount 3-phase 1,500 to 2,500 kVA, 2024 on: 15 awards, five buyers in four states, $26.88/kVA at p25, $27.86 at the median, $30.36 at p75, with Austin Energy's contract price as 8 of the 15 rows. At 2,500 kVA that is $67,200 to $75,900 around a median of $69,650, and the Fayetteville unit lands 6% under the median and $1,670 under p25: the tool prints "below the 25th percentile, check the spec matches and the lead time is real." Switch the window to all years and the same $26.21/kVA is +30% on a median of $20.13 across 72 awards back to 2019, 65% of them Austin's. **The window moves a 2,500 kVA pad-mount from 6% cheap to 30% dear without changing the quote, which is why the tool prints the window and the dates next to every band.** The [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool includes the same bands by kVA for budgeting before a quote exists. The Monitor's published lead time for a pad-mount is the one figure here with no OEM behind it: ["more than two years"](https://www.electric.coop/nreca-joins-coalition-urging-senators-to-fund-distribution-transformers) per the NRECA coalition letter of January 4, 2024, printed as 104+ weeks, with Wood Mackenzie's October 2025 note adding that pad-mount lead times "recorded steady increases" in Q2 2025. The Fayetteville tab holds twelve lead times from 10 to 12 weeks to 10 to 15 months for the same drawing on the same day. The tool does not know which of those twelve vendors you asked, so it prints the survey figure and the source, and leaves the tab for you to open. ## Quote Three: A 500 kW Genset in Its Own Band The Army's Mission and Installation Contracting Command at Fort Drum signed [W911S225PB065](https://www.usaspending.gov/award/CONT_AWD_W911S225PB065_9700_-NONE-_-NONE-) on September 19, 2025: one 500 kW standby generator set for Cleveland, Ohio, awarded to Veterans Electrical Group LLC at $247,403.09 under a total small-business set-aside with nine offers received, period of performance September 20, 2025 to June 17, 2026. That is $494.81/kW. In Quote Check: diesel generator set, 500 kW, 480 V, Midwest, quote $247,403. The stationary 500 to 999 kW band is ten awards from seven buyers in six states, 2019 to 2025: $271.85/kW at p25, $359.55 at the median, $491.45 at p75. The Fort Drum quote is +38% on the median and $3.36/kW over p75, so the tool prints "above the 75th percentile, ask for the bid tab behind the band before accepting." Two things to know before acting on that. First, the quote is in its own band twice: Fort Drum signed a [twin award, W911S225PB078](https://www.usaspending.gov/award/CONT_AWD_W911S225PB078_9700_-NONE-_-NONE-), to the same vendor at the same $247,403.09 the same day under a second solicitation with 13 offers, and both rows count. Without the twins the band is eight awards with a p75 of $408.67/kW and the quote is 21% above it, not 0.7%. Second, federal rows are obligations at award, not invoices. Seven of the ten rows come from USAspending descriptions typed by contracting officers, and the [Price Book](https://open-factory.com/models/bid-tab-price-book/) flags every one. The portable 500 to 999 kW band next to it, also n=10, has a median of $527.72/kW, 47% above the stationary median for the same rating on a trailer. The lead time line for this class is honest and empty: as of October 2025 no OEM guide or survey the Monitor tracks publishes a lead time for a 500 kW genset, so the tool prints "no published lead time in our sources." The award itself does: a period of performance of 39 weeks. The context is in [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/), which in September put Caterpillar's [firm backlog at $37.5 billion at the end of Q2 2025](https://www.sec.gov/Archives/edgar/data/18230/000001823025000040/cat-20250630.htm) with $10.3 billion beyond twelve months; the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool prices what 39 weeks of rented power costs while you wait. ## The Band Table and the n >= 5 Rule *Data: [data/02-band-table.csv](data/02-band-table.csv)* Seven cells, five published. The rule is five awards per cell, and the reason is arithmetic: with four values a 25th percentile is a point between the two lowest, and one utility's stock order can be half the band. The metal-clad 2024-on cell has four lineups from three buyers and says nothing. The stationary genset 2024-on cell has two rows from one buyer, the Fort Drum twins, and says nothing. The all-years cells that do publish carry their own warning in the last column: 42% of the metal-clad band is Clark PUD, 65% of the all-years pad-mount band is Austin Energy, 40% of the genset band is the Army. **A band with a top-buyer share above 80% is one utility's price history, not a market, and the tool prints the share so you can see which you are looking at.** Coverage as of October 18, 2025: 2,504 award lines dated to that day, 44 class and size cells in the all-years window, 30 published and 14 insufficient. Pad-mount and pole-mount transformers hold 19 cells (15 published), chillers and gensets 8 each (5 published each), MV switchgear 3 (all published), power transformers 4 (1 published), CNC machining centers 1 (published), LV switchgear 1 (insufficient). A reader who types a 20 MVA 115 kV substation transformer or a 3,000 A switchboard gets the contribute state today, and the [$2,850 category report](https://open-factory.com/articles/2850-report-says-nothing/) would have given a PPI chart instead. The empty state is the more useful answer, because it names the count it needs. ## What the Tool Does Not Know Quote Check compares a lump sum to a distribution of lump sums, and six things live inside a lump sum that the band cannot see. The worked table shows the three quotes as the tool prints them. *Data: [data/03-worked-gap-table.csv](data/03-worked-gap-table.csv)* Three positions, and a lead-time column that reads 44 against 131, 104+ against 18 to 20, and none against 39. Installation is the first blind spot: the Price Book excludes any award whose document bundles construction (`includes_install` = Y) and keeps rows where freight, startup or permits sit inside the price with a note, so a contractor's furnish-and-install quote will always read high against an equipment-only band. Options are the second: a 15 kV lineup with arc-resistant construction, a control house and spare breakers is one lump sum in the Riverside record and a different lump sum in Clark's, and the band treats them as the same product. Tax and freight are the third: Riverside's $2.25M includes California sales tax, Austin's PO lines are tax-exempt, and federal awards usually include delivery, so a single-digit gap can be tax and freight before any negotiation begins. Region is the fourth. Version 0.1 has no regional cells that clear n=5, so every region falls back to the US band with a note. The Southeast pad-mount buyer is compared with Texas and Florida prices whether or not the freight matches. Sections are the fifth: the tool asks for a section count because that is how a switchgear quote is built, but the metal-clad cell is per lineup, since none of the 12 public awards states how many sections it bought, so the tool compares the total and says so. Time is the sixth: the all-years window mixes 2019 pad-mount prices with 2025 prices across a period when the [transformer PPI rose 81%](https://fred.stlouisfed.org/series/WPU117409), April 2019 to September 2025, which is why the recent window exists and why the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) and the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) sit next to the band rather than inside it. ## When Above p75 Is Still Fair *Data: [data/09-above-p75-and-fair.csv](data/09-above-p75-and-fair.csv)* Ocala Electric Utility's [procurement exception of March 28, 2024](https://ocala.legistar1.com/ocala/attachments/9ffb1400-a660-4ebc-933b-e26ac56c98fb.pdf) is the cleanest public record of a buyer paying above p75 on purpose. Its contract price for a 2,500 kVA pad-mount was $69,647.39, and its distributor had told it the unit carried "a placeholder of 156 weeks, with no guarantee of slot availability." Irby quoted an in-stock GE/Prolec unit at [$163,750 with 4 to 6 weeks](https://ocala.legistar1.com/ocala/attachments/9dbf03b1-3497-4a57-a66f-b074c251ec4e.pdf); Wesco quoted a Hitachi unit at [$159,905 with 156 to 158 weeks](https://ocala.legistar1.com/ocala/attachments/c7ce8f7a-37ba-4f38-a21a-498de85a6eca.pdf). Ocala paid $65.50/kVA, 2.16 times the band's p75, for the same steel and copper as the $27.86 median, and bought 150 weeks with the difference. Fayetteville did the same in [January 2023](https://www.faypwc.com/wp-content/uploads/2023/02/board-minutes-20230125.pdf): it awarded JST Power Equipment at $102,000 a unit and 28 weeks over WEG's lower total at 85 to 94 weeks, and two lower bidders withdrew after opening because they could not honor their prices. **A quote above p75 is fair when the vendor is selling a delivery week the band cannot see, and unfair when it is not. The way to tell is to ask for the week in writing with a liquidated damages clause behind it.** Ocala's exception form gives the reason in one line, "Electric does not have the option to wait," and names the customer it would have lost. [Liquidated Damages at 0.5% a Week](https://open-factory.com/articles/liquidated-damages/) prices the week. A vendor who accepts LDs against an in-stock date is telling you the premium is real. ## What Contributing a Quote Changes Every cell that prints "insufficient" today is waiting for public awards, which arrive at the speed of board meetings. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) is the second source: contributed quotes and invoices, normalized to the same $/kVA, $/section and $/kW conventions, that merge into the Price Book bands once a cell holds five or more contributors and eight or more quotes, no contributor above 40% of the cell, every quote at least three months old. The threshold is stricter than the public rule because the rows are private, and it is why the Benchmark page prints a contribution count of zero rather than a band. A reader who types a 3,000 A switchboard into Quote Check today sees the contribute link. The same reader, three months after uploading the quote that made the cell, sees the band first. What changes is the n behind the opening sentence. The 15 kV metal-clad band is 12 awards over nine years because that is how many public utilities posted a bid tab with a price on it. The datacenter EPCs, developers and contractors who buy most of the lineups in the country post nothing, and the [RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) in this series asks them to write a bid form that could. Until they do, the band is a public buyer's band, and the tool says so on every line. ## What to Do Monday Run the quote before the meeting, not after. Type the spec as the vendor wrote it, read the position before the percent, and if the cell says insufficient, carry the unit price the tool prints into the RFQ as your own reference. Open the bid tab behind the band, which the [Price Book](https://open-factory.com/models/bid-tab-price-book/) links on every row, and find the award closest to your spec and year. The Fayetteville tab's twelve lead times are worth more than the band's median. Put the lead time the Monitor prints and the week the vendor quoted on the same page of the PO, so the gap between 44 weeks and 131 is a number the vendor has signed rather than a surprise the board reads about in 2027. Pull the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) with the quoted week, not the survey week, and read off which month the PO must leave for the plant to energize on time. Behind the paywall: the negotiation script keyed to where the quote sits in the band, below p25, inside the interquartile range, above p75, with the one clause to add in each case and the numbers filled in for the **$2,254,388** Riverside lineup, the $65,531 pad-mount and the $247,403 genset, calibrated to the [Public Bid-Tab Price Book's](https://open-factory.com/models/bid-tab-price-book/) bands. --- # One Steel Mill Gates Every US Transformer URL: https://open-factory.com/articles/butler-works-goes/ Date: 2025-10-12 Electrical / Deep dive Cleveland-Cliffs Butler Works, 920 mm coil width, 39% higher core losses than Japanese GOES, the $400M DoW contract, DPA Title III, what the DOE efficiency rule did and undid Every pound of domestic grain-oriented electrical steel in a US transformer core is melted in one electric arc furnace in Butler, Pennsylvania, at a plant with [0.3 million net tons of configured raw steel capacity](https://www.sec.gov/Archives/edgar/data/764065/000076406526000025/clf-20251231.htm) and a maximum finished coil width of [920 mm](https://www.bis.gov/media/documents/redacted-goes-report-updated-10-26-21.pdf). The US consumes about 220,000 metric tons of GOES a year. The transformer PPI is up 91% since December 2019. Those three numbers are most of the story behind pad-mounts quoted at [40 to 65 weeks and GSUs at 100 to 150 weeks](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) in June 2026, and this piece walks through the mill, the loss gap in watts per pound, the import channels, the DOE rule that was written around Butler, the $400 million Pentagon stockpile, and what the Defense Production Act can and cannot do about any of it. The index sat at 200.3 in December 2019, 351.2 in August 2025 and 381.8 in July 2026, so the last eleven months added 30 points, more than the entire decade before 2021. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists this series next to the published lead times it explains: Wood Mackenzie's survey put power transformers at 128 weeks and GSUs at 144 weeks in Q2 2025 ([as reported by POWER Magazine](https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/)), and our [160-week transformer piece](https://open-factory.com/articles/transformer-160-weeks/) covers who gets delivered first. This one covers the steel. ## One Mill, 1,300 Workers, 920 Millimeters Butler Works sits an hour north of Pittsburgh on 1,300 acres, melts in an EAF feeding an argon-oxygen decarburization unit, casts on two double-strand casters, and runs a three-stand tandem cold mill inside [3.5 million square feet of buildings](https://www.clevelandcliffs.com/operations/steelmaking/butler-works). It makes stainless, carbon and electrical steel. The GOES is finished partly at Zanesville Works in Ohio. Commerce's Section 232 investigation counted [about 1,300 employees at Butler and about 100 at Zanesville](https://www.bis.gov/media/documents/redacted-goes-report-updated-10-26-21.pdf) in 2021; Cliffs told DOE in July 2026 the two plants employ [approximately 1,500](https://downloads.regulations.gov/EERE-2026-BT-STD-0133-0027/attachment_1.pdf), all UAW. **The United States has had exactly one GOES producer since Allegheny Ludlum quit in 2016, and that producer's coil is 12 mm too narrow for large power transformer cores.** Commerce's 2021 report says it plainly: AK Steel "does not produce steel wide enough (>932mm) to form the laminations and cores of larger transformers," and "the maximum width of its domain-refined products (TRAN-COR) is 920mm." Two strips can be patched, but "this process leads to increased production costs and loss of efficiencies in the core." The same report records that AK did not offer the intermediate MOH grade sold by Korean, Japanese and Chinese mills, and that manufacturers "are unable to obtain permanent, heat resistant domain-refined grain oriented steel (PDR GOES) from the sole domestic manufacturer." The width matters because Wood Mackenzie counts imports covering [80% of US power-transformer supply](https://www.woodmac.com/press-releases/power-transformers-and-distribution-transformers-will-face-supply-deficits-of-30-and-10-in-2025/), and NIAC puts domestic LPT share at [about 20%](https://www.cisa.gov/sites/default/files/2024-09/NIAC_Addressing%20the%20Critical%20Shortage%20of%20Power%20Transformers%20to%20Ensure%20Reliability%20of%20the%20U.S.%20Grid_Report_06112024_508c_pdf_0.pdf): the units the US cannot build are the ones Butler cannot supply steel for. How much GOES flows through the market is the number nobody publishes. Commerce's transformer report estimated US consumption at approximately 220,000 metric tons a year, and Cliffs told DOE that [200,000 tons go into distribution transformers](https://downloads.regulations.gov/EERE-2019-BT-STD-0018-0105/attachment_1.pdf). Butler's own GOES tonnage is redacted in every public filing; WEG's comment to Commerce put AK Steel at "a 70% market share of the current industry." Cliffs reports stainless and electrical steel together: 552,000 net tons shipped in 2025, down 3% from 567,000 in 2024, for [$1,748 million of revenue](https://www.sec.gov/Archives/edgar/data/764065/000076406526000025/clf-20251231.htm), and the 10-K says the electrical steel business "is expected to continue to achieve strong profitability in the coming years." GOES is about 25% of a transformer's cost on Commerce's survey basis, which is why the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) tracks awards per kVA rather than trusting a list price. ## The Loss Gap, in Watts per Pound Every DOE transformer analysis rides on one table of core-steel properties, and it is the fairest scorecard available because the measurements come from the same test frame. The designations are DOE's: M-grades are conventional GOES by thickness, "hib" is high-permeability, "dr" is laser domain-refined, "pdr" is heat-proof (permanently) domain-refined, and "am" is amorphous. M3 loses 0.60 W/lb at 1.7 T, 23hib090 0.52, laser-refined 23dr080 0.47, heat-proof 23pdr085 0.46, and amorphous 0.108 W/lb at its 1.35 T ceiling, per the [DOE technical support document](https://downloads.regulations.gov/EERE-2019-BT-STD-0018-0165/content.pdf). On one sheet, then, Cliffs' laser-scribed TRAN-COR H and the Japanese heat-proof grades are near-equal. The gap opens inside a wound core. Distribution cores are wound, then stress-relief annealed at high temperature; DOE's TSD explains that laser domain refinement does not "survive the high-temperature annealing process used to relieve stresses in wound core distribution transformer designs," so laser-refined steel "is primarily used in stacked core designs," while the heat-proof grooves "do survive high-temperature annealing." Cliffs does not sell a heat-proof grade. Nippon Steel and JFE do, and Breakthrough Journal's August 2026 analysis concluded that in distribution cores Cliffs' steel [loses up to 39% more energy than GOES from Nippon Steel and JFE](https://www.breakthroughjournal.org/p/america-makes-the-wrong-steel-for), using a groove-etching process JFE developed in the 1970s that only Nippon Steel, JFE and POSCO run. The datasheets say the same thing in W/kg. JFE guarantees [23JGSD075 at 0.75 W/kg at 1.7 T and 50 Hz, 0.99 W/kg at 60 Hz](https://www.jfe-steel.co.jp/en/products/electrical/catalog/f1e-001.pdf), tested after anneal, in coils from 50 to 1,050 mm wide. Nippon Steel guarantees [23ZDKH70 at 0.70 W/kg](https://www.nipponsteel.com/product/catalog_download/pdf/D004je.pdf) on a single-sheet tester without anneal, and its heat-proof ZDMH80 at 0.80 W/kg after anneal. Metglas guarantees its amorphous [2605SA1 and 2605HB1M at 0.20 W/kg at 1.4 T and 60 Hz](https://metglas.com/wp-content/uploads/2016/12/Amorphous-Alloys-for-Transformer-Cores-.pdf), in 25 micron ribbon no wider than 213 mm, which is why amorphous cores are bought finished and why nobody builds a 300 MVA amorphous unit. Cliffs told DOE in July 2026: "We are not aware of any companies globally that mass produces an electrical steel product that has lower core loss" than TRAN-COR H. Its current product bulletin was unreachable when we checked (the CDN returns 403), so the only published Cliffs width and loss figures we can cite are the ones Commerce cited in 2021, and Cliffs is welcome to send newer ones. DOE's five-year average prices were $1.70/lb for M3, $1.96 for laser-refined 23dr080, $2.12 for heat-proof 23pdr085 and $2.02 for finished amorphous cores, with 23pdr085 modeled to fall to $1.80 if capacity expanded. The same document says the single domestic supplier's low-loss capacity "has historically been directed towards the power transformer market," and that "lower loss grades of GOES have traditionally been sold at a significant price premium relative to M3 grades." Central Moloney told DOE it buys pdr cores for 90% of its designs when it can get them. That premium is the first place to look when the [Quote Check](https://open-factory.com/tools/quote-check/) flags a pad-mount quote above the Bid-Tab band: not copper, steel. ## Amorphous Is 213 Millimeters Wide The steel that DOE's 2023 proposal would have forced into 95% of distribution cores is made by one company in one town. Metglas in Conway, South Carolina is the [only US producer of amorphous cores](https://www.utilitydive.com/news/doe-finalizes-distribution-transformer-rule/712229/); Cliffs' 2023 comment put its output at [20,000 to 25,000 tons a year with 15,000 tons of headroom](https://downloads.regulations.gov/EERE-2019-BT-STD-0018-0105/attachment_1.pdf) against the 200,000 tons of GOES in distribution cores, and amorphous at about 3% of the distribution market. The ribbon is 25 microns thick and comes in [142, 170 or 213 mm widths](https://metglas.com/wp-content/uploads/2016/12/Amorphous-Alloys-for-Transformer-Cores-.pdf), it saturates at 1.56 T against 1.87 to 1.90 T for high-permeability GOES, and DOE's TSD notes that amorphous cores "are generally purchased as finished cores and not directly manufactured by distribution transformer manufacturers." Cliffs' comment claimed an amorphous unit carries an estimated 25% more weight for the same rating, and that the substrate for "domestic" amorphous is imported predominantly from China and Japan. **Amorphous cuts core loss by two thirds at the sheet and cannot be scaled past distribution class, so it removes GOES demand at the small end and leaves the LPT problem untouched.** Metglas itself claims [35% of the core loss of M3](https://metglas.com/magnetic-materials/) for finished cores. That is real money on a pole-top unit that idles at 20% load for 30 years and irrelevant to a 300 MVA GSU that cannot be wound from 213 mm ribbon. NIAC's June 2024 report asked the government to "allow enough time for amorphous steel production capacity to expand in the U.S. or other near-shore areas," which is a polite way of saying the capacity did not exist when the rule was written and, on Cliffs' numbers, still covers about a fifth of the tonnage the 2029 standard would move. Every one of those tons comes off Butler's order book, which is the arithmetic behind the reversal in the next section. ## Japan Sells the Coil, Mexico Sells the Core Because Butler cannot make the width or the heat-proof grade, the US imports the difference, and it does so through two doors. The first is GOES coil under HTS 7225.11 and 7226.11. Coil imports peaked at $149 million and 67,800 tonnes in 2017, fell to $54 million in 2019 after the 25% Section 232 duty ([Proclamation 9705, 83 FR 11625](https://www.federalregister.gov/documents/2018/03/15/2018-05478/adjusting-imports-of-steel-into-the-united-states)), recovered to $128 million in 2023, and dropped to $66 million and 19,900 tonnes in 2025, with Japan at $39.6 million (60%) and Korea at $14.0 million (21%). Russia shipped $9.6 million in 2015 and zero since 2023. The UK and Czech volumes of 2015 to 2018 are gone with the closure of Orb and the shrinking of thyssenkrupp's plants. The 2025 collapse tracks the 50% duty imposed on [June 4, 2025 by Proclamation 10947](https://www.federalregister.gov/documents/2025/06/09/2025-10524/adjusting-imports-of-aluminum-and-steel-into-the-united-states). That coil is not cheap to begin with. Dividing customs value by net weight gives $2,500/t in 2015, $3,448/t in 2022 and $3,327/t in 2025, before duty; at 50% the landed cost of Japanese coil is ~$5,000/t, or ~$2.27/lb against DOE's $1.70/lb for domestic M3. No contract price for Butler GOES is public, and the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) will not publish one until it holds eight or more contributed quotes from five or more contributors. Until then the customs unit value above is the only GOES price series a buyer can check a quote against, and it says the steel itself has risen a third in a decade while the finished transformer has nearly doubled. The second door is bigger. Since 2018 the duty has applied to coil but, until August 2025, not to the laminations and cores cut from it, so core shops moved to Canada and Mexico and imported Japanese steel there. Commerce's survey found [88% import penetration for laminations in 2019](https://www.bis.gov/media/documents/redacted-goes-report-updated-10-26-21.pdf), 75% for wound cores and 54% for stacked cores; core imports rose from $22 million in 2015 to $167 million in 2019, Canada 52% and Mexico 45%, and "neither Mexico nor Canada has indigenous production capability for GOES." Breakthrough's series has transformer-core imports at [$126 million in 2018 and $524 million in 2025](https://www.breakthroughjournal.org/p/america-makes-the-wrong-steel-for). The broader parts heading shows the same shape. HS 8504.90 imports went from $1,287 million in 2018 to $3,649 million in 2025, with Mexico at $986 million, Thailand $542 million, Canada $390 million, China $351 million and India $246 million. The heading also covers inverter and converter parts, so read it as an upper bound on cores. The direction is not in doubt. Commerce concluded in 2021 that laminations and cores "are being imported into the United States in such quantities and under such circumstances as to threaten to impair the national security," and it took four more years for the [August 19, 2025 inclusions notice (90 FR 40326)](https://www.federalregister.gov/documents/2025/08/19/2025-15819/adoption-and-procedures-of-the-section-232-steel-and-aluminum-tariff-inclusions-process) to add 8504.90.9634, 9638 and 9642 (laminations, wound cores, stacked cores) and 8504.23.00 and 8504.33.00 to the derivative list, effective August 18, 2025. Our [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) prices each of those lines by origin, and the [tariff bill of materials](https://open-factory.com/articles/tariff-bom-electrical/) shows what the change adds to a finished lineup. **Cliffs has been asking the government to close this door since 2013 and lost every time until 2025.** AK Steel, Allegheny Ludlum and the Steelworkers petitioned in September 2013 against GOES from seven countries. The ITC found [no material injury from Germany, Japan and Poland](https://www.federalregister.gov/documents/2014/09/12/2014-21724/grain-oriented-electrical-steel-from-germany-japan-and-poland-determinations) in September 2014 and [none from China, Czech Republic, Korea and Russia](https://www.federalregister.gov/documents/2014/11/10/2014-26585/grain-oriented-electrical-steel-from-china-czech-republic-korea-and-russia) that November, one commissioner dissenting each time. Lourenco Goncalves told the Congressional Steel Caucus in March 2020 that Butler and Zanesville would close without action on cores and laminations. Europe, meanwhile, has anti-dumping minimum import prices on GOES from [China, Japan, Korea, Russia and the United States](https://policy.trade.ec.europa.eu/news/commission-initiates-safeguard-investigation-imports-grain-oriented-electrical-steel-2026-03-27_en), and on March 27, 2026 opened a safeguard investigation that also covers laminations and cores: Butler's steel is a dumping target in Brussels and a national-security asset in Washington in the same year. ## What the DOE Rule Did and Undid The one policy that moved GOES demand rather than GOES tariffs was the distribution transformer efficiency standard, and it moved it twice. Thirteen dated actions, two of them about demand: the April 2024 final rule that kept roughly 75% of core steel on GOES, and the April 2026 DPA determination that named electrical core steel a defense resource. DOE's [January 11, 2023 proposal (88 FR 1722)](https://www.federalregister.gov/documents/2023/01/11/2022-28590/energy-conservation-program-energy-conservation-standards-for-distribution-transformers) set levels that, in Secretary Granholm's words, would have required "about 95% of the electrical steel in distribution transformers to be amorphous," with compliance in 2027 ([as reported by Utility Dive](https://www.utilitydive.com/news/doe-finalizes-distribution-transformer-rule/712229/)). Cliffs' [March 2023 comment](https://downloads.regulations.gov/EERE-2019-BT-STD-0018-0105/attachment_1.pdf) said the proposal would "eliminate the market" for Butler's product and put nearly 1,500 jobs at risk, that amorphous was about 3% of the distribution market, and that the sole US amorphous producer ran at 20,000 to 25,000 tons a year with 15,000 tons of headroom against 200,000 tons of GOES to replace. Manufacturers put conversion at $30 to $50 million per plant and $500 to $800 million for the industry. DOE folded. The [April 22, 2024 final rule (89 FR 29834)](https://www.federalregister.gov/documents/2024/04/22/2024-07480/energy-conservation-program-energy-conservation-standards-for-distribution-transformers) adopted standards "expected to require significantly less amorphous material," extended compliance two years to April 23, 2029, and estimated $14 billion of energy savings and 85 million tonnes of CO2 over 30 years, down from $15 billion and 340 million tonnes in the proposal. Cliffs [applauded it on April 4, 2024](https://www.clevelandcliffs.com/news/news-releases/detail/633/cleveland-cliffs-applauds-does-final-transformer): "we expect to actually see an increase in demand for our GOES, opening the possibility of future investments and expansion of our plants." On July 22, 2024 it announced a [$150 million distribution transformer plant at Weirton](https://www.clevelandcliffs.com/news/news-releases/detail/644/cleveland-cliffs-announces-its-new-state-of-the-art), $50 million of it a West Virginia forgivable loan, 600 USW jobs, online in the first half of 2026, citing the efficiency standards as "ensuring the viability of this investment." NIAC's June 2024 report credited the rule with a "carefully synchronized path" and asked for a virtual transformer reserve with the government as buyer of last resort. **Ten months later the transformer plant was dead and two years later Cliffs asked DOE to repeal the rule it had applauded.** Cliffs [scrapped Weirton on May 7, 2025](https://wvmetronews.com/2025/05/07/cleveland-cliffs-pulling-the-plug-on-development-of-transformers-production-plant-at-weirton-facility/), the same week it posted a $483 million first-quarter loss and idled five other facilities. On June 15, 2026 DOE issued a [request for information (docket EERE-2026-BT-STD-0133)](https://www.federalregister.gov/documents/2026/06/15/2026-11971/energy-conservation-program-energy-conservation-standards-for-distribution-transformers) on how the 2029 standards "interact with national security considerations." Of the 33 comments filed, the utilities and the transformer makers defended the rule: EEI asked only for longer compliance, APPA "does not advocate for any changes," NRECA said it "strikes the right balance," NEMA urged DOE "to refrain from actions that would disrupt the certainty," and ERMCO called it "reasonable, balanced, and achievable" ([as reported by Utility Dive](https://www.utilitydive.com/news/doe-mulls-changes-to-biden-era-transformer-rule-raising-utility-concerns/826440/)). Cliffs' [July 15, 2026 comment](https://downloads.regulations.gov/EERE-2026-BT-STD-0133-0027/attachment_1.pdf) asked DOE to "continue to apply the current 2016 efficiency standards indefinitely" or cap the standard at EL2, and called the 2024 rule "a disincentive for further investment in domestic GOES capacity." Every buyer with a 2029 delivery in the [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) now includes a spec risk on the core, and the [large-load policy tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) tracks the docket. ## The $400 Million Stockpile In September 2025 the Defense Logistics Agency awarded Cleveland-Cliffs Steel Corp. a [maximum $400 million indefinite-delivery, indefinite-quantity contract for grain-oriented electrical steel](https://www.war.gov/News/Contracts/Contract/Article/4531882/contracts-for-july-1-2026/), a sole-source, five-year award with no option periods, performance complete September 8, 2030, funded with fiscal 2025 through 2029 transaction funds for the Army, Navy, Air Force, Marine Corps and Space Force. Goncalves disclosed it on the October 20, 2025 earnings call and put the volume at up to 53,000 short tons, adding that "the government tends to stockpile the material for national security purposes" ([as reported by Steel Market Update](https://www.steelmarketupdate.com/2025/10/21/cliffs-aligns-with-us-critical-mineral-policy-moves-forward-with-doe-funded-projects/)). The Department of War published the award on July 1, 2026, nine months after the fact. Three arithmetic facts follow. First, 53,000 short tons over five years is ~10,600 short tons a year, about 4 to 5% of the ~220,000 tonnes the country consumes. It is a stockpile draw on Butler, not a capacity addition. Second, $400 million over 53,000 tons is a ceiling of ~$7,500 a short ton, more than double the ~$3,300 a tonne at which Japanese coil clears customs before duty. The contract is a maximum, not a price, but it is the first federal number attached to Butler steel. Third, the stockpile buys the 920 mm laser-scribed product Butler makes today. It does not buy a wider mill, a heat-proof line, or an amorphous caster. Cliffs' own July 2026 comment describes the only expansion under way: new furnaces at Butler "supported by a DOE award of $75 million," with the total investment and the capacity gain redacted; the project replaces two gas-fired slab reheat furnaces with four induction furnaces and, per Goncalves, proceeds "on schedule" after surviving the October 2025 grant cancellations ([SMU](https://www.steelmarketupdate.com/2025/10/21/cliffs-aligns-with-us-critical-mineral-policy-moves-forward-with-doe-funded-projects/)). The three disclosures the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) asks a supplier for, capacity in units, published lead time and delivery performance, are all absent from Cliffs' filings; GOES capacity appears only as a redacted bracket. The disclosure lag is itself a data point for buyers. A federal customer took roughly 5% of the only domestic GOES supply off the merchant market in September 2025, the seller mentioned it on an earnings call a month later, and the buyer confirmed it the following July. Nothing in between reached a transformer OEM's purchasing desk except a tighter allocation and a longer quote, which is exactly the lag the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) exists to close: it takes the Schneider lead-time guide, Census M3 unfilled orders and 10-Q backlog disclosures and publishes them monthly, so that a September allocation shows up as an October number rather than a July press release. Goncalves told investors on the same October 2025 call that the US steel market "is now hostile territory for dumped steel from abroad" ([Q3 2025 release](https://www.sec.gov/Archives/edgar/data/764065/000076406525000146/clf-202509308xkex991.htm)); for a GOES buyer that hostility is a 50% duty on the only coil wide enough for the job. ## The Determination, and the Three Doors On April 20, 2026 the President signed a [Section 303 determination](https://www.whitehouse.gov/presidential-actions/2026/04/presidential-determination-pursuant-to-section-303-of-the-defense-production-act-of-1950-as-amended-on-grid-infrastructure-equipment-and-supply-chain-capacity/) finding that "transformers, transmission lines and conductors, substations, high-voltage circuit breakers, power control electronics, protective relay systems, capacitor banks, electrical core steel, and related raw materials and manufacturing tools" are essential to national defense, that industry "cannot reasonably be expected to provide these capabilities" without action, waiving the requirements of section 303(a)(1) to (a)(6), and directing the Secretary of Energy to make "purchases, commitments, and financial instruments." It was published in the Federal Register on April 23, 2026, and DOE's June RFI cites it as the reason to reopen the transformer rule. The determination arrived 22 months after NIAC's seven recommendations, which asked for CHIPS-style incentives tied to verifiable capacity, a DOE task force to size the subsidy, long-term purchase commitments of the kind European utilities sign for five to ten years, the virtual reserve, design standardization, coordinated steel, standards and trade policy, and a workforce pipeline. It explicitly named "Defense Production Act funding to support the increased development of large transformer domestic capacity." The DPA Title III fund had about $323 million remaining for fiscal 2026 ([as reported by Utility Dive](https://www.utilitydive.com/news/what-does-trumps-wartime-powers-flex-mean-for-the-transformer-shortage/818159/)), against Breakthrough's estimate of $1 to 2 billion for a Butler retrofit or a greenfield GOES mill. That is the free version of the bottleneck: one EAF, 920 mm, no heat-proof grade, a rule that first threatened and then protected it and is now being reopened at its request, a stockpile that pulls from it, and a determination with less cash than one mill costs. What a buyer does with this on Monday is narrower and more useful. If the transformer is under 5 MVA, ask the OEM in the RFQ which steel designator is in the core and whether the core was annealed. A pdr or amorphous core is a different loss curve and a different import exposure, and the [Transformer Sizing and Cost tool](https://open-factory.com/tools/transformer-sizing-cost/) prices the unit either way. If the unit is above 50 MVA, assume the core steel is Japanese or Korean regardless of the nameplate, and price the tariff on the route it took with the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/). Either way, place the PO for anything with a 2029 delivery before the docket closes, because a re-spec after DOE acts is a lost slot, and the [slot economy piece](https://open-factory.com/articles/slot-economy/) shows what a lost slot resells for. Behind the paywall: what the DPA determination can and cannot fix, and the **three** import routes, with the mills behind them, the duty on each door as of July 2026, JFE's 1,050 mm heat-proof coil against Butler's 920, and the one route whose tariff is capped at 15% through 2027, all tied to the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) classes they feed. --- # MES Is a Consulting Product URL: https://open-factory.com/articles/mes-is-consulting/ Date: 2025-10-10 Machinery / How-to BEP's $110M services line vs $4.6M licence, Riverside's 1% software share, $138 to $300 integrator hours, KPI acceptance by phase, the SOW template and three pricing traps The Bureau of Engraving and Printing has paid Deloitte [$110.1 million since May 2021](https://www.usaspending.gov/award/CONT_AWD_2031ZB21F00015_2041_47QTCK18D0029_4732) for "Manufacturing Execution System (MES) support services" at its two currency plants. The Oracle licences the system runs on cost [$4.64 million this year](https://www.usaspending.gov/award/CONT_AWD_2031ZB25F00050_2041_NNG15SC69B_8000). Five dollars of consulting for every dollar of software, on a firm-fixed-price order that drew one offer, coded to the federal product code for labor, not software. On Monday BEP posted a [notice for an MES Industry Day on December 11](https://sam.gov/opp/36451846906c481fa13173e49089d892/view) to find the next contractor. Every MES buyer gets what BEP got: a small licence, a large services contract, and a support agreement that outlives both. This piece pulls 44 public MES and SCADA awards from USAspending, SAM.gov and city records, shows why the licence is the smallest line, prices the hours, lays out the phases and acceptance tests for the statement of work, and compares the same plant on a low-code or spreadsheet interim; behind the wall, the SOW template and the three integrator pricing structures that go wrong. Oracle renewals at BEP ran [$3.64 million in FY2021, $3.79 million, $4.09 million, $4.44 million and $4.64 million in FY2025](data/01-bep-licence-vs-services.csv), a 6% a year escalator. The Deloitte order took $14.9 million, $21.7 million, $23.9 million, $23.1 million and $25.9 million over the same five fiscal years, $110.1 million against a $130.7 million ceiling. The program has now run ten times longer than the [171-day capex lead time](https://open-factory.com/articles/171-day-capex-lead-time/) ISM reports for the equipment it tracks, and the Industry Day notice says the next contractor will integrate 33 new presses into it. ## 44 Public Awards, Four Kinds of Line MES awards are rare in the public record because most plants that buy one are private. We found six federal buyers with MES by name: BEP, NIH's Center for Cellular Engineering ([Körber PAS-X, $884,492 base year](https://www.usaspending.gov/award/CONT_AWD_75N90024C00029_7529_-NONE-_-NONE-), sole source, "install and configure within 12 months"), the Defense Microelectronics Activity ([Eyelit, $58,200 a year through 2027](https://www.usaspending.gov/award/CONT_AWD_HQ072722C0005_9700_-NONE-_-NONE-)), NASA Marshall ([iBASEt Solumina, $41,509 for the FY2025 support subscription](https://www.usaspending.gov/award/CONT_AWD_80NSSC25PA947_8000_-NONE-_-NONE-), up from $29,253 in FY2022), West Point's Cadet Uniform Factory ([one shop-floor licence, $67,451](https://www.usaspending.gov/award/CONT_AWD_W911SD23P0075_9700_-NONE-_-NONE-)), and the Army Corps at Bonneville Dam ([three Ignition packages with support, $95,930](https://www.usaspending.gov/award/CONT_AWD_W9127N25CA043_9700_-NONE-_-NONE-)). To get a usable sample we widened to the system public buyers own in quantity: SCADA and historian platforms at water and electric utilities, bought from the same integrators on the same licence-plus-hours structure, with the staff report attached. The method is the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) method we [launched in September](https://open-factory.com/articles/bid-tab-price-book-launch/) for transformers and switchgear, applied to software: USAspending API v2 keyword and PSC sweeps (DA01, DA10, 7A20, 7A21), SAM.gov notices for the solicitation documents, and the Legistar API across 150 city and utility boards for matters with SCADA, Ignition, Wonderware or MES in the title. The [44 rows](data/mes-awards.csv) carry buyer, vendor, integrator, licence and services dollars, term and the URL of the record. No market report we have seen prints one of them. The [$2,850 reports](https://open-factory.com/articles/2850-report-says-nothing/) price MES in ranges and cite no award. Four kinds of line recur: a licence or renewal, competed among resellers; a services order to an integrator or consultancy, usually the largest number in the file; a support agreement priced as a percentage of the licences at the site; and hardware, hosting and contingency, often bought through a separate vehicle. **In every award where the buyer itemized, software was 1% to 14% of the program and services were 60% to 95%.** The ratio holds from a $452,000 control-room upgrade to a $144 million federal program. In the [five stacks](data/02-licence-services-share.csv), software is 14.3% at BEP (Oracle renewals against Deloitte plus Oracle managed cloud hosting), 1.0% in Riverside's water SCADA program, 5.2% at the Western Area Power Administration (a $610,646 licence-fee order against $11.0 million of implementation orders), 4.3% in Santa Clara's itemized HSQ quote, and 30.6% only in Symestic's own published on-premise price stack, where the vendor is the one drawing the picture. Riverside is the cleanest case. The city selected its water SCADA software first, through [RFP 2031](https://riversideca.legistar1.com/riversideca/attachments/d769e478-24e4-41fb-bab3-4a0db9efacd3.docx), then required every design-build bidder to use Inductive Automation's Ignition. The program the council approved in December 2022 is $11,989,775: $9,318,000 to Tesco Controls for design-build (78%), $1,303,286 to Eramosa International as owner's representative (11%), $1,062,129 of contingency, $186,360 of the utility's own operations time, and $120,000 for four years of Ignition software maintenance, 1% of the total. The losing design-build bid from Leed Electric was $7,257,987, $2.06 million lower, and lost on the technical score. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) holds both bids. The [14 awards on the chart](data/03-award-by-scope.csv) span $40,300 (one year of Solumina support at NASA Marshall) to $110.15 million (BEP), with the municipal implementations clustered between $1.05 million (Fresno, 300 groundwater wells) and $4.57 million (Columbus's Dublin Road water plant, 730 calendar days, 20% contingency included), and the program that separated software from build, Riverside, at $11.99 million. ## Why the Licence Is the Smallest Line Three kinds of work sit between a licence key and a running plant, and none is in the box. Integration hours: Riverside's water system, Fresno's 300-plus wells, Corona's "over 100 programmable logic controllers" across seven control centers, each tag mapped by a person billing by the hour. Data-model work: the order, routing, part, lot and equipment structures that have to agree between the ERP and the floor before one screen means anything; BEP's Oracle renewal covers "MES, FlowSys, and other in-house developed systems," and the Bureau pays Koenig & Bauer a separate [$3,173,750 for "100% of the development and operations and maintenance support"](https://www.usaspending.gov/award/CONT_AWD_2031ZB25C00002_2041_-NONE-_-NONE-) of FlowSys, the system "that pulls raw machine data from the production presses." Change management: BEP's Industry Day notice asks vendors to run SAFe Agile, "sprint-based delivery with 2-week cycles," and "User Acceptance Testing (UAT) prior to production release," while integrating "33 new presses." None of that is licence. The vendors' own accounts say the same. When Rockwell Automation bought Plex Systems in August 2021, its [10-K allocated the $2,205.5 million purchase price](https://www.sec.gov/Archives/edgar/data/1024478/000102447822000093/rok-20220930.htm) as $1,730.0 million of goodwill, $276.4 million of customer relationships, $232.8 million of technology and $22.2 million of trade names. **The code was 10.6% of what Rockwell paid for an MES company. The customer relationships were worth more than the software.** Emerson paid [$1.6 billion for Open Systems International](https://www.sec.gov/Archives/edgar/data/32604/000003260420000041/emr-20200930.htm) in October 2020, a business with "annual sales of approximately $170 million," 9.4 times revenue for the SCADA vendor that Riverside, WAPA and the Navy could not leave. Rockwell's FY2024 10-K reports [total annual recurring revenue up 16%](https://www.sec.gov/Archives/edgar/data/1024478/000102447824000107/rok-20240930.htm) in a year when its sales fell 9%, the pattern the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) shows for every automation OEM that has moved revenue into software; the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) tracks which of them publish a price at all. The lock-in is documented in the buyers' files. Riverside's cyber-security consultant, AESI, reviewed the [$2,166,116 OSI Monarch renewal](https://riversideca.legistar1.com/riversideca/attachments/66567d4e-b6ba-486e-89bd-b37481e24395.pdf) and wrote that "comparable products exist in the market from several vendors with similar pricing for the software itself, however, a transition from one vendor to another would incur significant additional migration costs." The city waived competition under a purchasing rule for "renewal of maintenance, license(s), support, or a similar need for existing technology systems." Fresno's groundwater SCADA upgrade drew [one proposal, $1,045,914, from Enterprise Automation](https://legistar2.granicus.com/fresno/attachments/406c0db6-7415-46d0-a169-d36c2f75077f.pdf), the firm that "implemented the original SCADA platform nearly 12 years ago." Columbus's Division of Power bought its [PRISM historian add-on sole source for $106,045](https://webapi.legistar.com/v1/columbus/matters/134603/texts/134128) from Minsait ACS, "the only vendor equipped with the proprietary tools, source code access, and technical know-how," on top of $27,369 of annual licensing in 2024 and $42,502 in 2023. Support is where the small line grows. OSI's five-year support for Riverside runs [$147,184, $149,392, $151,143, $348,671 and $155,216](https://riversideca.legistar1.com/riversideca/attachments/66567d4e-b6ba-486e-89bd-b37481e24395.pdf), $951,606 in all, with a system upgrade pre-sold into year four; the annual fee is 12.6% of the $1,169,510 upgrade price. AVEVA's distributor E&M quotes Riverside's wastewater plant [$57,000 a year of "Customer FIRST Premium" support](https://riversideca.legistar1.com/riversideca/attachments/98ba2e81-7ab5-4290-af9a-e3866a88d644.pdf) and states the rule on the quote: "Support is calculated based on the list price of all licenses at the site." Silicon Valley Power let its Wonderware support lapse and paid a [10% reinstatement fee of $3,688](https://santaclara.legistar1.com/santaclara/attachments/9a515fb2-cf9c-4571-b3cf-cc8af48c9fee.pdf) to get back on. Inductive Automation and Sepasoft publish the same structure openly: support at [16%, 20% or 24% of retail](https://inductiveautomation.com/pricing/ignition) a year, and a major-version upgrade without support at 65% of the current retail price. Vendors price the licence low because the annuity is priced on it. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) will include support rates as a percent of licence once the cell-size rules are met. ## What an Hour Costs on the Record Hours are the argument in every services line; public agreements print the rate. HSQ Technology's 2022 agreement with Santa Clara bills [project management at $300, senior engineers at $250 and engineers at $200 an hour](https://santaclara.legistar1.com/santaclara/attachments/3ac01ce9-b3bb-45b1-968e-474d85eaf6a5.pdf) for work outside the $1,390,300 lump sum. Systems Integrated's rate card for Fullerton runs from [$80 for shop labor to $250 for a senior project manager](https://fullerton.legistar1.com/fullerton/attachments/05c3c618-e71a-4b2f-a46a-f96e4d318733.pdf), emergency work at 1.25 times, materials at cost plus 25%. South Coast Automation Systems bills Corona [$184, $164 and $138 an hour](https://corona.legistar1.com/corona/attachments/9b6a0152-2ea1-48d6-b430-f6b21bb48477.pdf) for senior, project and associate engineers, no after-hours premium, 15% on materials. The [five rate cards](data/04-hourly-rates.csv) put integrator engineering at $138 to $300 an hour, against $109 an hour billed by an electrical contractor on a Riverside change order and the $46.59 an hour labor basis Craftsman's 2025 National Electrical Estimator uses for an electrician. The rate is public; the hours are the variable, and only one buyer in our set made the integrator write them down before the award. Fullerton's RFP asked for a master plan priced by task, and Systems Integrated answered with [592.6 estimated hours at an average $150, $88,890](https://fullerton.legistar1.com/fullerton/attachments/05c3c618-e71a-4b2f-a46a-f96e4d318733.pdf): 16 hours per workshop, 2 hours of drawings per site across 61 sites, 24 hours of cybersecurity exposure analysis, 160 hours to write the plan. That table is the whole discipline of scoping in one page: quantity times hours per element, with the elements named. Corona's [$700,000 a year through FY2025](https://corona.legistar1.com/corona/attachments/f000bbf1-3660-4f8a-a2a2-f59a9d794e28.pdf) and Fort Worth's [up to $1,000,000 a year with two renewals](https://fortworthgov.legistar1.com/fortworthgov/attachments/15d63964-06f7-4cf9-b6ad-b53b0e240457.pdf) for SCADA integration on-call are the opposite form: a rate card and a cap, no hours. BEP's Deloitte order is fixed price and still carried [ten "surge" funding modifications](https://www.usaspending.gov/award/CONT_AWD_2031ZB21F00015_2041_47QTCK18D0029_4732) and a June 2025 descope that "captures savings" of $755,763. Use the [Quote Check](https://open-factory.com/tools/quote-check/) hours-per-element sheet before you sign either form. It prices a task from the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) rate cards. ## Scoping the SOW: Phases, KPIs, Fixed or T&M The public agreements agree on the skeleton even when they get the proportions wrong. HSQ's Santa Clara contract pays [30% at the city's approval of design, 30% at completion of manufacturing and factory test, 30% at installation and start-up, 10% at site acceptance](https://santaclara.legistar1.com/santaclara/attachments/3ac01ce9-b3bb-45b1-968e-474d85eaf6a5.pdf), with "no obligation to pay unless Contractor has successfully completed, and City has approved, the milestone." Naperville's Survalent award runs [four phases](https://webapi.legistar.com/v1/naperville/matters/14374/texts/14759): 1A base SCADA with training, 1B the GIS model, 2 the outage system, 3 the DERMS, then five years of support, with hardware carved out to a separate purchase and a 3% contingency on implementation only. Carving out the hardware is right; server, RTU and switchgear deliveries sit on the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/), not on the integrator's schedule. NIH gave Körber [12 months to a configured MES](https://sam.gov/opp/3aef64fa236c45a1bdcf63679f3c4e76/view). BEP tests every release against users before it ships. Riverside chose the software before it chose the builder. The [model schedule](data/05-phase-acceptance.csv) pays 10% at the signed integration inventory, 20% each at approved design, passed FAT, pilot sign-off and rollout, and 10% at final acceptance, so 50% of the price is still unpaid when the first line goes live, against HSQ's 40%. Acceptance by KPI is the part no public agreement in our set wrote down, and it decides whether the services line ends. Define three numbers before the design phase, and put them in the RFP the way Riverside put the software choice ahead of the [$9.3 million build](https://open-factory.com/models/bid-tab-price-book/): the OEE formula per line, first-pass yield per routing step, and WIP age per order, each computed by hand from the paper logs for 30 production days. The acceptance test is that the system's number matches the manual number within 2% for 30 consecutive days on the pilot line, and then on each line at rollout. If the integrator's data model cannot reproduce the plant's own count, the phase is not done, and the payment waits. This is the FAT-as-payment-trigger logic from capital equipment, applied to software. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) treats the MES go-live as one more gated milestone on the plant schedule. Fixed versus time-and-materials follows one rule: fixed price for what has an acceptance test, T&M with a per-task ceiling for what does not. Design, build, FAT, pilot and rollout have tests, so they are fixed, per phase, with the integrator's hours-per-element estimate attached as the basis. Discovery, ERP-side data cleanup, and post-acceptance changes do not have a test yet, so they are T&M at the published rate card with a not-to-exceed per task order and a monthly hours report, the Fullerton form. Corona's single $700,000 PO and BEP's surge lines are what happens when the second form is used for the first kind of work. ## The Interim: Low-Code or a Spreadsheet A plant that cannot fund the $375,000 to $1.2 million implementation range that ScienceSoft [publishes](https://www.scnsoft.com/manufacturing/execution-system) for midsize and large sites has three interim routes, and all three are priced in public. Tulip lists [$100 per interface per month for Essentials and $250 for Professional](https://tulip.co/plans/), billed annually, minimum 10 interfaces, so a 10-station line is $12,000 or $30,000 a year in software before machine monitoring, vision or services add-ons. Sepasoft's MES modules on Ignition are [perpetual: OEE Downtime $22,700, Track and Trace $22,700, SPC $9,100, Batch $35,000](https://www.sepasoft.com/pricing/), plus Ignition's [$13,500 Application Building suite](https://inductiveautomation.com/pricing/ignition), so $58,900 buys OEE, traceability and the HMI outright, plus 16% to 24% a year for support. Symestic prices its cloud MES from [$900 a month for five machines](https://www.symestic.com/en-us/blog/mes-system-prices-2026) and puts traditional on-premise MES at a $35,000 to $100,000 licence, $60,000 to $180,000 of consulting and a $250,000 to $500,000 three-year cost. For [one line, 10 stations, 20 machines](data/06-interim-options.csv), the software line runs from $0 (spreadsheet) through $10,800 (Symestic, five machines), $12,000 and $30,000 (Tulip), to $58,900 (Ignition plus Sepasoft), while the services line runs from $28,000 to $37,000 for 200 hours of a controls engineer on the spreadsheet route to $90,000 to $300,000 for 600 to 1,200 integrator hours on the Ignition route. The spreadsheet is not free; it is 200 hours at $138 to $184 with no machine data and no audit trail. Its value is that it produces the 30 days of manual KPI logs the acceptance test needs, so the interim is phase 1 of the SOW, not a detour. Bonneville Dam took the middle route in June 2025: it bought [three Ignition Pro packages with primary and redundant gateways and PriorityCare support for $95,930](https://sam.gov/opp/a57d656d92bb4d4488ba610a68ea2651/view), specified "the Pro package that Inductive Automation sells has all the modules that we would like," took 21 quotes, and is doing the conversion from FactoryTalk View with its own staff, citing the incumbent's "limitations in licensing flexibility." That is what a licence looks like when the buyer owns the hours: a purchase order, not a program. The [Automation Roadmap](https://open-factory.com/roadmaps/automation-roadmap/) tracks which platforms publish prices, which is the precondition for buying that way. ## What to Do Monday Pick the software before the integrator, the way Riverside ran RFP 2031 before RFP 2056, and get the licence and support quote in writing for 24 months so the annuity is fixed before the hours start. Send the [Quote Check](https://open-factory.com/tools/quote-check/) integration inventory to three integrators and require the Fullerton form back: elements, quantities, hours per element, rate per classification, total. Write the 30-day KPI acceptance test into the RFP with the formulas attached, and put 50% of the price behind pilot and rollout acceptance. Price support as a fixed dollar schedule with an escalator capped at the lesser of 3% or CPI, the clause HSQ already accepted in Santa Clara, not as a percentage of list. If the capex is not there this year, run the spreadsheet interim for 90 days and treat it as phase 1: the logs it produces are the test data. Every award in this piece is a row in the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), and the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) contribution form takes MES and SCADA quotes as of today. Behind the paywall: the eight-section SOW template with model language for each, and the three integrator pricing structures that go wrong on the public record, including the support agreement that pre-sells a **$348,671** year-four upgrade and the milestone schedule that pays 60% before anything runs on site. The Bid-Tab Price Book's [MES and SCADA table](https://open-factory.com/models/bid-tab-price-book/) links every clause to the agreement it came from. --- # Tariff Bill of Materials: What Sections 232 and 301 Add to a UPS, a Lineup, a Genset and a Megapack URL: https://open-factory.com/articles/tariff-bom-electrical/ Date: 2025-10-10 Procurement / Deep dive Copper at 50% on content, steel and aluminum tiers, Chinese cells at 58.4%, SKU by SKU landed cost, the calculator behind it A 1 MW UPS assembled in the United States from the parts a US OEM actually buys carries about 17 points of federal duty in its ex-works price. The same UPS imported finished from China carries 55. The "50% copper tariff" that every quote now cites is 3.8 of those 17 points, and on a 3.9 MWh Megapack it is $5,700 on a $1.39 million unit. This piece builds the bill of materials for four products by weight and by value, puts each line on its HTS code and origin, applies the stack in force on October 10, 2025, and shows which layer actually moves the landed cost: the metal content, the Chinese electronics, or the country the finished good ships from. The US-assembly base case lands at 17.1 points for the UPS, 12.7 for a six-section 15 kV lineup, 6.7 for a 2 MW enclosed diesel genset and 12.6 for the Megapack, rising to 16.5 when the Chinese battery rate steps up on January 1, 2026. Every rate here is in the [Open Factory Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) with its Chapter 99 heading and Federal Register citation. Every BOM share that is our estimate says so in the CSV. ## Three Bases, Not One Rate The reason a tariff on electrical gear is a bill of materials and not a percentage is that the six layers in force are assessed on three different bases. Section 232 copper, under [Proclamation 10962 of July 30, 2025](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states) (90 FR 37727, effective August 1), is 50% "upon the value of the copper content" of semi-finished copper and a short list of derivatives, and the importer reports that content in kilograms on the entry line. Section 232 steel and aluminum, raised to 50% by [Proclamation 10947](https://www.federalregister.gov/documents/2025/06/09/2025-10524/adjusting-imports-of-aluminum-and-steel-into-the-united-states) (90 FR 24199, effective June 4, 2025), is on full value for chapter 72, 73 and 76 goods and on metal content for derivatives outside those chapters. The IEEPA reciprocal tariff under [EO 14257](https://www.federalregister.gov/documents/2025/04/07/2025-06063/regulating-imports-with-a-reciprocal-tariff-to-rectify-trade-practices-that-contribute-to-large-and) (90 FR 15041) applies only to the non-metal remainder of a 232 good, per HTSUS chapter 99 U.S. note 2(v). Section 301 and the China fentanyl duty apply to full value and stack on everything. Six layers, three bases, and one non-stacking order: [EO 14289](https://www.federalregister.gov/documents/2025/05/02/2025-07835/addressing-certain-tariffs-on-imported-articles) (90 FR 18907) stops the auto, Canada, Mexico, steel and aluminum actions from piling on each other, and says nothing about Section 301 or the China fentanyl duty, so those two stack on a Chinese good with everything else. **A US OEM therefore pays 50% on the copper and steel it imports as metal, and nothing under 232 on the same copper and steel when it arrives inside a finished switchgear section from Monterrey**, because 8537.20.00 is on no 232 list as of October 10, 2025. That inversion is the procurement story of this piece, and it is live in every RFQ on a sourcing manager's desk this quarter. The [Tariff Schedule 2026](https://open-factory.com/articles/tariff-schedule-2026/) piece tracks the instruments that can change it. ## The Eighteen HTS Lines The four finished goods sit on four HTS lines: 8504.40.95 for the static converter that is a UPS (column 1 Free), 8537.20.00 for boards and panels over 1,000 V (2.7%), 8502.13.00 for diesel generating sets over 375 kVA (2.0%) and 8507.60.00 for lithium-ion batteries (3.4%), with the BESS statistical suffix 8507.60.00.30 "fully encased within a housing." Rates are from [HTSUS 2025 Revision 25](https://hts.usitc.gov/) and match the 2026 Revision 18 schedule line for line. None of the four is on a Section 232 list in October 2025. All four are on Section 301 lists for Chinese origin: List 3 at 25% for the UPS, List 1 at 25% for the lineup and the genset, List 4A at 7.5% for the battery. Eighteen lines, four of them finished goods with no 232 exposure and fourteen inputs of which nine carry a 232 heading. The inputs are where the duty attaches for a US assembler: refined copper bar 7407.10.50 and winding wire 7408.11.60 under 9903.78.01 at 50% on copper content, flat steel 7208 and 7210 under 9903.81.87 at 50% on full value, aluminum sheet 7606 under 9903.85.02, and, since the [BIS inclusions notice of August 18, 2025](https://www.federalregister.gov/documents/2025/08/19/2025-15819/adoption-and-procedures-of-the-section-232-steel-and-aluminum-tariff-inclusions-process) (90 FR 40326) added 407 codes, the instrument transformers of 8504.31, the transformer cores of 8504.90.96, the cable lines of 8544 and the radiators of 8419.50, each at 50% on their steel or aluminum content. The Chinese lines carry [USTR's four-year review](https://www.federalregister.gov/documents/2024/09/18/2024-21217/notice-of-modification-chinas-acts-policies-and-practices-related-to-technology-transfer) (89 FR 76581): semiconductors 8541 and 8542 at 50% since January 1, 2025, and non-EV lithium-ion batteries moving from 7.5% to 25% on January 1, 2026. | HTS | Description | MFN | Section 232 (Oct 10, 2025) | Section 301 (China) | |---|---|---|---|---| | 8504.40.95 | Static converters (UPS) | Free | none | List 3, 25% | | 8537.20.00 | Panels over 1,000 V (switchgear) | 2.7% | none | List 1, 25% | | 8502.13.00 | Diesel gensets over 375 kVA | 2.0% | none | List 1, 25% | | 8507.60.00 | Lithium-ion batteries (BESS) | 3.4% | none | List 4A 7.5%; 25% from Jan 1, 2026 | | 7407.10.50 | Copper bars and rods (busbar) | 1.0% | 9903.78.01, 50% on copper content | List 3, 25% | | 7408.11.60 | Copper winding wire | 3.0% | 9903.78.01, 50% on copper content | List 3, 25% | | 7208 / 7210 | Flat steel | Free | 9903.81.87, 50% on full value | List 3, 25% | | 7606 | Aluminum sheet | 3.0% | 9903.85.02, 50% on full value | List 3, 25% | | 8541.29 | IGBT modules | Free | none | 9903.91.05, 50% | | 8532.25 | Film capacitors | Free | none | List 1, 25% | | 8504.31.40 | Instrument transformers (CT, VT) | 6.6% | 9903.85.08, 50% on aluminum content | List 3, 25% | | 8535.90.80 | Vacuum interrupters | 2.7% | none | List 1, 25% | | 8544.42.90 | LV cable with connectors | 2.6% | 9903.78.01 copper; 9903.85.08 aluminum | List 3, 25% | | 8419.50.50 | Radiators, heat exchangers | Free | 9903.85.08, 50% on aluminum content | List 1, 25% | | 8504.23.00 | Transformers over 10 MVA | 1.6% | 9903.81.91, 50% on steel content | List 1, 25% | [CSV: data/hts-codes.csv](data/hts-codes.csv) ## Four Bills of Materials by Weight We built each product two ways: by mass, because the 232 metal layers are declared in kilograms, and by value, because every other layer is ad valorem. The Megapack 2 XL is the only one of the four with a public mass: [Tesla's specification](https://en.wikipedia.org/wiki/Tesla_Megapack) lists 38,100 kg (84,000 lb) for 3,916 kWh and 1,927 kW, and a list price of $1.39 million per unit, which is $355/kWh. For the UPS, the lineup and the genset the OEM sheets we could reach do not publish frame mass, so the unit masses (2,000 kg for a 1,250 kW class frame, 1,400 kg per 15 kV section, 22,000 kg for an enclosed 3516-class package) are our engineering estimates, and the CSV says so on every row. By mass the lineup is 68% steel and 12% copper, the genset is 80% iron and steel and 4% copper, the UPS is 45% steel and 18% copper, and the Megapack is 64% LFP cells with 15% steel and 3% copper. Multiply the copper shares by the unit masses and the copper in the four products is 360 kg, 1,008 kg, 880 kg and 1,143 kg. [Powell's 10-K](https://www.sec.gov/Archives/edgar/data/80420/000008042024000086/powl-20240930.htm) names the raw materials a switchgear builder buys as "steel, copper and aluminum, as well as various engineered electrical components," which is the whole 232 exposure: the metal is dutied on the way in, the engineered components are dutied by origin. ## The Copper Tariff Alone Take the copper on its own. At the IMF global copper price of $9,995 per tonne for September 2025 ([FRED PCOPPUSDM](https://fred.stlouisfed.org/series/PCOPPUSDM)), the copper content of the UPS is worth about $3,600, the lineup's $10,100, the genset's $8,800 and the Megapack's $11,400. If every kilogram enters as imported semi-finished copper, the 50% duty is $1,800, $5,000, $4,400 and $5,700 respectively. On the Megapack that is 0.41% of the $1.39 million unit price. **The copper tariff a vendor names in the surcharge letter is worth $1,800 to $5,700 per unit, and a letter that quotes a percentage of contract price instead of kilograms cannot be checked against that number.** The reason a larger surcharge is not simply wrong is the price channel: US copper wire and cable PPI rose 11.9% in the year to September 2025 while the IMF world copper price rose 7.9%, and aluminum mill shapes rose 25.8% against 8.0% for world aluminum ([BLS PPI](https://www.bls.gov/ppi/) WPU10260314, WPU102501; FRED PCOPPUSDM, PALUMUSDM). The gap between the US mill price and the world price is the tariff showing up in the domestic price, and a US OEM pays it whether its copper is imported or not. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) lists both series monthly so a buyer can see which one a surcharge is indexed to. Steel mill products rose 12.4% and the switchgear PPI rose 11.1% over the same twelve months, so the switchgear index moved with its metal, not ahead of it. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes the switchgear and transformer PPI beside the published lead times. The price series is the one input a buyer can check without an RFQ. ## Declaring the Metal in Kilograms The content basis is not a convenience for the importer, it is a filing obligation. Proclamation 10962 directs CBP to issue "authoritative guidance mandating strict compliance with declaration requirements for copper content in imported articles" and names the penalties for underreporting: "significant monetary penalties, loss of import privileges, and criminal liability." The BIS notice of August 19 says the same thing for the 407 new derivatives: 232 duty "will apply to the steel and aluminum content" and "the non-steel and non-aluminum content will remain subject to the reciprocal and other applicable tariffs." Two numbers therefore go on every entry line for a derivative: the kilograms of covered metal and the value of that metal, and the reciprocal layer is computed on what is left. For a US OEM buying semis the declaration is trivial, because a copper bar is copper and the 85% content share we use in the model only strips out fabrication. For an importer of a finished 8504.23 transformer it is a 40-tonne engineering exercise: the GOES core, the tank steel and the clamping structure have to be weighed and valued separately from the copper windings, the oil and the bushings, and the mill certificates for the core steel have to be on file, because the 10% rate under 9903.81.92 for US-melted steel is only available with proof of melt. The 25% steel share we use for a substation transformer in the paid section is our estimate. The buyer's declared share will be audited against the vendor's drawings, not ours. Foreign trade zones and bonded warehouses do not change any of this. Duty is assessed at the rate in force on the date of withdrawal for consumption, so a lineup parked in a zone through a rate change pays the new rate, and 232 metal duty cannot be avoided by zone status on goods that enter US commerce. What a zone does buy is time: a Megapack integrator holding Chinese cells in bond in December 2025 pays 7.5% Section 301 if it withdraws before January 1 and 25% after, a 17.5-point difference on the cell value that the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) prints as two dated rows. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) includes the COMEX and LME prices a declared copper value will be checked against. ## Where the Duty Sits, Line by Line Now the value BOM with origins. For the US-assembly base case we gave each input the origin a US OEM most plausibly uses in 2025: copper, steel and aluminum semis from Mexico under USMCA (which removes MFN and the IEEPA layers but not 232, since USMCA has no carve-out from the metal proclamations), IGBT modules from Japan (Annex II of EO 14257 exempts semiconductors from the reciprocal tariff), film capacitors and half the control boards from China, vacuum interrupters from the EU, engines, alternator winding, relays, inverters and assembly in the United States, and LFP cells from China. The largest single line is the Megapack's cells at 9.0 points of unit value (22% of value at 40.9% duty), then the lineup's steel enclosure at 6.0 points, the UPS's Chinese capacitors at 4.4, the Chinese half of its control boards at 3.5, and the lineup's copper bus at 3.4. IGBT modules, 18% of UPS value, have zero: Free under column 1, exempt from the reciprocal, and not Chinese. **The UPS's 17.1 points split 9.2 on metal content and 7.7 on Chinese electronics, and switching the capacitors and boards to Japanese or European suppliers would cut the electronics half to about 2 points**, because EU and Japanese goods pay a flat 15% inclusive of MFN under [EO 14326](https://www.federalregister.gov/documents/2025/08/06/2025-15010/further-modifying-the-reciprocal-tariff-rates) (90 FR 37963) rather than 55%. The genset is the lightest case at 6.7 points because its two largest value lines, a US-built engine (42% of value) and a US-wound alternator (15%), carry duty only on the imported winding wire, and its steel and radiator lines are small. That result holds for Caterpillar and Cummins packages built in Indiana. A mtu or Rolls-Royce set imported from Germany pays the 15% inclusive rate on the whole unit. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool takes either number as a landed-cost input, and the [Gensets Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/) piece covers why the origin choice is mostly about slots, not duty, this year. ## The Same Four Products, Finished, From China Import the finished good instead and the metal layers vanish while the origin layers take over. A Chinese UPS pays 0 MFN plus 25 (List 3) plus 20 (fentanyl, [EO 14228](https://www.federalregister.gov/documents/2025/03/07/2025-03775), 90 FR 11463) plus 10 (reciprocal, [EO 14334](https://www.federalregister.gov/documents/2025/08/14/2025-15554/further-modifying-reciprocal-tariff-rates-to-reflect-ongoing-discussions-with-the-peoples-republic), 90 FR 39305, through November 10, 2025): 55.0%. The lineup pays 2.7 plus 25 plus 20 plus 10, or 57.7%. The genset 57.0%. The BESS 3.4 plus 7.5 plus 20 plus 10, or 40.9%, becoming 58.4% when List 4A gives way to the 25% battery rate on January 1, 2026. Forty-one to fifty-eight points against 7 to 17 for US assembly. Nobody buys a Chinese 15 kV ANSI lineup for a US datacenter, but plenty of buyers were quoted Chinese BESS in 2024, and the 40.9% stack is why Powin, which relied on Chinese LFP cells, filed for Chapter 11 in June 2025 with more than $300 million of debt ([TechCrunch, June 12, 2025](https://techcrunch.com/2025/06/12/battery-manufacturer-powin-files-for-bankruptcy-months-after-landing-200m-loan/)) citing "uncertainties over levels of US import tariffs" ([Energy-Storage.News, June 11, 2025](https://www.energy-storage.news/powin-files-for-chapter-11-bankruptcy-protection-and-spins-off-project-services-business/)); its former CEO said the company had been "searching for alternative domestic suppliers, but the supply chain wasn't sufficiently mature." The [BESS as UPS](https://open-factory.com/articles/bess-as-ups/) TCO uses the 58.4% figure as the 2026 base case for Chinese cells. ## The Lineup, Nine Ways The switchgear lineup is where the inversion bites hardest, because a US builder's exposure is almost entirely metal and a Mexican builder's is zero. Under USMCA a qualifying lineup from Monterrey or Apodaca pays no MFN (column 1 special), no fentanyl duty ([EO 14232](https://www.federalregister.gov/documents/2025/03/11/2025-03991/amendment-to-duties-to-address-the-flow-of-illicit-drugs-across-our-southern-border), 90 FR 11787), no reciprocal (HTS 9903.01.27), and no 232 because 8537.20.00 is not a derivative. The US-built lineup pays 4.2 points on imported copper bus, 6.0 on imported steel sheet, 0.4 on the aluminum in its CTs, and 1.7 on European interrupters: 12.7 points. Zero for Mexico under USMCA, 12.7 for US-built, 15.0 for the EU and Japan, 17.7 for Korea (2.7 plus the 15% Korea rate), 22.7 for Vietnam, 27.7 for a non-USMCA Mexican unit (25% fentanyl), 52.7 for India since [EO 14329](https://www.federalregister.gov/documents/2025/08/11/2025-15267/addressing-threats-to-the-united-states-by-the-government-of-the-russian-federation) (90 FR 38701) added 25 points on August 27, and 57.7 for China. **Schneider, ABB and Siemens build ANSI metal-clad gear in Mexico, and on October 10, 2025 that gear enters at 0% while Eaton's and Powell's US-built sections carry 12.7 points of metal duty in their cost base.** Whether that shows in the quote depends on how much of the 232 cost a US builder can pass through in a market where the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) has 15 kV metal-clad at 60 to 80 published weeks. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is where the $/section awards will show it first. The second BIS inclusions window that [opened September 15, 2025](https://www.federalregister.gov/documents/2025/09/17/2025-18008/notice-of-the-opening-of-the-inclusions-window-for-the-section-232-steel-and-aluminum-tariff) (90 FR 44799) is the mechanism by which 8537.20 could join the derivative list and close the gap; until it does, a Mexican origin certificate is worth 12.7 points. ## The Megapack Stack, Cells Only The Megapack case is the mirror image: the metal is small, the cells are everything. Tesla builds US Megapacks at Lathrop, California (target 40 GWh a year), so the unit itself never crosses a border; the cells do. With cells at 22% of unit value ($78/kWh of $355/kWh, under the $94/kWh China pack average [BloombergNEF](https://about.bnef.com/blog/lithium-ion-battery-pack-prices-see-largest-drop-since-2017-falling-to-115-per-kilowatt-hour-bloombergnef/) reported for 2024), Chinese LFP cells at 40.9% add 9.0 points to the unit and 12.8 points from January 1, 2026. Add 3.7 points of 232 metal and Mexican cable and the Lathrop unit carries 12.6 points today and 16.5 next year. Nine points today, 12.8 in January, 4.0 with Korean cells (3.4 MFN plus 15% reciprocal on 22% of value), 3.3 with Japanese cells, zero with US cells. The 58.4% headline applies only to a finished Chinese BESS. A US-assembled block with Chinese cells pays it on a fifth of its value. For a buyer comparing a Megapack quote against a Chinese-integrated system, the [MEP Cost Table](https://open-factory.com/models/mep-cost-table/) includes BESS at $/kWh installed and the calculator carries the duty on top. The two numbers move in opposite directions after January 1. ## Thirteen Instruments in Twelve Months Every number above changed at least once between February and October 2025, and the January battery step is already scheduled. The metal proclamations came in February (25%), June (50%) and August (copper); the reciprocal order in April, its country rates in August, India's second 25 points on August 27; the non-stacking order in April; the BIS inclusions on August 18. A PO signed in January 2025 against a 2.7% column 1 rate now sits under six layers, three of which did not exist when it was signed. Thirteen rows, eight of them effective between June 4 and August 27, 2025. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) scores whether a vendor publishes the HTS code and duty layers behind a surcharge or sends a percentage. A percentage cannot be reproduced from the schedule, a heading can. The [Datacenter Electrical Roadmap](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) lists the two dates that matter next: November 10, 2025 for the China reciprocal truce and January 1, 2026 for the battery rate. ## What the Calculator Is and Is Not The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) is a version 0 compilation, not a customs ruling. Its inputs are an HTS line, a country of origin, and, for a derivative, the declared metal content share. Its outputs are the layers above with the Chapter 99 heading, the Federal Register citation and the date we last checked each one. Every rate in it comes from the HTSUS chapter 99 notes and the proclamations and orders cited above, and a change in any of them shows up as a new dated row, never as a silent edit. It holds 18 lines for the four products here and about 200 in total across electrical gear, machine tools and copper. It does not determine origin. Whether a Mexican lineup with Chinese boards qualifies under USMCA is a rules-of-origin question decided by the tariff shift and regional value content tests in the agreement's product-specific rules, and a customs broker or counsel answers it, not a calculator. It does not carry antidumping or countervailing duties, none of which apply to the four finished lines here but several of which apply to inputs such as aluminum extrusions. And it does not know your price. A duty percentage without a price is a rate, and a rate is what the surcharge letter already gave you. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/), launched in September with award prices for transformers, switchgear sections and gensets from public procurement records, is where the percentage becomes dollars per section. ## What To Do Monday Ask every open electrical quote for three things before you sign: the HTS line the vendor will declare, the country of origin of the finished good, and the metal content in kilograms the vendor will report on the entry. With those three you can reproduce the vendor's tariff number in the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) in five minutes, and you will find that a US-built lineup's exposure is about 12.7 points of metal, not the 25 or 50 the surcharge letter implies, and that a USMCA-qualifying Mexican unit's is zero. For BESS, ask for the cell origin and the cell value share. The difference between Chinese and Korean cells is 5 points of unit price today and 9 in January. Run the quote through [Quote Check](https://open-factory.com/tools/quote-check/) against the Bid-Tab bands so the tariff line is judged against a price, not a percentage. Then write the clause: name the HTS code, the origin, the baseline stack by heading on the quote date, and who receives any refund. Most POs signed before 2025 have no such clause, which is why the refund fights of 2026 will be about contracts, not customs. Behind the paywall: the same four products with origin switched (Mexico assembly for the UPS, lineup and genset, Korean and US cells for the Megapack, and a substation transformer sourced from Korea, India and Mexico), the landed-cost delta per product, with the **40.0-point** penalty an Indian lineup carries and the **26.2-point** gap between a Korean and an Indian 10 MVA transformer, plus the eight clauses of a tariff pass-through that can be audited against a CBP entry summary. All of it runs in the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/). --- # The Copper and Wire Price Sheet, Weekly URL: https://open-factory.com/articles/copper-wire-sheet-launch/ Date: 2025-10-02 Procurement / Model launch IMF copper at $6.14 via FRED as the reference, the COMEX gap stated not carried, NEC Table 8 to $/ft for 13 conductor sizes, the escalation-clause calculator, copper wire PPI +17.9%, 14 Border States bulletins logged, what version 1 adds A foot of 500 kcmil copper holds $9.48 of metal at the [$6.14 a pound the IMF recorded for July 2026](https://fred.stlouisfed.org/series/PCOPPUSDM). The distributor who sells it priced it off a COMEX open of $6.28 that month, so the same foot was $9.70 before insulation, margin or freight. The [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) launches today as version 0.1 to put those two numbers, and the conductor weights that turn a pound into a foot, in one place with a source on every cell. It is a compilation of public series and code tables, refreshed every Monday, and it carries an escalation-clause calculator so a bid can name the index it will move with. This piece sets out what the sheet publishes, the exact method from the FRED series to $/ft by NEC conductor size, why the default is an LME-based price and what that means for a US bid, how a distributor bulletin is logged, who the sources are by name, what the sheet cannot tell you yet, and what version 1 adds. The monthly series bottomed at $2.03 a pound in January 2016, averaged $2.73 in 2019 and $4.15 in 2024, and ran from $4.43 in July 2025 to $6.14 in July 2026, a 38.6% rise in twelve months that the sheet prints as twelve monthly cells and nothing in between. ## What the Sheet Publishes Every Monday Version 0.1 publishes four things. First, a copper and an aluminum reference price in $/lb, converted from the IMF's global price series in $/tonne at 2,204.62 lb per tonne; July 2026 is [$13,543 a tonne for copper](https://fred.stlouisfed.org/series/PCOPPUSDM) and [$3,158 for aluminum](https://fred.stlouisfed.org/series/PALUMUSDM), which is $6.14 and $1.43 a pound. Second, a price sheet of bare metal value in $/ft and $/kft for thirteen NEC conductor sizes, #4 AWG to 600 kcmil in copper and 4/0 AWG to 900 kcmil in aluminum, each with its 75°C ampacity beside it. Third, an escalation-clause calculator that takes the conductor line price at bid, the index on the bid date and on the PO date, the copper content in pounds or as a share of the line, and a deadband, and returns the adjustment. Fourth, a log of distributor commodity bulletins, fourteen rows from Border States between July 2025 and August 2026, with the COMEX and aluminum opening prices each bulletin printed. Every input is mirrored to the URL, so a scenario is a link: enter the COMEX settlement and the conductor, copy the link, and the estimator who opens it sees the same $/ft and the same adjustment. **The sheet does not publish a finished wire price, and it says so on every result: metal value is a third to a half of what 12 AWG THHN costs at the counter and most of what 600 kcmil costs.** The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is the model the sheet is built on, and finished THHN and MC cable bands from public awards arrive there, not here, when the cell rules are met. ## The Reference Price Is LME and Your Bid Clears on COMEX The default copper price is the IMF's global copper price as published by the St. Louis Fed's FRED, series [PCOPPUSDM](https://fred.stlouisfed.org/series/PCOPPUSDM), a monthly average of the LME cash price for grade A cathode, in US dollars per metric ton. The aluminum default is the sibling series [PALUMUSDM](https://fred.stlouisfed.org/series/PALUMUSDM). We chose them because they are free, documented, dated and identical for every reader. They are monthly, and the sheet treats them as monthly. **The July 2026 cell is one number, $6.14, from the first of the month to the last, and the sheet does not interpolate it to a weekly print.** When FRED posts August, the sheet posts August. Between releases the Monday refresh re-reads the series, re-reads the PPI and adds any new bulletin to the log. The reference price changes only when the source does. The problem with an LME basis is that US building wire is priced off COMEX, and since the copper Section 232 action the two have diverged. The [copper tariff piece](https://open-factory.com/articles/copper-tariff-panel/) compiled the COMEX premium over LME from exchange reporting in the press: [about $2,937 a tonne in late July 2025](https://think.ing.com/articles/whats-next-for-us-copper-import-tariffs/) when a duty on all copper was feared, [$130 through mid-September](https://www.metalminer.com/2025/09/16/comex-price-of-copper-holds-premium/) after refined copper was exempted, and [close to $1,000 for the December 2026 contract by mid-August 2026](https://www.cnbc.com/2026/08/14/copper-trump-tariffs-metal-commodities-trade-war.html). The sheet does not carry a COMEX series; CME's settlement data is licensed, and a free delayed quote is not a series anyone can cite in a contract. So the sheet states the gap instead, prints the LME-based default beside a field that says "enter the COMEX settlement of the day", and shows the gap as the distributor reports it. Border States printed a [COMEX open of $5.45 on July 14, 2025](https://tedmag.com/border-states-commodity-update-july-2025/) against an IMF monthly average of $4.43, a $1.02 gap in the week the market priced a tariff on all copper; the gap was $0.01 in February and March 2026, $0.02 in April, and then $0.14, $0.16 and $0.14 in May, June and July as the refined-copper question reopened. A daily open against a monthly mean is an indicative gap, not a settlement spread. It is also the only free, dated, distributor-printed COMEX record we have found, which is why the sheet logs it. The tariff itself is stated with its instruments, because the sheet's users write contracts. [Proclamation 10962 of July 30, 2025](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states) put 50% on semi-finished copper products and intensive copper derivatives from August 1, 2025, and left refined copper out, the content rule the [Tariff Bill of Materials](https://open-factory.com/articles/tariff-bom-electrical/) worked through a UPS, a lineup and a Megapack; [Proclamation 11021 of April 2, 2026](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states) applied the duty to full customs value from April 6, 2026; the investigation began with [Executive Order 14220 of February 25, 2025](https://www.federalregister.gov/documents/2025/02/28/2025-03439/addressing-the-threat-to-national-security-from-imports-of-copper). The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) lists each instrument by HTS line with its Federal Register citation, and the [September 2026 tariff schedule](https://open-factory.com/articles/tariff-schedule-2026/) has the metals stack at 50, 25, 15 and 10 percent. The sheet links to those and does not restate the rates. ## From a Pound to a Foot Metal value per foot is conductor metal weight in pounds per thousand feet, times the metal price in $/lb, divided by 1,000. The weight is computed, not read from a catalog. [NEC Chapter 9 Table 8](https://www.nfpa.org/codes-and-standards/nfpa-70-standard-development/70) gives the conductor area in circular mils for every size: 41,740 for #4 AWG, 133,100 for 2/0, 211,600 for 4/0, and the kcmil sizes are their own area. One kcmil of cross-section over 1,000 ft is 9.425 cubic inches; copper at 8.89 g/cm³ is 0.3212 lb per cubic inch, so a kcmil-kft of copper weighs 3.027 lb, and aluminum at 2.70 g/cm³ weighs 0.919 lb. Stranded conductors carry about 2% more metal than the nominal area because each strand spirals around the axis, so the sheet multiplies by 1.02. The result is 411 lb per kft for 2/0 copper, 653 for 4/0, 1,544 for 500 kcmil and 1,853 for 600 kcmil; 4/0 aluminum is 198 lb and 900 kcmil aluminum 844. At the 2024 average of $4.15, 2/0 copper carried $1.71 of metal a foot and 600 kcmil $7.69; at the sheet's July 2026 default of $6.14 they are $2.52 and $11.38; at the [$6.60 COMEX open Border States printed on August 10, 2026](https://tedmag.com/border-states-commodity-update-august-2026/) they are $2.71 and $12.23, so the LME-to-COMEX choice is worth $0.19 a foot on 2/0 and $0.85 on 600 kcmil that week. Ampacities in the 75°C column come from [Cerrowire's chart](https://www.cerrowire.com/products/resources/tables-calculators/ampacity-charts/), which reproduces NEC 310.16 on a 2017 basis: 175 A for 2/0 copper, 230 A for 4/0, 380 A for 500 kcmil, 420 A for 600. The 900 kcmil aluminum row has no ampacity in the sheet because Cerrowire's chart stops at 750, and we print a blank rather than a number we did not source. Two cautions travel with the weights. They have not yet been checked against a Southwire or Encore spec sheet, which state bare conductor weight to the pound; expected agreement is within 2%, and the version history will record the check. And they are bare metal: a 500 kcmil THHN conductor weighs more than 1,544 lb per kft once the PVC and nylon are on, and the sheet does not price insulation. ## The Escalation Calculator, With the Numbers In The clause the calculator applies is the one published in [the copper tariff piece](https://open-factory.com/articles/copper-tariff-panel/), on the [ConsensusDocs 200.1](https://www.consensusdocs.org/contract/200-1/) form: the conductor line is priced at the COMEX front-month settlement on the bid date, applied to a stated number of pounds of copper content per the attached schedule, and if the settlement on the PO date differs from the bid price by more than 5%, the line moves by the full difference times the pounds, in either direction, with no markup on the adjustment. The calculator does exactly that arithmetic. Adjustment equals index at PO minus index at bid, times pounds; if the move is inside the deadband the adjustment is zero; if you have the line price and the copper share but not the pounds, pounds equals line price times share divided by the index at bid. Take 1,000 conductor feet of 500 kcmil copper, 1,544 lb, bid on February 9, 2026 at the [$5.88 COMEX open Border States printed that morning](https://tedmag.com/border-states-commodity-update-february-2026/) and released to PO on July 13 at [$6.28](https://tedmag.com/border-states-commodity-update-july-2026/). The move is 6.8%, outside a 5% deadband, so the line rises by $0.40 times 1,544 lb, or $618 on $9,079 of metal at bid. The trigger at $5.88 was a $0.29 move. A 400 A feeder in parallel 3/0 over 100 ft holds 414 lb and moves $166 on the same dates. A fixed-price bid signed in February ate that, and a bid with the clause billed it, in both directions. **The clause needs the copper content in pounds, and the sheet prints the pounds for any conductor schedule, which is the one number most bids do not carry.** Two public documents show a clause without an index. The [Eaton terms Silicon Valley Power filed in January 2026](https://open-factory.com/articles/switchgear-quote-anatomy/) reserve the right to invoice at prices in effect at shipment regardless of any prior bid, with no index, base date or cap. The [City of Ocala autotransformer order](https://open-factory.com/articles/rfq-capital-equipment/) included a seller-chosen PPI escalator that added $744,800, 9.2% of the contract. The calculator is the buyer's version: name the series, the base date, the pounds and the deadband, and the number is arithmetic anyone can check. ## The Wire Index Beside the Metal The sheet lists one price that is not metal: the BLS producer price index for copper wire and cable, [WPU10260314](https://fred.stlouisfed.org/series/WPU10260314), because it is the only public series of what US mills actually charged for finished conductor. It stood at 578.2 in July 2026, up 17.9% on July 2025 and 95% on December 2019. From January 2015 to July 2026 the metal index went from 100 to 232 and the wire index from 100 to 198. The wire index sat above the metal from mid-2022 to mid-2024, when copper fell and mills held price, and below it from 2025, when copper ran faster than mills could reprice. The gap is the mill's and distributor's margin moving, and the [switchgear cost decomposition](https://open-factory.com/articles/switchgear-50-percent/) used the same two series to show copper up 70% and switchgear PPI up 78% since January 2021. The sheet prints the PPI as a monthly index beside the monthly metal price, as the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) prints the transformer and switchgear PPIs, and neither pretends an index is a price. ## How a Distributor Bulletin Is Logged Southwire and Encore price letters do not reach the public; distributor commodity bulletins do, and Border States publishes one on the second Monday of each month that [tED magazine](https://tedmag.com/tag/border-states/), the NAED's trade title, reprints in full. Each bulletin prints the COMEX copper opening price and the US delivered aluminum opening price that morning, and reports any wire price action suppliers announced. We log one row per bulletin: date, both opening prices, the IMF monthly average for the same month for comparison, the action, and the tED URL. Fourteen bulletins from July 2025 to August 2026 are in the log; thirteen carry a copper price, and the November 2025 reprint had none, which the row says. [CSV: data/distributor-bulletin-log.csv](data/distributor-bulletin-log.csv) Two rows are dated price actions: [finished aluminum goods up 7% to 10% on March 9, 2026](https://tedmag.com/border-states-commodity-update-march-2026/) and [copper suppliers up 7.5% on July 10, 2026](https://tedmag.com/border-states-commodity-update-july-2026/). One row is a count that no other public source carries: the [January 12, 2026 bulletin](https://tedmag.com/border-states-commodity-update-january-2026/) reports 27 price increases for copper wire and six for aluminum wire during 2025, against a 2025 COMEX average of $4.82, 60 cents above 2024. Twenty-seven increases is one every 13 days, the cadence a fixed-price bid is exposed to. A row is logged only from the reprinted bulletin, never from a call or a screenshot, and every row says "Border States" because it is one distributor's view of the market, not the market. The aluminum column of the log is the Midwest Premium made visible. The sheet's aluminum default is the LME-based IMF price, $1.43 a pound in July 2026, and Border States opened aluminum at $2.53 that month, because US delivered aluminum carries the premium under the [50% Section 232 aluminum tariff](https://www.federalregister.gov/documents/2026/04/09/2026-06960/strengthening-actions-taken-to-adjust-imports-of-aluminum-steel-and-copper-into-the-united-states). The gap was $0.68 in July 2025, crossed $1.00 in February 2026 and reached $1.27 in June 2026 when US delivered aluminum opened at $2.83 against an LME month of $1.56, so an aluminum feeder priced off the sheet's default understates US metal by 70 to 80%, and the sheet's aluminum field says to enter the delivered price for the same reason the copper field says to enter COMEX. The aluminum substitution math for a 400 A feeder is in [the copper tariff piece](https://open-factory.com/articles/copper-tariff-panel/), and the imported-rod duty is in the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/). ## Every Field and Its Source [CSV: data/sheet-fields.csv](data/sheet-fields.csv) Ten fields, and the sources by name are the International Monetary Fund and the Federal Reserve Bank of St. Louis for the metal prices, the Bureau of Labor Statistics for the PPI, the National Fire Protection Association for NEC Chapter 9 Table 8, Cerrowire for the ampacity table, Border States Electric and tED magazine for the bulletins, CME Group's COMEX for the settlement the sheet asks you to enter, ConsensusDocs for the 200.1 form, and the Federal Register for the tariff instruments. The [Lead-Time Monitor launch](https://open-factory.com/articles/lead-time-monitor-launch/) and the [Bid-Tab Price Book launch](https://open-factory.com/articles/bid-tab-price-book-launch/) describe their compilations the same way. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades who publishes at all. ## Known Gaps The sheet has no finished-cable price. Metal value per foot is not a THHN, XHHW or MC cable price; drawing, insulation, jacket, reel, mill margin and distributor margin sit on top, and the share they take falls as the conductor grows. On 12 AWG the copper is about 20 lb per kft, so at $6 copper the metal is $121 of a counter price near $300; on 600 kcmil the metal is most of the price. The sheet says this in the method note under every result and prints no counter price, because we have none we can source. The sheet has no distributor net price. Graybar's [2025 gross margin was 19.3%](https://www.sec.gov/Archives/edgar/data/205402/000020540226000015/c402-20251231x10k.htm) and WESCO's [21.1%](https://www.sec.gov/Archives/edgar/data/929008/000092900826000008/wcc-20251231.htm), both blended across every product line, and neither says what a contractor pays for 500 kcmil on a Tuesday. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) will publish wire lines from municipal and utility bid tabulations, by conductor size and award date with n, once a cell has five or more awards; nothing prints before that. The sheet has no COMEX series, only a field and a log of one distributor's opening prices, which are not settlements. The conductor weights are unchecked against a mill sheet, the 900 kcmil aluminum ampacity is blank, and the reference price is a month old on the day it posts, because the IMF publishes after the month ends. The sheet names the month on every result rather than guessing at a weekly value. ## What to Do Monday Open the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/), enter the COMEX settlement from your distributor's quote or your own terminal, pick the conductors on your schedule and read the pounds. Put three things in the bid: the settlement, the date, and the pounds per line. Then paste the clause and set the deadband; at $6.14 copper, 5% is a $0.31 move, and the calculator shows what a $0.40 move costs on your schedule before the estimator signs. Check the gear lines of the same package against [Quote Check](https://open-factory.com/tools/quote-check/), which returns the Bid-Tab band and the Monitor's published lead time for a switchboard or transformer spec, and run the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) so the conductor PO lands the week the gear is released, not the week the bid was won. If the run is 400 A or larger, price the aluminum case at the delivered price, not the sheet's LME default, with the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) open for imported rod. Contribute the wire lines from your last three quotes to the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/). It publishes nothing until a cell has five contributors and eight quotes, and wire is the class where that will happen first. ## What Version 1 Adds Three things, in the order they are likely to land. The bulletin log grows from one distributor to several: Graybar, WESCO, CED and Rexel publish commodity notes to customers, and mills publish price letters that reach distributors; version 1 logs every one obtainable as a public document, same row format, same rule that a row comes from the document and not from a call. Finished THHN and MC cable price bands arrive from the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/): municipal and utility bid tabulations for building wire by conductor size, normalized to $/kft with the award date, region and n, so the sheet can show metal value and awarded price on the same row for the same size. And the COMEX settlement replaces the blank field, under a data license from CME Group, so the sheet's default becomes the US price a bid actually clears on and the LME-based series moves to the comparison column. Until then the reference price is monthly, the gap is stated, and the pounds are the number to put in the bid. --- # Introducing the Large-Load Cost Table: 15 Markets Side by Side URL: https://open-factory.com/articles/large-load-cost-table-launch/ Date: 2025-10-01 Power / Model launch Study fees, security per MW, minimum bills, terms, exit fees and realistic energization year for Dominion, AEP Ohio, Georgia Power, Oncor, CenterPoint, Duke, APS, SRP, PacifiCorp and six more, 53 cells from primary documents and 45 not published Fifteen utilities, eight tariff cells each, and 45 of the 120 cells have no published number. That is the honest headline of the [Open Factory Large-Load Cost Table](https://open-factory.com/models/large-load-cost-table/), which launches as version 0.1: for a 50 MW connection in fifteen US markets, the study fee, security per MW, minimum bill, contract term, ramp, exit fee, CIAC exposure and a realistic energization year, each cell tied to the tariff sheet, commission order, statute or dated statement it came from. Fifty-three cells come from a primary document, 20 from a trade-press report of a filing we could not open, 2 from our arithmetic on a disclosed total, and 45 from nothing, because the utility has put the number nowhere public. The [15-market walk-through](https://open-factory.com/articles/what-50mw-costs-15-markets/) argues what the numbers mean. This piece is about the table as a product. It covers what each column means and how it was read from a tariff or a docket, the verification flag and what "not published" means, how the fifteen markets were chosen, the known biases, how the table is refreshed, how to use it with the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool, and what version 1 adds. Read the columns down: the threshold is a primary document in 10 markets and a press report in 2, the CIAC column is documented in 14 of 15, and the study fee is published by one utility of the fifteen (AEP Ohio, on its own page) and reported for one more (Xcel Colorado, by the Colorado Sun). The full table, 15 rows by 22 columns with a source URL and date in every row, is [large-load-cost-table-v0.1.csv](data/large-load-cost-table-v0.1.csv), and the cell-by-cell coding behind the figure is [verification-matrix.csv](data/verification-matrix.csv). ## Nine Columns, One Row per Market Each row is one utility territory and the instrument governing a new large load there. The nine cells a buyer reads first, and how each was read: **Threshold** is the MW at which the large-load regime starts, read from the tariff's applicability clause or the statute's definitions. Dominion's Schedule GS-5 applies at 25 MW of contract demand per the SCC's [November 25, 2025 order in PUR-2025-00058](https://www.scc.virginia.gov/docketsearch/DOCS/8dt001!.PDF) as carried in the [LBNL and Brattle brief](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf); AEP Ohio's Schedule DCT at 25 MW aggregated across affiliates per the [utility's own tariff page](https://www.aepohio.com/company/about/rates/data-center-tariff/); Texas SB 6 at 75 MW; Georgia's PSC rule at 100 MW per the [January 23, 2025 release](https://psc.ga.gov/site/assets/files/8617/media_advisory_data_centers_rule_1-23-2025.pdf). A yes/no column, `applies_to_50mw`, records which side of the line a 50 MW load falls on, because nothing else in the row applies until you know. **Study fee** is the one-time fee to enter the utility's load study. AEP Ohio prints three tiers on its page: $10,000 for more than 25 to under 50 MW, $50,000 for 50 to under 100 MW, $100,000 at 100 MW and above, payable within 45 days. The PUCT's proposed 16 TAC 25.194 sets $100,000 for 75 to 250 MW and $300,000 above, plus actual study costs, per [Greenberg Traurig's summary of the March 12, 2026 proposal](https://www.gtlaw.com/en/insights/2026/3/texas-senate-bill-6-update-what-data-centers-large-load-customers-should-know-about-proposed-interconnection-standards); below 75 MW the TDSP's own study fee is not published, so the Texas cells at 50 MW say so. Thirteen of fifteen study-fee cells are "not published." **Security or collateral** is what you post before energization, and the table keeps three sub-columns: the instrument as written, the dollar figure for 50 MW where it can be computed, and the basis of that computation. Dominion's $1.5 million per MW of contracted capacity is a tariff figure ([LBNL brief, p. 6](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)), so the 50 MW cell is $75 million with basis "tariff." ComEd's cell is $12.5 million with basis "implied," because Exelon told analysts that [about 4 GW of signed Transmission Security Agreements had posted $1 billion](https://www.utilitydive.com/news/exelon-data-center-load-new-jersey-battery-earnings/826686/), a portfolio average of $250,000 per MW, not a quote. AEP Ohio's collateral, 50% of total minimum charges over the term, is recorded as a formula and not dollarized, because that needs your demand charge. **Minimum bill** is the share of contract demand billed whether or not consumed, and the table records the formula, not the headline. AEP Ohio's page says 85%, but its formula for 25,001 to 75,000 kW, 15,000 kW plus 85% of capacity above 25,000 kW, gives 36,250 kW at 50 MW, or 72.5%. Dominion's is three numbers, 85% for distribution, 85% for transmission and 60% for generation ([LBNL brief, p. 5](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf)). Where the regime has no minimum by design, the cell says "none" and why: an ERCOT TDSP delivers power a retail electric provider sells, so no take-or-pay contract carries a floor. **Contract term**, **ramp** and **exit fee** are read together, because they interact. AEP Ohio's term is a ramp of at most four years at 50, 65, 80 and 90% of contract plus eight years, its exit fee 36 months of minimum charges allowed only after the fifth year following the ramp ([AEP Ohio tariff page](https://www.aepohio.com/company/about/rates/data-center-tariff/)). Indiana Michigan Power's settlement, approved February 19, 2025 in IURC Cause 46097, runs at least 12 years after a ramp of up to five and lets a customer cut up to 20% after year five with 42 months' notice free ([AEP release](https://www.aep.com/news/stories/view/10037/); [Utility Dive](https://www.utilitydive.com/news/indiana-michigan-power-aep-amazon-google-microsoft-data-center-interconnect/733850/)). Where a tariff says an exit fee exists but not how much, as Dominion's GS-5 does in the documents we opened, the cell says "amount not published." **CIAC exposure** is what you pay toward the wires, and at 14 of 15 it is the best-documented column because it is where regulators have written the most: the SCC's [July 31, 2026 final order in PUR-2026-00056](https://www.scc.virginia.gov/docketsearch/DOCS/8dt001!.PDF) makes a contribution mandatory for "direct connect" substations and lines net of expected transmission revenue; PG&E's [Electric Rule 30](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_RULES_30.pdf) takes an advance equal to the project-specific estimate for service facilities, interconnection upgrades and network upgrades at 50 to 230 kV and refunds it over up to ten years; the Texas proposal takes 100% of radial lines and substations with no revenue credit. **Realistic energization year** is the one cell that is an estimate. Its `energization_basis` column says what it rests on. ## How a Cell Is Read The source hierarchy is fixed and the CSV's `source` column shows which rung each row reached. First, the tariff sheet or the commission's order, read directly: PG&E Rule 30, the SCC's two Dominion orders, Georgia Power's TOU-SC-16 sheet, Oregon's [Order 26-154](https://apps.puc.state.or.us/orders/2026ords/26-154.pdf). Second, the statute or the utility's own page or release: AEP Ohio's tariff page and its [February 13, 2026 load update](https://www.aepohio.com/company/news/view?releaseID=10753), the Georgia PSC's release and [fact sheet](https://psc.ga.gov/site/downloads/datacenterfactsheet.pdf), ERCOT's [June 18, 2026 Batch Zero release](https://www.ercot.com/news/release/06182026-puct-approves-ercots) and [April 1, 2026 Senate deck](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf), the [Governor's August 3, 2026 letter](https://gov.texas.gov/uploads/files/press/Thomas_Gleeson_Pablo_Vegas_Data_Centers_Directive_Letter_to_PUCT_ERCOT_August_2026_.pdf). Third, a national-lab compilation that read the tariff and cites the docket, which means the LBNL and Brattle brief of August 10, 2026 with its Appendix 1 of docket numbers. A page-cited LBNL figure counts as primary and the cell says "LBNL p. 6." Fourth, a trade-press report of a filing we could not open, printed "as reported by": Latitude Media on [Duke's Schedule HLF filing](https://www.latitudemedia.com/news/duke-energys-light-touch-large-load-tariff-could-be-a-problem/) in NCUC E-7 Sub 1329, the Colorado Sun on [Xcel's 26AL-0137E filing](https://coloradosun.com/2026/04/02/data-centers-xcel-energy-colorado-rate-tariff-consumers/), Utility Dive on the [FERC orders approving ComEd's TSAs](https://www.utilitydive.com/news/ferc-comed-exelon-data-center-service-agreement/814414/). Fifth, arithmetic on a disclosed total, labelled "implied." AEP Ohio is 8 of 8 primary, PG&E and Oncor 7, Dominion and AEP Texas 6, CenterPoint 5 plus one implied; at the other end Xcel Colorado and Duke are 7 and 6 press-reported cells with none primary, ComEd 4 press plus one implied, Rocky Mountain Power 2 press, and APS and Entergy one primary cell each. The market-level flag follows from that: VERIFIED where every filled cell is primary (Dominion, AEP Ohio, PG&E), PARTIAL where the regime is verified but cells rest on press or are missing, THIN where only the statute and the docket number are confirmed (Rocky Mountain Power, Entergy Louisiana). **"Not published" means the number was not in any document we could open, and it is a finding about the utility, not a gap in our reading.** Dominion's public [business rates page](https://www.dominionenergy.com/virginia/rates-and-tariffs/business-rates) lists Schedules GS-1 through GS-4 and no GS-5 sheet as of September 2026, ten months after the SCC created the class; the collateral, term and minimums are in the order and the LBNL brief, the study fee and the exit-fee dollar amount are not. Georgia Power's 100 MW-plus contracts are customer-specific by design and filed with the PSC 30 days before execution, so the fee and collateral cells will stay "not published" until a contract is made public. APS's XHLF update is a [pending rate case](https://www.aps.com/en/Utility/Company-wide-Events-and-Messages/Data_Centers_How_Were_Protecting_Customers_While_Planning-for_Big_Energy_Needs), E-01345A-25-0105, with the formulas in testimony we did not open. The 45 cells fall as 13 study fees, 7 ramps, 7 exit fees, 6 security formulas, 4 minimum bills, 4 terms, 3 thresholds and 1 CIAC. A cell that says "none" is different: it means the regime has no such term, and the document that shows that is cited. Every cell also includes a status and a date. Proposals are marked (p) in the figures and "pending" in the `status_as_of` column. The PUCT's 25.194, Duke's HLF, Xcel's Colorado tariff and APS's XHLF are opening positions, not rules. An approved tariff carries its approval date; PG&E's Rule 30 carries "interim," because its refund rules are still being set in Application 24-11-007. ## How the Fifteen Markets Were Chosen Three filters, applied in order. First, the territories holding the most requested or contracted large load: Dominion's queue of about [70 GW against a 24.6 GW peak](https://www.utilitydive.com/news/states-ferc-large-load-tariff-doe-christie/816672/), Oncor's [298 GW of requests](https://www.utilitydive.com/news/fate-of-oncors-nearly-300-gw-load-pipeline-unclear-following-texas-data-ce/827303/), AEP Ohio's 17,861 MW under contract, Southern's [17 GW contracted](https://www.utilitydive.com/news/southern-co-contracted-large-load-data-centers/826919/), CenterPoint's [14 GW of Batch Zero-eligible load](https://www.utilitydive.com/news/centerpoint-energy-14-gw-texas-batch-zero-large-load-earnings/826329/), Duke's [7.6 GW executed](https://www.utilitydive.com/news/duke-energy-earnings-anderson-gas-plant/819428/), ComEd's roughly 9 GW of high-probability load, PG&E's [12.7 GW pipeline](https://www.utilitydive.com/news/pge-claims-127-gw-in-data-center-pipeline-as-utility-courts-smaller-loads/826099/). Second, the five markets JLL charts by name for a new 50 MW data center in its [2026 outlook](https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-reports/26-research-global-data-center-outlook-new.pdf): Northern Virginia, Dallas, Phoenix, Atlanta and Chicago. Third, one instance of each regime type, so a factory in a territory we do not cover can find its nearest analogue: PJM take-or-pay tariffs (Dominion, AEP Ohio, I&M), a PJM wires company on FERC-jurisdictional agreements (ComEd), ERCOT TDSPs under a statute (Oncor, CenterPoint, AEP Texas), Southeastern non-RTO utilities (Georgia Power, Duke) and a MISO one (Entergy), Western utilities under a rate case (APS), a public power board (SRP), a state law (PacifiCorp under Utah SB 132) and a proposed tariff (Xcel Colorado), and CAISO under a cost mechanism rather than a tariff (PG&E Rule 30). What that leaves out is the subject of the biases section: no cooperative, one public power district, nothing in New York, the Pacific Northwest or the Plains. Portland General's Schedule 96 is a context row, not a market. ## What the Table Dollarizes Only two of fifteen markets publish a security figure per MW. Everything else the table can put a dollar sign on is a fee, a deposit or our arithmetic, and the chart says which. Dominion's collateral is $75.0 million at 50 MW and $150.0 million at 100 MW; ComEd's implied credit is $12.5 million and $25.0 million; a Texas project posts nothing at 50 MW and about $10.1 million at 100 MW under the proposed 25.194: $5 million of security of which 80% is forfeited on withdrawal, a $5 million non-refundable interconnection fee and a $100,000 study ([ERCOT deck, slide 9](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf)). CenterPoint's implied deposits are $3.2 million and $6.4 million on top of that, Xcel's proposed $600,000 does not scale with MW, and AEP Ohio's study fee is $50,000 and $100,000 while its collateral waits for your rate. Georgia Power at 100 MW enters the PSC rule and the collateral becomes customer-specific, so that bar cannot be drawn. The minimum bill dollarizes for four tariffs, because four publish both a floor and a term. Xcel Colorado's proposed 80% over 15 years commits 600 MW-years for a 50 MW contract, Dominion's 85% over 14 years with its 20%-a-year ramp commits 510, AEP Ohio's 72.5% formula over 12 years with its 50/65/80/90 ramp commits 386, and Duke's proposed 75% over 10 years commits 375. SRP's E-67 floor of 80% of forecast has no published term and Georgia's is contract-specific, so neither gets a bar. At $10 per kW-month, which is an assumption the [Interconnection Cost tool](https://open-factory.com/tools/interconnection-cost/) makes you replace, Dominion's 510 MW-years is $61.2 million of demand charges due whether or not the plant runs. Before any of that, the threshold. It is the first test the tool runs and the first column in the CSV. SRP at 20 MW of forecast load, Dominion and AEP Ohio at 25, Duke's proposal and Xcel's at 50 sit at or below a 50 MW request; I&M at 70 MW per site, the three Texas TDSPs at 75, and Georgia Power and Rocky Mountain Power at 100 sit above it, per [SRP's general manager](https://azcapitoltimes.com/news/2026/03/31/the-facts-about-srp-and-data-centers/), the [Utah SB 132 summary](https://www.latitudemedia.com/news/utah-is-taking-a-different-approach-to-new-data-center-load/) and the documents above. APS, Entergy, ComEd and PG&E publish no MW threshold; ComEd's TSAs and PG&E's Rule 30 are defined by transmission-level service, not by megawatts. ## Energization Year Is an Estimate, and Says So The last column is the one buyers ask about first and the only one no utility publishes as a number. The table carries it as an Open Factory estimate with the basis printed beside it. Houston, Phoenix under SRP, Salt Lake City and northern Indiana are 2028 to 2029; Dallas, AEP Texas, Phoenix under APS and Louisiana 2029; the Carolinas and Colorado 2029 to 2030; Atlanta and Chicago 2031; the Bay Area 2031 to 2032; Columbus 2031 to 2033; Northern Virginia 2033. The bases are of three kinds. JLL's 2026 outlook charts average grid-connection lead time for a new 50 MW data center by market, about 2.5 years in Dallas and Phoenix, 5 in Atlanta and Chicago, 7 in Northern Virginia ([JLL, p. 18](https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-reports/26-research-global-data-center-outlook-new.pdf)), and we read those values off a bar graphic, not from text, which the basis column discloses. A CPS Energy planning presentation at PowerGen 2026 puts Columbus at 84 months and Silicon Valley at 72, and reaches us second-hand through [Avanza Energy](https://avanzaenergy.substack.com/p/data-centers-are-killing-the-grid). Utility statements fill the rest: CenterPoint's [8 GW energized by 2029 with 3.5 GW under construction](https://www.utilitydive.com/news/centerpoint-energy-data-center-load-earnings/818293/), Meta's Entergy contract effective 2028 per [UCS's reading of the April 2026 filing](https://blog.ucs.org/paul-arbaje/meta-hides-data-center-info-as-it-pushes-more-risks-to-entergys-customers/). **The year in that column is gated by transformers as much as by dockets**: the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists 128 weeks for power transformers, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tells you whether a utility promising 2028 has ordered the unit yet. ## Twenty Months of Rule Changes, and the Refresh Every cell has a date because most of them changed between January 2025 and August 2026. The [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) keeps the list. The table records which entry changed which cell. Seven actions in 2025, from Georgia's rule on January 23 to Dominion's GS-5 on November 25, then fourteen in the five months from the PUCT's forecasting rule on March 1, 2026 to its good-cause exception for ERCOT on August 20, with Batch Zero and FERC's show-cause orders landing on the same day, June 18. The rows and their documents are in [04-rule-timeline.csv](data/04-rule-timeline.csv). The refresh follows the dockets: every row is reviewed on the first business day of the month, and a cell changes the day its document lands, with the old value kept in the history and the docket entry dated. Already on the calendar: AEP Ohio's second study tranche opens October 1, 2026; the PUCT's 25.194 proposal reaches the six-month mark from publication on September 12, 2026, after which Texas law treats an unadopted rule as withdrawn ([PUCT SB 6 roadmap memo](https://interchange.puc.texas.gov/Documents/58317_5_1526835.PDF)); Dominion's amended line-extension policy is due within 90 days of July 31; ERCOT's audit report is [targeted for December 10](https://www.utilitydive.com/news/ercot-texas-puc-data-center-audit/828472/); PJM's backstop auction results land December 2; Duke's HLF would start in 2027. The headline table stays free. The history of every cell is Pro, and docket alerts, sent when a cell in a market you name changes, are institutional. ## Known Biases Six, and each is a reason to read the flag before the number. First, the table is large investor-owned utilities plus one public power district. No cooperative or municipal utility is in it, although co-ops serve much of the land where 50 MW plants get built. Second, sub-75 MW regimes are underrepresented by construction: the three Texas rows describe an SB 6 regime a 50 MW load is outside of, and the TDSP facilities-extension fees that govern it are the cells we could not find. The sub-5 MW line-extension tariffs we did read for Ohio, Georgia, California and Illinois live in the [2 MW service upgrade piece](https://open-factory.com/articles/2mw-plant-service-upgrade/), not yet here. Third, docket portals fought the fetcher: PUCO's DIS, the NCUC, Utah PSC, ACC, LPSC and ICC systems are JavaScript-only or returned errors, Oncor publishes its tariff as a PDF our text fetcher refuses, georgiapower.com and srpnet.com returned 403 or 404 during compilation, and Dominion's rates page opens with no GS-5 sheet on it. That is why 20 cells rest on press reports, and version 0.2 fixes it first. Fourth, the two implied figures are portfolio averages: CenterPoint's $64,000 per MW is $900 million over 14 GW, and a 50 MW load beside an existing 138 kV substation will post less and a greenfield one more. Fifth, the JLL and CPS Energy lead times were read from a graphic and a second-hand report. Sixth, the table is a snapshot with four proposals in it: Duke's 75% floor and 25% exit penalty are what Duke asked for, not what the NCUC will order. ## Using It With the Interconnection Cost Tool The [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) tool is the table with your numbers in it. It takes a market, a requested MW and your demand charge in $/kW-month, and returns whether the regime applies at your size, the study fee and security scaled by MW where the source gives a per-MW figure (labelled implied where it does not), the minimum-bill exposure in MW-years and dollars with the published ramp applied, and the energization year with its basis. Cells the sources do not publish render as "not published" in the tool too. It estimates nothing the table does not. Run 50 MW in Northern Virginia at $10 per kW-month and it returns GS-5 applies, $75 million of collateral, 510 MW-years and $61.2 million of minimum demand charges over 14 years, exit fee amount not published, 2033. Run the same 50 MW in Houston and it returns SB 6 does not apply, no rule-mandated security, deposits implied at about $3.2 million, no minimum bill, 2028 to 2029, and a note that the Governor's audit reaches down to 25 MW. Run 100 MW in Dallas and the Texas stack appears: $100,000 study, $5 million security, $5 million fee, 100% CIAC. In a PJM market the tool hands you to the [PJM Capacity Charge Estimator](https://open-factory.com/tools/pjm-capacity-bill/), because no tariff here waives the capacity charge and at $333.44 per MW-day a 50 MW load carries about $6.1 million a year of it ([the PJM guide](https://open-factory.com/articles/pjm-capacity-factory-bill/)). Where the minimum bill exceeds a bridge's cost, the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool and [Recips vs Turbines vs Fuel Cells](https://open-factory.com/articles/turbines-recips-fuel-cells/) are next. Where a CIAC hands you the transformer, the [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool prices it. ## What to Do Monday Pull [the CSV](data/large-load-cost-table-v0.1.csv) and sort on the `verification` column before anything else. A VERIFIED row can go into a board memo with the docket number beside it; a PARTIAL row goes in with "as reported by" and the outlet; a THIN row is a phone call, not a number. For every cell in your market that reads "not published," ask the utility for it in writing before the letter of intent, in the table's units: study fee in dollars, collateral formula in dollars for your load, minimum demand in kW for your contract capacity, exit fee in months. Then run the threshold test at your true size and one hall smaller: a 74 MW request in Texas is outside SB 6 and Batch Zero; a 49 MW request in Columbus saves $40,000 on the study fee and nothing else, because the DCT reaches down to 25 MW. If the document you receive contradicts a cell, send it to us. The table changes the day we can read it, the correction is dated, and the history keeps the old value. ## What Version 1 Adds Version 0.1 is the public record, read and dated. Version 1 adds four things. Developer interviews on actual study outcomes: what the study fee bought, how many months the study took against the tariff's 45 or 60 days, what the facilities estimate came back at, and the energization date offered against the year in our column; interview counts print on the model page only when true, the rule the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) follows for its panel. Sub-100 MW regimes as their own rows: the standard industrial and line-extension tariffs a 5 to 75 MW plant lands in, starting with the sub-75 MW TDSP facilities-extension terms in Texas and the four CIAC formulas already read for the 2 MW piece. Alerts, so a cell change in a market you name reaches you the day the docket moves. And the docket documents behind the twenty press-reported cells, opened in a browser session, so that Duke, Xcel, ComEd and Rocky Mountain Power move from PARTIAL and THIN to VERIFIED or to an honest "not in the filing." The [Texas funnel piece](https://open-factory.com/articles/ercot-funnel/) and the [15-market walk-through](https://open-factory.com/articles/what-50mw-costs-15-markets/) are the first two articles built on the table. The CSV under them is this file, and it will say version 0.2 when any of the above lands. --- # The $2,850 Report That Says Nothing URL: https://open-factory.com/articles/2850-report-says-nothing/ Date: 2025-09-29 Procurement / Take A $37,300 benchmark for a $12k to $3.8M category, no transaction prices, PPI resold to 2033, four tariff changes in 2025, IBISWorld vs SpendEdge vs Beroe, ten-category free recipe IBISWorld's ProcurementIQ profile for "Switchgear Systems in the US" prices the whole category at [one benchmark, $37,300 per system](https://www.ibisworld.com/united-states/procurement/switchgear-systems/27812691/). The 118 public switchgear awards in the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) run from [$12,492 for a two-way S&C pad-mount switch Naperville bought in 2019](https://naperville.legistar.com/LegislationDetail.aspx?ID=3637&GUID=C3AD00FB-2C58-428F-A53E-D68D4FA95577) to [$3,827,000 for the 34.5 kV lineup the same utility bought in September 2024](https://naperville.legistar.com/LegislationDetail.aspx?ID=12613&GUID=F54CCBA8-A39B-4B5C-A90E-EE7D33725DCF). **A $2,850 category report on switchgear has no transaction price, no lead time and no buyer-side benchmark, and every series it is built from is free, monthly and reproducible in about eight hours.** This piece audits six switchgear reports from their public pages, rebuilds the same category from PPI, Census M3, filings and bid tabs, prices the rebuild in hours, and, behind the wall, gives the free-source recipe for ten categories and the three cases where a paid report still earns its price. The 46 manual air-insulated pad-mount switches in the Book have a median unit price of $18,294 (n=46, 2019 to 2025, four buyers), the 66 fault-interrupter and automatic-transfer pad-mounts a median of $88,107 (n=66, 2019 to 2025), and the six metal-clad lineups a median of $2.38 million (n=6, 2018 to 2025). The $37,300 benchmark sits above the first band, below the second, and at 1.6% of the third. ## What the Sample Pages Show We checked six products a switchgear buyer would find in an afternoon of searching: IBISWorld's industry report and its ProcurementIQ profile, SpendEdge, Beroe, MarketsandMarkets, and the free Mordor Intelligence summary. We read only public pages: tables of contents, methodology notes, pricing pages and free previews. Every cell below carries its URL in [the data file](data/02-report-contents-matrix.csv). Six products, zero transaction prices, zero lead times, two buyer-side scores (IBISWorld's buyer power score on a -5 to +5 scale and Beroe's 18-question benchmark), and one published refresh cadence. IBISWorld sells the [Industry Research Report at US $2,850 and the Spotlight at $800](https://help.ibisworld.com/en/articles/8119074-how-much-does-it-cost-to-purchase-one-report). Its US collection has no switchgear-only title: the report is [Electrical Equipment Manufacturing, NAICS 33531](https://www.ibisworld.com/united-states/industry/electrical-equipment-manufacturing/795/), which covers transformers, motors, generators, switchgear and relays together, and its chapters are Performance, Products and Markets, Geographic Breakdown, Competitive Forces, Companies, External Environment, Financial Benchmarks and Industry Data: revenue, cost structure, market share above 5%, valuation multiples. The help center says [reports update "1-4 times per year depending on changes in the market"](https://help.ibisworld.com/en/articles/8156155-about-ibisworld). Nothing on that chapter list is a price a buyer pays or a week a buyer waits. The ProcurementIQ profile is the one built for buyers, and its public page is specific about what it holds: a [Market Price "per system," a five-year price forecast, a buyer power score, vendor market share and cost structure, wage rates by geography, and "buying lead time" listed under RFP planning elements](https://www.ibisworld.com/united-states/procurement/switchgear-systems/27812691/). The same page says the category covers low-voltage indoor, low-voltage outdoor, medium-voltage indoor and outdoor, high-voltage, metal-enclosed, metal-clad, gas-insulated and open-air switchgear, and prices all of it at $37,300 per system, with prices "declined at a CAGR of -0.41 from 2023-26." The Dover, Delaware bid tabs show the [low bid for one PME-9 stock switch rising from $20,703 in November 2019 to $31,771 in September 2024](https://www.cityofdover.gov/media/Purchasing/Tabulations/FY%2025%20Info/25-0004WH%20Bid%20Tab.pdf), a 53% increase on one stock number, and Riverside's [15 kV metal-clad low bid went from $1,343,878 in 2018 to $2,509,063 in 2023](https://riversideca.legistar1.com/riversideca/attachments/ae7986e2-8ee3-4675-8f9c-b8dbb6c3040f.docx). A declining category average and a doubling unit price can both be true when the category is nine products wide. SpendEdge publishes its prices, which is more than most: [$99 a month read-only, $199 a month for 12 report downloads a year, $399 a month for 60, Enterprise by quote](https://www.spendedge.com/pricing/), against "100,000+ reports across 5,000+ categories." It publishes no switchgear sample: the [category index](https://www.spendedge.com/category-intelligence-reports) features paid-search software, architectural services and crypto asset management, and every "Unlock Report" button on it leads to the seven-day trial signup. Beroe's [switchgear category URL returns a 404](https://www.beroeinc.com/category-intelligence/switchgear-market/); its [Category Watch page](https://www.beroeinc.com/category-watch/) describes 3,000 categories, a Category Health Score across risk, price, availability and competition, an 18-question benchmark, and a "Request a demo" button where the price would be. MarketsandMarkets prices its [Switchgear Market report at $4,950](https://www.marketsandmarkets.com/search.asp?Search=switchgear); its [methodology text](https://www.marketsandmarkets.com/Market-Reports/switchgear-market-1162268.html) describes secondary sources (Factiva, OANDA) plus interviews, its units are USD billion and shipment units, and its "pricing analysis" is a chapter about a $110 billion global market, not a quote. Mordor's [free summary](https://www.mordorintelligence.com/industry-reports/switchgear-market) gives the global market at $131.64 billion for 2025 with segment shares to two decimals and no price for anything you can put on a pad. ## The Same Data, Free, in a Day Every number in those reports that is not an interview comes from a public series, and the biggest tell is that IBISWorld sells the series back. Its "Price Index" pages include [Switchgear, switchboard, industrial controls equipment, "available from 1994 with forecasts out to 2033"](https://www.ibisworld.com/united-states/price-index/switchgear-switchboard-industrial-controls-equipment/11.75/). That is BLS series WPU1175, downloadable free from [BLS](https://data.bls.gov/timeseries/WPU1175) or as a [FRED CSV](https://fred.stlouisfed.org/graph/fredgraph.csv?id=WPU1175) in one click, and it is the first input to the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). WPU1175 stood at 228.6 in January 2019 and 383.2 in August 2025, up 67.6%, with 9.1% of that since January; Census M3 unfilled orders for NAICS 3353 were 3.2 months of shipments in January 2019, peaked at 7.1 months in October 2022, and were 5.7 months in July 2025. The [Monitor's launch note](https://open-factory.com/articles/lead-time-monitor-launch/) explains why we treat the M3 ratio as the free backlog gauge: it moves before OEM guidance does. The filings are free too. Powell Industries' [August 5, 2025 release](https://www.sec.gov/Archives/edgar/data/80420/000008042025000095/ex991-powlq3xfy2025earning.htm) put backlog at $1.4 billion on June 30, 2025, up 7% in a quarter, with a book-to-bill of 1.3x; the [Book-to-Bill Table](https://open-factory.com/models/book-to-bill-table/) lists that row next to Eaton, ABB and the rest, and the [Q2 league table](https://open-factory.com/articles/book-to-bill-league/) is the quarterly read. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) tracks which of those vendors publishes anything more than a backlog figure. Transaction prices are where the paid reports stop and the public record starts. Riverside Public Utilities publishes every bid on its 15 kV metal-clad lineups: [four bids from $1,343,878 to $1,469,875 in 2018](https://riversideca.legistar1.com/riversideca/attachments/28dbba7d-a43b-456d-99cc-3148790fb517.pdf), [three from $2,561,552 to $3,176,381 in 2024](https://riversideca.legistar1.com/riversideca/attachments/37cb493a-5440-49a2-a51f-c84cbe1795ce.pdf), [four from $2,254,388 to $3,205,871 in 2025](https://riversideca.legistar1.com/riversideca/attachments/99a5fd23-cb4f-4c0e-b009-d7390ecf7818.pdf). Austin Energy's purchase orders sit in a [public Socrata table](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX90237) that returns a 15 kV Model 624 pad-mount at $85,933 in January 2020 as JSON. Dover posts [bid tabulations as PDFs](https://www.cityofdover.gov/bid-tabulations). The [Price Book launch](https://open-factory.com/articles/bid-tab-price-book-launch/) describes how we normalize them. The bands in the first chart are the result, and [Quote Check](https://open-factory.com/tools/quote-check/) returns the band for a spec. Lead times are in the same files. Riverside's June 2024 board slides list ["High-cost and long lead-time (Average 70 weeks)"](https://riversideca.legistar1.com/riversideca/attachments/34f6d7f6-a219-4872-95f4-9220c7bb1edf.pdf) as the reason to order switchgear two years ahead; Greenville Utilities' [2019 bid tab for two S&C pad-mounts](https://www.guc.com/sites/default/files/uploads/bid-tab-sheets/19_25_bid_tab.pdf) records 18 weeks delivery at $15,370 a unit. Those two documents, five years apart, are a lead-time series no category report on the list publishes. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) turns the Monitor's current figures into PO dates. ## Refresh Against a Moving Tariff **A report refreshed one to four times a year met four Section 232 changes between February and August 2025.** [Proclamations 10895 and 10896](https://www.federalregister.gov/documents/2025/02/18/2025-02833/adjusting-imports-of-steel-into-the-united-states) (signed February 10, 2025, 90 FR 9807 and 9817) took aluminum from 10% to 25% and ended the steel exemptions from March 12; [Proclamation 10947](https://www.federalregister.gov/documents/2025/06/09/2025-10524/adjusting-imports-of-aluminum-and-steel-into-the-united-states) (June 3, 2025, 90 FR 24199) doubled both to 50% from June 4; [Proclamation 10962](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states) (July 30, 2025, 90 FR 37727) put 50% on semi-finished copper and copper-intensive derivatives from August 1; and the [BIS inclusions notice](https://www.federalregister.gov/documents/2025/08/19/2025-15819/adoption-and-procedures-of-the-section-232-steel-and-aluminum-tariff-inclusions-process) (90 FR 40326) added derivative HTS codes to the 50% duty from August 18. | Instrument | Dates | What changed | Citation | |---|---|---|---| | Proclamations 10895 and 10896 | Signed Feb 10, effective Mar 12, 2025 | Aluminum 10% to 25%; steel exemptions and exclusions ended | 90 FR 9807; 90 FR 9817 | | Proclamation 10947 | Signed Jun 3, effective Jun 4, 2025 | Steel and aluminum 25% to 50% (UK stays 25%) | 90 FR 24199 | | Proclamation 10962 | Signed Jul 30, effective Aug 1, 2025 | 50% on semi-finished copper and derivatives | 90 FR 37727 | | BIS inclusions process | Notice Aug 19, effective Aug 18, 2025 | Derivative HTS codes added to the 50% duty | 90 FR 40326 | *Source: Federal Register, documents 2025-02832, 2025-02833, 2025-10524, 2025-14893, 2025-15819. [CSV](data/05-tariff-changes-2025.csv)* The buyers already price this in themselves. Riverside's May 2025 memo carries a [20% contingency on the switchgear and 25% on the WEG transformer "to account for potential import duties or tariffs that may arise after the quoted price"](https://riversideca.legistar1.com/riversideca/attachments/c543fa97-4254-4b6d-8413-8103f7cc5f04.docx), double the 10% it used in 2024. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) lists each instrument with its citation and a checked-on date, and the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) tracks the metal that the copper proclamation moved. A category report written before June 3 and refreshed once a year describes a duty schedule that no longer exists. ## What It Costs in Hours We priced the reconstruction the way a sourcing manager would: the steps in this article, the time each one takes on a first pass, and the [BLS median wage for buyers and purchasing agents, $37.36 an hour](https://www.onetonline.org/link/summary/13-1023.00). Eight hours in total, half of them transcribing awards, at $37.36 an hour is about $300 against $2,850 for one IBISWorld report or $4,950 for MarketsandMarkets, and the eight hours produce 118 dated unit prices with buyer names and two lead times that neither report contains. The one thing the eight hours do not buy is a PPI escalator that matches your gear. Riverside's own memo for the 2023 award says [staff "used an escalator to adjust the old pricing for inflation, but the actual cost inflation of such specialized equipment has exceeded our estimates"](https://riversideca.legistar1.com/riversideca/attachments/ae7986e2-8ee3-4675-8f9c-b8dbb6c3040f.docx): the engineer's estimate was $2.1 million against a $2.51 million low bid. WPU1175 rose 43% between the two awards; the low bid rose 87%. Dover's stock switch, by contrast, tracked the index within five points, up 53% from November 2019 to September 2024 against 49% for WPU1175. The index works for a commodity stock item and fails for an engineered lineup, and no report on the list says which of the nine switchgear products it is describing when it quotes a CAGR. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) exists to fill that gap with contributed quotes once its cells reach five contributors and eight quotes. ## What to Do Monday Before renewing any category subscription, run the test above on the one category you buy most: open the public page, count the transaction prices and lead times it lists, and compare the count to what your own purchasing files hold. If you are quoting switchgear, pull WPU1175 and the M3 ratio, read the last Powell release, and put your last three awards next to the Price Book's bands before you open a quote. The [Quote Check](https://open-factory.com/tools/quote-check/) tool does the last step in one form. Write the tariff clause against the Federal Register citation, not against a report's "trade" paragraph. Spend the $2,850 only where the recipe below says the public record is thin. Behind the paywall: the free-source recipe for **ten categories**, from switchgear to industrial pumps, with the exact PPI series, Census M3 group, filings, award portals and lead-time sources for each, and the three cases where a paid report is still worth buying, drawn from the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/). --- # Building the Missing Index: Realized Auction Prices for Used CNC Machines URL: https://open-factory.com/articles/used-cnc-auction-index/ Date: 2025-09-25 Machinery / Deep dive 144 hammers from three houses that publish them, a 3.9-year half-life, Haas at 44 to 55% of list under eight years old, the $100 control cohort, $8,900 to $22,400 to land a $36,600 machine, the five checks before you wire money, the lookup table A five-year-old Haas VF-2SS with the Next Generation Control, probing, a 30-station side-mount changer and a 4th-axis drive hammered at [$36,600 on 18 November 2025](https://www.machinesused.com/lots/Haas-VF-2SS-CNC-Vertical-Machining-Center,-12K-Spindle,-30-Station-SMTC,-Probing,-4th-Axis-Brushless-Drive---NEW-2020-47518) after 69 bids, which is 43.6% of the [$83,995 Haas lists it for today](https://machinetoolindex.com/machines/haas-vf-2ss), and the buyer paid 15% on top of that. Nobody publishes numbers like that as a series. The dealers hide their prices, the big industrial auction houses hide their results, and the two government surplus platforms either hide closed prices behind a login or leave them public but unindexed. So we went to the houses that do publish and read every closed CNC lot we could reach: 144 CNC machine tools, 142 of them sold, across 15 Machinesused.com auctions (13 of them with a CNC machine tool), 7 Hansen Auction Group sales and 2 K-BID liquidations, from June 2023 to 8 September 2026, every row with a URL in [data/realized-prices.csv](data/realized-prices.csv). This piece covers what the hammer does with age, Haas against everyone else, the controller cohort that trades at $100, what the three houses charge and how fast they want the machine gone, the bill between the hammer and the first chip, the five checks before wiring money, and behind the wall the model-and-year lookup table and the premium ranking. The 31 Haas lots with a model year fit an exponential with a half-life of 3.9 years (annual retention 0.836), the 82 other builders fit one with a half-life of 4.1 years (0.845), and the Haas curve starts from a higher intercept, so at ten years the fitted Haas hammer is ~$30,000 against ~$12,000 for the field. The spread of the dots around either line is the reason the curve is a guide and not a quote. ## What a Hammer Price Is and Who Publishes One A hammer price is the final accepted bid before the buyer's premium, sales tax, rigging and freight. It is the only used-machinery price that is not an ask. In [What a Haas VF-2 Really Costs in 2026](https://open-factory.com/articles/haas-vf2-true-cost/) we opened 63 used VF-2 listings on 10 September 2026 and found 3 with a price; in [The Used and Refurbished Market Nobody Indexes](https://open-factory.com/articles/used-refurbished-market/) we found the same wall around transformers and gensets, with GovDeals showing current bids on open lots and nothing on closed ones. The [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) exists because public procurement awards are the one place transaction prices for electrical gear appear by law. Machine tools have no equivalent, so this file is the v0 substitute: realized prices from auction houses that leave their closed catalogues on the open web. Three did. [Machinesused.com](https://www.machinesused.com/past-auctions) in Eagan, Minnesota keeps every past auction's catalogue with the final bid, the bid count and the close timestamp on each lot, and its lot pages carry manufacturer, model, year, serial, control and weight; 15 past auctions from 19 August 2025 to 18 August 2026 gave us 122 CNC machine tools. [Hansen Auction Group](https://www.hansenauctiongroup.com/auctions/results) in Wisconsin runs on a bidding platform whose API returns each lot's status, winning bid and bid count, and its 8 September 2026 New Berlin dispersal is the freshest sale in the file. [K-BID](https://www.k-bid.com/terms), a Minnesota network of independent auctioneers, leaves closed item pages public with the high bid, the close time and a "Reserve Not Met" flag where it applies. The other 20 or so venues we opened are in the access log in research/notes.md: GovDeals closed prices behind a login, PublicSurplus closed pages public but not listable, GSA Auctions a JavaScript shell, HiBid's search API returning zero ended lots for any status value, BidSpotter and Proxibid blocking, eBay serving an error page. **The result is 171 rows, of which 144 are CNC machine tools, 142 of those sold, with a model year on 115 and a control named on 89.** The remaining 27 rows are EDMs, desktop and educational CNC trainers, two 3D printers and a router, kept in the file and left out of every statistic below. It is a Minnesota and Wisconsin file: 113 of the 144 machine-tool lots sat in Minnesota and 20 in Wisconsin. The roadmap for this piece promised 1,800 lots. We have 144 with a public URL each, and 144 is the number this article uses. ## The Curve: Half Its Value Every Four Years The scatter above is the whole market in one frame. The median hammer across the 142 sold machine tools is $3,919; 32% went for under $1,000 and 55% for under $5,000; the gross was $1,009,566 before premiums. By type, the 61 vertical machining centers ran a median $5,526, the 44 lathes and turning centers $3,312, the 15 CNC knee and bed mills $3,536, the 18 Swiss and CNC automatics $874, the 3 horizontals $7,355, and the single 5-axis, a [2013 Mazak VCU-400A 5X](https://www.machinesused.com/lots/Mazak-VCU-400A-5X-5-Axis-CNC-Vertical-Machining-Center,-Mazatrol-Matrix-2-CNC,-12K-RPM,-30-ATC---New-2013-48003), $11,150. The median lot drew 40 bids and the busiest drew 202. These are no-reserve liquidations by and large: Machinesused sold 122 of 122 and Hansen 19 of 19. **A used CNC machine tool in this file loses about half its hammer value every 3.9 years if it is a Haas and every 4.1 years if it is not, and past 25 years the builder's name is worth almost nothing.** Both fits are least squares on the log of the hammer against age at sale, and both are in [data/fit.json](data/fit.json) with the points in [data/hammer-vs-age.csv](data/hammer-vs-age.csv). The top of the frame belongs to a [2015 Doosan Mynx 6500/40 at $66,500](https://www.machinesused.com/lots/Doosan-Mynx-6500%2F40-CNC-Vertical-Machining-Center-with-Doosan-Fanuc-i-Series-CNC-Control,--30-ATC-Thru-Spindle-Coolant,-12,000-RPM,-CAT-40,-48757) on 13 January 2026 and a [2018 Haas VF-4SS at $59,000](https://www.machinesused.com/lots/Haas-VF-4SS-5-Axis-CNC-Vertical-Machining-Center,--12K-RPM-Spindle,-30-Station-SMTC,-TSC,-Probing,-Chip-Auger,-4th-and-5th-Axis-Drives-with-Haas-TR-200Y-Trunnion-Table--NEW-2018-47543) on 18 November 2025. The bottom belongs to a [1998 Haas VF-4 at $252](https://www.machinesused.com/lots/Haas-VF-4-CNC-Vertical-Machining-Center,-20-Station-Carousel--Tool-Changer,-7500-RPM-Spindle,-Chip-Auger,-Programmable-Coolant-49640) on 17 March 2026 and a [2001 Roku-Roku GR-658N graphite machining center on a Fanuc 16i at $2](https://www.machinesused.com/lots/Roku-Roku-GR-658N-CNC-Graphite-Machining-Center,-Fanuc-16i-M,-32K-RPM,-HSK-Taper,-Dust-Collector-50349) on 28 April 2026. Both of those were running machines with a listed control. Age did that. The shape matters for anyone modelling a purchase. The [Haas VF-2 TCO](https://open-factory.com/articles/haas-vf2-true-cost/) we published in March carried a $30,000 resale assumption at year ten on a $70,995 machine, 42% of list, anchored on one dealer's sold price. The auction file says a ten-year-old Haas hammers at a median 24% of today's list (n=6 in the 11 to 15 year bucket, min 14%, max 34%), and the buyer then pays a 15% premium, so the asking-to-hammer gap the dealers protect is real and it is about a third. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) will replace both numbers with contributed invoices when it has eight of them in a cell; until then the hammer is the only realized price anyone can point at. ## Haas Against the Field, at List Haas is the one builder that publishes list prices, which is why it sits where it does on the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) and why we can express its hammers as a share of a current price. We read 22 current US list prices from [machinetoolindex.com](https://machinetoolindex.com/machines/haas-vf-2), which republishes Haas's list with a "verified 2026-08-31" stamp: VF-2 $70,995, VF-2SS $83,995, VF-3SS $101,995, VF-4 $92,995, VF-4SS $107,995, ST-20 $77,995, TM-1 $46,995, EC-400 $244,995 and the rest in research/notes.md. Twenty-three of the 32 Haas lots are models Haas still sells. The HL-2, SL-10, SL-20, VF-0, VF-0E and VM-3 are not, and get no percentage. The buckets read 44% at 0 to 5 years (one lot, the 2020 VF-2SS), 46% to 55% with a median of 50% at 6 to 10 years (the 2019 ST-20 at $36,100 and the 2018 VF-4SS at $59,000), 14% to 34% with a median of 24% at 11 to 15 years (n=6), 3% to 19% with a median of 9% at 16 to 20 years (n=8), and 0.3% to 10% with a median of 4.5% past 21 years (n=6). **Under eight years old, a Haas hammers at roughly half of what Haas charges for the same model new, and a buyer who adds the 15% premium and $9,000 to $22,000 of landing costs is paying 65% to 80% of list for a machine with no warranty and no installation.** That is the arithmetic that the dealers' asking prices hide. Machinio's asking band for 2016 to 2022 VF-2s was $41,900 to $62,900 when [the Haas piece read it in September 2026](https://open-factory.com/articles/haas-vf2-true-cost/). The 2020 VF-2SS hammer plus premium comes to $42,090, the floor of that band, for a machine one grade up. A [2014 VF-2 hammered at $23,550](https://www.machinesused.com/lots/Haas-VF-2-CNC-Vertical-Machining-Center,-30%22x16%22x20%22Travels,-CT40,-20-ATC,-Probing,-Thru-Spindle-Coolant,-4th-Axis-Drive---New-2014-47326) on 14 October 2025 with 2,031 hours stated, against the [$22,750 KD Machinery marked as sold on a 2008 VF-2](https://surplusrecord.com/listing/haas-vf-2-30-x-16-y-20-z-10000-rpm-24-automatic-tool-changer-cat-40-haas-cnc-2008-1088411/): a dealer's sold price for a machine six years older than the auction lot, and the two numbers are $800 apart. Below eight years the Haas curve and the field's are far apart, and the reason is supply as much as brand. Of 113 dated machine-tool lots, only 9 were younger than ten years, and 3 of those 9 were Haas. The young machines that do reach auction are the ones a shop bought on a five-year note and lost. The [Machinesused Superior Machine sale](https://www.machinesused.com/auctions/264) that produced the 2020 VF-2SS and the 2018 VF-4SS was a complete-shop auction "by order of secured party", and the [October 2025 Regal Machine sale](https://www.machinesused.com/auctions/261) that produced the 2019 ST-20 at $36,100 and a 2013 VF-2SS at $28,455 was the same kind of event. When [USMTO orders hit $3.44 billion in the first half of 2026](https://open-factory.com/articles/machine-tool-record-fewer-units/) on fewer units, the machines that came off shop floors were the ones that had been paid for. Haas machines 0 to 10 years old ran a median $36,600 (n=3), 11 to 20 years $11,562 (n=18), and 21 years and older $3,450 (n=10); the Japanese and German majors (Mazak, DMG Mori, Okuma, Kitamura, Matsuura, Brother, Miyano, Star, Citizen) ran $18,300 at 11 to 20 years (n=5) and $880 past 21 (n=20); the Korean and Taiwanese builders (Doosan, Hyundai Wia, YCM, Sharp, Acer, Femco, Kia) $6,187 at 11 to 20 (n=18) and $100 past 21 (n=5); the US toolroom and legacy names (Milltronics, Fadal, Hardinge, ProtoTRAK, Fryer, Bridgeport, Tree) $17,050 at 0 to 10 (n=5), $6,706 at 11 to 20 (n=10) and $2,027 past 21 (n=10). **Between eleven and twenty years a Mazak or Mori holds a median $18,300 against $6,187 for the Korean and Taiwanese tier, a 3x gap that vanishes after year twenty when both trade for the price of the rigging.** The named lots make the point. A [2012 Mazak QuickTurn Nexus 200-II](https://www.machinesused.com/lots/Mazak-QuickTurn-Nexus-200-II-2-Axis-CNC-Lathe-Turning-Center,-Tailstock,-3-Jaw-Chuck,-Chip-Con---NEW-2012-47330) took $21,600 and a [2013 Mazak VCN 510C-II](https://www.machinesused.com/lots/Mazak-VCN-510C-II-CNC-4-Axis-Vertical-Machining-Center,-Rotary-Table,-12K-RPM,-Probing-%28New-2013%29-49578) took $28,111 on Mazatrol controls that Mazak has since replaced; a [2010 Hyundai Wia F500 Plus](https://www.machinesused.com/lots/Hyundai-Wia-F500-Plus-CNC-Vertical-Machining-Center,-10K-Spindle,-30-Station-SMTC,-Chip-Conveyor,-High-Pressure-Coolant-System-47573) on a Fanuc took $5,775 and a [2006 Daewoo Doosan 320/40](https://www.machinesused.com/lots/Daewoo-Doosan-320%2F40-CNC-Vertical-Machining-Center-with-2-Station-Pallet-Changer,-12,000-RPM-Spindle,-14-Station-Rotary-Turret,-Chip-Conveyor-47709) took $600. The exceptions run the other way too: the 2015 Doosan Mynx at $66,500 was the most expensive lot in the file, and a 2020 Milltronics VM4020 at [$46,500](https://www.machinesused.com/lots/Milltronics-VM4020-CNC-Vertical-Machining-Center,-40%22x20%22x20%22,-10K-RPM,-Probing---NEW-2020-46972) beat every Haas but the VF-4SS. What the tiers agree on is the floor. Past twenty years, 20 Japanese and German machines ran a median $880 and 5 Korean and Taiwanese machines a median $100. For a buyer, that floor is the interesting number: it says a 1996 Okuma LB15 II on an OSP700 control [changes hands for $100](https://www.machinesused.com/lots/Okuma-LB15-II-CNC-Lathe-Turning-Center,-OSP700L-CNC,-12-Station-Turret,-Tailstock,-8%22-Chuck,-Chip-Conveyor-48011), and the risk is entirely in the control. ## The $100 Control Cohort Every lot page names the control when the auctioneer bothered to, and 89 of the 144 did. The pattern in those 89 is the clearest single finding in the file, and it is not about age. Eleven machines on Yasnac controls, median age 24 years, hammered at a median $100, and 73% of them went for under $1,000; eleven on Fanuc 16, 18 or 21 series controls, median age 25, hammered at a median $356 with 55% under $1,000; two Fadals ran $388; twelve on pre-Smooth Mazatrol, median age 19.5 years, ran $6,301; nine Haas machines on the Classic control, median age 18, ran $7,355; five Milltronics on Centurion 8 and later ran $6,875; two on Mitsubishi M7 or M8 controls ran $11,946; and the three Haas machines on the Next Generation Control, median age 6, ran $36,600. **A working machine with a Yasnac or a 1990s Fanuc control is worth its scrap weight plus the cost of the rigger's time, and the same iron on a control the builder still supports is worth $5,000 to $7,000.** [FANUC America's own product page](https://www.fanucamerica.com/products/cnc/cnc-systems) marks the 0i-MF, 0i-TF, 31i-B, 31i-B5 and 32i-B as "Legacy" and sells the Plus versions as current, and Wikipedia dates the last Series 0 controls to a [production run that ended in 1998](https://en.wikipedia.org/wiki/FANUC); the 16, 18 and 21 series in our lots are a generation older than what FANUC now calls legacy. Yaskawa's Yasnac line is gone from every current machine in the file. The lots say what that means at the hammer: a [2001 Miyano MTV-T310 on Yasnac](https://www.machinesused.com/lots/Miyano-MTV-T310-CNC-Vertical-Machining-Center-46844) for $505, a [1999 Kia Super KiaTurn 15 on Yasnac](https://www.machinesused.com/lots/Kia-Super-KiaTurn-15-CNC-Turning-Center-with-Yasnac-J50L-CNC-Control,-6%22-3-Jaw-Chuck,-Tool-Setter,-12-Tools,-Tailstock,-Chip-Conveyor-and-MORE%21-48379) for $100, a [1992 Matsuura MC-800VF on Yasnac](https://www.machinesused.com/lots/Matsuura-MC-800VF-CNC-Vertical-Machining-Center,-Yasnac-i80-CNC,-CT40,-8K-RPM,-30-ATC-48009) for $100, a [1998 Femco Durga 25E on a Fanuc OT](https://www.machinesused.com/lots/Femco-Durga-25E-CNC-Turning-Center,-Fanuc-OT-CNC-Control,-8%22-3-Jaw-Chuck,-Tool-Setter,-12-Station-Turret,-Tailstock-and-MORE%21-48378) for $100, a [2003 Mazak PFH-4800 horizontal](https://www.machinesused.com/lots/Mazak-PFH-4800-CNC-Horizontal-Machining-Center,-Mazatrol-640M-Control,-%282%29-400mm-Pallets,-CT40,-80-ATC,--.001-Degree-Index,-2-Station-Pallet-Changer---CURRENTLY-NOT-OPERATIONAL---Oil-Leak-under-pallet%29-48012) for $100. Against those, a [2007 Doosan Puma 300MSC on a Fanuc 18i-TB](https://bid.hansenauctiongroup.com/ui/auctions/165047) drew 60 bids and $7,100 at Hansen on 8 September 2026: an 18i, but the i-series, with a live-tool spindle and a chip conveyor, and the description said "machine was tested and all axes appear to function per seller". We cannot separate the controller discount from the age discount inside one brand with this n. Haas moved to the Next Generation Control in the middle of the last decade, so every Haas in the file younger than nine years has it and every Haas older than eleven does not. The 43% to 55% of list on the NGC machines and the 3% to 34% on the Classic machines is one effect measured twice. What the file does show is the floor by control family, and the floor is where a buyer's risk lives: a board for a Yasnac or a Fanuc 16 is a used part from a breaker, not a catalog item, and the hammer already knows it. ## Where the Machines Sell, and How Fast They Want Them Gone The volume in this file is lumpy because liquidations are lumpy. Machinesused runs a multi-shop online auction most months and a complete-shop sale when a secured party calls; Hansen sells a machine shop when an owner retires or a bank repossesses. From August 2025 to September 2026 the three houses closed 126 CNC machine-tool lots, 25 of them in November 2025 and 23 in December, none in February or May 2026, and the gross hammer ran $274,000 in October 2025 and $253,000 in November against $8,000 in September 2026; 18 earlier Hansen lots from June 2023 to August 2024 sit in the CSV and not in the chart. **Two months, October and November 2025, carried $527,000 of the $1,009,566 gross in the file, because that is when the two complete-shop liquidations closed.** For a buyer the implication is scheduling, not price: the good machines come to market in the weeks after a lender moves, and the backlog arithmetic in [The $306k Machine](https://open-factory.com/articles/machine-tool-record-fewer-units/) put a Japanese builder's order-to-ship at about seven months in mid-2026, so the shop that can watch three catalogues in October buys a 2020 machine in November and cuts chips in December. The [EV plant liquidations](https://open-factory.com/articles/ev-plant-machines-second-hand/) we covered in January are the same mechanism at a scale where BidSpotter houses run the sales and the results never reach the open web. Machinesused sold 122 of 122 CNC machine-tool lots at a median 42.5 bids and a median hammer of $5,015 with a 15% premium; Hansen sold 19 of 19 at a median 23 bids and a median $220 with a 10% buyer's fee; K-BID sold 1 of 3, the other two being a [2014 Haas VF-2SS bid to $10,000](https://www.k-bid.com/auction/60151/item/2) and a [2013 Haas ST-20 bid to $7,500](https://www.k-bid.com/auction/60151/item/1) that both closed "Reserve Not Met" on 22 July 2025, under a [15% premium capped at $1,500 per item](https://www.k-bid.com/terms). The terms are the other half of the price. Machinesused's [Superior Machine lot page](https://www.machinesused.com/lots/Haas-VF-2SS-CNC-Vertical-Machining-Center,-12K-Spindle,-30-Station-SMTC,-Probing,-4th-Axis-Brushless-Drive---NEW-2020-47518) says "ALL PAYMENTS DUE BY WEDNESDAY, NOVEMBER, 19, 2025 4PM CT", one day after the 18 November close, and "This item must be removed between Monday, December 1, 2025 and Thursday, December 4, 2025", with an exclusive rigger on site those days and a $350 shipping-bracket fee "automatically added to this invoice"; cards are taken only to $2,500 with a 3% charge, so a $42,090 invoice is a wire. Hansen's New Berlin terms give the buyer [60 days to remove and charge 3.75% on cards](https://www.hansenauctiongroup.com/auctions/detail/bw165047), and every lot description ends "buyer is responsible for rigging and removal of item". GovDeals' Norco transformer lot in our [used-market piece](https://open-factory.com/articles/used-refurbished-market/) gave ten business days at a 15% premium. The removal window is a cost: a buyer without a rigger booked and a truck scheduled before the hammer falls is paying storage or forfeiting the deposit, and the [RFQ template](https://open-factory.com/articles/rfq-capital-equipment/) we published for new equipment has no clause for it because a factory order does not need one. ## From Hammer to First Chip The hammer is the largest line in landing a used machine and not the whole bill. We built the rest from the public rate cards that exist and marked the lines that do not have one. On the 2020 VF-2SS the premium is $3,660 at Hansen's 10% or $5,490 at Machinesused's 15%; the bracket fee is $350; loadout, freight for 150 miles and set-in-place run $2,500 to $7,500 across the three lines; the electrical drop for a machine that wants [220 VAC three-phase at 70 A or 440 VAC at 35 A](https://machinetoolindex.com/machines/haas-vf-2) runs $1,200 to $4,000; distributor startup at [$150 to $225 an hour](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/) for 8 to 16 hours runs $1,200 to $3,600 plus $300 to $1,500 of travel at $1.50 to $2.00 a mile; the total is $8,860 low, $15,740 base, $22,440 high, or 24%, 43% and 61% of the hammer. **A $36,600 hammer is a $45,000 to $59,000 machine on the floor, and the 15% premium is the only line in that bill that a bidder can price to the dollar before bidding.** The rows marked (A) in [data/rigging-cost-table.csv](data/rigging-cost-table.csv) are ours: no rigger or flatbed carrier we could reach publishes a rate card, DAT's flatbed benchmark blocked our fetch, and the auction houses name the exclusive rigger without naming his price. The rows that are not marked (A) come from Hymson's [maintenance benchmark table](https://www.hymsonlaser.com/resources/tables/maintenance-cost-benchmarks/), which gives $150 to $225 an hour for OEM distributor service labour, $250 to $400 for a 5-axis or laser specialist, $1.50 to $2.00 a mile for travel and the line "A simple 1-day repair often costs $2,500+ when travel is added", and from the auction terms themselves. The [Haas TCO piece](https://open-factory.com/articles/haas-vf2-true-cost/) listed $4,000 to $8,000 for freight, rigging and install on a new machine delivered by the distributor. The auction version is higher because the buyer is paying two riggers, a carrier and a startup call that the distributor would have bundled, and because the premium is a cost a factory order does not have. [Quote Check](https://open-factory.com/tools/quote-check/) exists to hold a distributor's bundled quote against these lines. The same arithmetic run in reverse says a distributor's $70,995 VF-2 delivered and installed is competing against a $45,000 to $59,000 five-year-old one, not a $36,600 one. The machine's weight is the other input a bidder needs and the lot page usually gives it. The Superior Machine VF-2SS listed "Main Unit Approx Weight: 9000.00 lbs" and shipping dimensions of 112 by 94 by 98 inches; Hansen's Doosan Puma gave 5,860 kg; a [2015 Fryer CM-15](https://www.machinesused.com/lots/Fryer-CM-15-CNC-Vertical-Machining-Center,-Siemens-828D-CNC-Control,-8-ATC,-CAT-40,-8000-RPM,-NEW-2015-50936) that took $7,725 in August 2026 came with a Siemens 828D and its own weight line. Weight sets the forklift class, the trailer and the permit, and a 9,000 lb machine is a 15,000 lb forklift and a step-deck, not a box truck. ## The Five Checks Before You Wire Money Every dollar in the file was wired within a day or two of a close on an "as is, where is" lot, and the descriptions tell a buyer exactly what was and was not checked. First, the hours. Only 4 of 144 lots stated an hour meter, and the one that stated all three was the [Haas VF-5 at Hansen](https://bid.hansenauctiongroup.com/ui/auctions/165047): "power on time reads 23380:11, cycle start time reads 4821:27:42, feed cutting time reads 3162:51:30". That is a machine that sat powered for 23,380 hours and cut for 3,163 of them, and it hammered at $675 with 63 bids. Ask for the three timers and photograph the screen. A Haas records them and the auctioneer will read them off if asked before the preview closes. Second, the control, by family and not by builder. The chart above says a Yasnac or a Fanuc 16/18/21 is a $100 to $400 machine and a Mazatrol Matrix or a Haas Classic is a $6,000 to $7,000 one at the same age. The lot page names it in the "Machine Control" field or the equipment line, and a buyer who cannot get a board for the control should bid the scrap price and not the iron. Third, the test statement. Hansen writes "machine was tested and all axes appear to function per seller" on the Puma and "attempted to test but machine tripped the breaker" on a lathe in the same sale; Machinesused writes "NOT CURRENTLY OPERATIONAL" in the title of a [1998 Brother TC-229](https://www.machinesused.com/lots/Brother-TC-229-CNC-Tapping-Center---NOT-CURRENTLY-OPERATIONAL-46829) that then took $100. The words are the warranty. There is no other one. Fourth, the paper and the clock. The premium (10% or 15%), the card limit, the payment deadline (one day at Machinesused), the removal window (four days there, 60 at Hansen), the exclusive rigger's dates and the $350 bracket fee are all in the terms before the auction opens, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) is the tool to run them against: a machine bought on 18 November that cannot be moved until 1 December and cannot be started up until an electrician has run a 70 A drop is a January machine. Fifth, the reserve. K-BID shows it and the others do not run one, and the two K-BID Haas lots that closed at $10,000 and $7,500 with "Reserve Not Met" are the best-documented asking prices in the file: a seller who wanted more than 12% of list for a 2014 VF-2SS and did not get it. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) contribution form takes the number a buyer actually paid after a private treaty like that, and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) will list the realized rows as the machine-tool sheet until there is a better source. ## What to Do Monday Pull the three closed catalogues before calling a dealer. [Machinesused's past auctions](https://www.machinesused.com/past-auctions), [Hansen's results](https://www.hansenauctiongroup.com/auctions/results) and any K-BID machine-shop liquidation give a buyer 144 realized prices with year, control and weight, and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) includes them as a file with a URL per row; a buyer who knows a 2013 to 2014 Haas VF-2SS hammers at $17,000 to $28,500 and a 2020 one at $36,600 is not negotiating from Machinio's $41,900 floor. Price the landing before bidding, at $9,000 to $22,000 on a $36,600 machine, and book the rigger for the auction's removal dates before the auction opens, because the rigger is named in the terms and the window is four days. Bid the control, not the badge: nothing on a Yasnac or a Fanuc 16 through 21 is worth more than the rigging, and the file has 22 lots to prove it. Ask for the three Haas timers or the Fanuc power-on counter in writing from the auctioneer during preview. And put every price paid, at auction or by private treaty, into the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/), because the houses that publish are three, the ones that do not are twenty, and the [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) can only index what someone shows it. Behind the paywall: the lookup table of all 32 Haas hammers by model and model year against today's list, with the 2013 VF-2SS at **33.9% of list** and the 1998 VF-4 at 0.3%, the 82 non-Haas machines by make, model and year in the CSV, and the ranking of nine auction platforms by buyer's premium and whether the hammer is public, from their own terms. --- # Introducing the Public Bid-Tab Price Book URL: https://open-factory.com/articles/bid-tab-price-book-launch/ Date: 2025-09-20 Procurement / Model launch 2,712 public award lines, 51 bands at n of five or more, pad-mount $30 to $153 per kVA since 2024, twelve bids on one 2,500 kVA unit, gensets $/kW, chillers $/ton, the five portals, Austin Energy at 61% of rows, what version 1 adds On March 19, 2024, Fayetteville Public Works Commission in North Carolina opened twelve bids for four 2,500 kVA three-phase pad-mount transformers and the unit prices ran from [$64,285 to $219,000](https://www.faypwc.com/wp-content/uploads/2024/04/PWC2324061-Bid-Tab.pdf), a 3.4x spread on one stock number, with quoted lead times from 18 weeks to 64. Every one of those twelve numbers is a public record, and until today none of them was in a table anyone could search. The [Open Factory Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) launches as version 0.1 with 2,712 award lines and 2,567 vendor bids, including the losing ones, transcribed from public utility, municipal, state and federal procurement documents, normalized to $/kVA, $/kW, $/ton and $ per unit, and published as 51 price bands that each carry n, the buyer count, the states and the award-date range. This piece gives away the headline bands, walks through the five portals the rows came from and the rules that turned a PDF line into a band, prints the coverage table and the biases, reads the pad-mount, switchgear, genset and chiller bands, credits every buyer whose tab we used, and says what version 1 adds. Across 297 award lines dated January 2024 to August 2026, the median public price was $115 per kVA for single-phase 10 to 50 kVA units (n=59), $89 for single-phase 75 to 167 kVA (n=90), $153 for three-phase 75 to 167 kVA (n=26), $69 for three-phase 225 to 500 kVA (n=60, 11 buyers in 11 states), $47 for three-phase 750 to 1,000 kVA (n=29) and $30 for three-phase 1,500 to 2,500 kVA (n=27, 6 buyers), with the interquartile range on the 225 to 500 kVA band running $53 to $84. ## What the Price Book Is **Version 0.1 is public record only: 2,712 award lines from 141 buyers in 44 states, award dates April 2016 to September 2026, every row with the URL of the document that states the price.** There is no panel, no contributed quote and no survey behind any number here. A row is one awarded line item (a 500 kVA pad-mount at $26,544) or, where the document gives only a total, one award with the unit price left blank. A bid row is one vendor's price on one line, so the Fayetteville tab above yields one award row and twelve bid rows. Bands are computed from award rows that carry a unit price, a rating and no bundled installation. The unit conventions are the ones a buyer uses to check a quote. Distribution transformers are $/kVA by phase, mount and kVA band. Generator sets are $/kW by stationary or portable and kW band. Chillers are $/ton by cooling type and tonnage. Substation and power transformers are $ per unit by HV voltage class, because only 14 of the 33 power transformer rows state MVA. MV switchgear is $ per lineup for metal-clad and $ per unit for pad-mounted switches. None of the 155 switchgear documents states a section count, so $/section waits for version 1. Each band prints n, buyers, states, min, p25, median, p75, max, the award-date range and the share of rows from the largest single buyer, in two windows: `all` (2015 on) and `recent` (awards dated January 1, 2024 or later). What it is not is a survey of what a datacenter developer pays. The buyers are Austin Energy, Ocala Electric, Greenville Utilities, the Air Force and 137 others who buy to a public spec in small lots. Those prices are still the only dated, linkable transaction prices that exist for this gear, and the [$2,850 category report](https://open-factory.com/articles/2850-report-says-nothing/) has none. ## Five Portals That Publish Prices We tried dozens of procurement sites. Most sit behind a login (BidNet, Bonfire, PlanetBids) or post the solicitation and never the tabulation. Five kinds of source produced 2,554 of the 2,712 rows; state portals and school boards gave the rest. Open-data purchase-order lines are 1,671 rows, Legistar board memoranda and attached bid tabs 379, USAspending federal awards 220, utility websites 177, state portals 129, city websites 107 and BoardDocs school boards 29. The first and largest is the [City of Austin open-data portal](https://data.austintexas.gov/resource/3ebq-e9iz.json), Socrata dataset 3ebq-e9iz, which publishes every commodity purchase-order line with description, quantity, unit price, vendor and date, no login required. Austin Energy's lines from 2019 on are 1,665 rows of the Book. One is a [2,500 kVA feed-through pad-mount at $68,851 from WESCO on July 8, 2026](https://data.austintexas.gov/resource/3ebq-e9iz.json?purchase_order=DOM1100MAX275238), $27.54 per kVA. The second is the Legistar public API. Any city on Granicus Legistar answers `webapi.legistar.com/v1//Matters` with a title filter, and the attachments endpoint returns the bid tab PDF or the staff report with the bidder table. Riverside, Ocala, Idaho Falls, Naperville, Mesa, Silicon Valley Power and a dozen counties came in this way, 28 clients in all. The [Idaho Falls Power tab of September 27, 2022](https://idahofalls.legistar1.com/idahofalls/attachments/ff7016e4-5fdf-4deb-ad09-2d42e6bea688.pdf) has seven vendor columns; Ocala's [quarterly stock-contract price list](https://ocala.legistar1.com/ocala/attachments/8e669c10-fd4f-4308-948e-acc15a0ac974.pdf) moves a 2,500 kVA ERMCO unit from $54,322 in Q2 2022 to $69,647 in Q1 2024. The third is the pair of North Carolina municipal utilities that post full multi-vendor tabs on their own sites. [Greenville Utilities Commission](https://www.guc.com/doing-business-us/past-bids) posts a tab sheet for every past bid, 46 of them from 2019 to 2025 with up to eleven bidders per line and a total-ownership-cost evaluation; its [RFB 25-43 of August 2025](https://www.guc.com/sites/default/files/uploads/bid-tab-sheets/25-43-bid-tab.pdf) awarded 150 kVA pad-mounts at $10,100.69 against eleven bids with a 14-week delivery. [Fayetteville PWC](https://www.faypwc.com/previous-bid-opportunities/) posts its tabs as PDFs and confirms the award in Commission minutes; it gave the twelve-bid 2,500 kVA line and an eight-bidder power transformer tab in June 2025. The fourth is the [City of Dover, Delaware bid tabulations index](https://www.cityofdover.gov/bid-tabulations), 39 PDFs from FY2020 to FY2027, 31 of them tabs for distribution transformers and S&C PME switchgear bought for stock, which we parse by column position. Dover's [July 2025 three-phase tab](https://www.cityofdover.gov/media/Purchasing/Tabulations/FY%2025%20Info/25-0023WH%20Bid%20Tab.pdf) has 30 priced columns on the 1,000 kVA line, from $24,435 to $95,610, awarded to PulseMac at $26,000. The fifth is the [USAspending API](https://api.usaspending.gov/api/v2/search/spending_by_award/), queried by product service code and keyword from FY2022 on; it supplies most genset and chiller rows. The Georgia Procurement Registry, Florida Vendor Bid System, Texas SmartBuy and South Carolina's SCEIS filled in the state rows. What was tried and what sat behind a login is in the model's provenance file. ## From Document to Band Every row goes through six steps, written down on the [model page](https://open-factory.com/models/bid-tab-price-book/) so a reader can rebuild a band. Classification comes from the commodity code and the description: liquid-filled distribution transformers (pole-mount flagged in `subclass`); power transformers, meaning an HV winding of 34.5 kV or more or 2,500 kVA and up at a substation; MV switchgear from 5 to 38 kV, split into metal-clad lineups and pad-mounted switches (PME, Vista, SF6, VFI); LV switchgear; gensets with a stated kW; chillers with stated tons. Dry-type and instrument transformers, UPS, parts, rentals, repairs, rewinds and testing are excluded at normalization. Unit price is taken as stated on PO lines and bid tabs. For USAspending awards it is award amount divided by quantity only when the description states the quantity ("TWO (2)", "QTY 3", "4 EA") or names one unit; otherwise the row keeps its extended price and never enters a band. Installation is flagged `Y` when the award bundles construction, and those rows stay out of every band; `partial` means freight, startup or permit fees sit inside the price, and the row stays in with a note. Sales tax is separated where the memo gives the pre-tax figure and flagged where it does not; California awards are commonly tax-included, Texas PO lines are tax-exempt. Of the 166 excluded rows, 138 are awards with a total and no stated quantity, 10 had a $/kW outside 90 to 3,000, 7 a $/kVA outside 8 to 600, 6 a $/ton outside 150 to 6,000, one a per-unit price outside its class bounds, 3 were dated before 2015 and one carried an option-period date after the build. **The plausibility bounds are deliberately wide, and they caught parts orders and lump-sum obligations, not real outliers.** A [$633,581 Navy task order for four 900 kW generator packages](https://www.usaspending.gov/award/CONT_AWD_N3943019F4141_9700_N3943019D2124_9700) is in the bands at $176 per kW because the quantity is stated; a $566,885 Texas Parks and Wildlife award titled "Generators" with no count is excluded. The bounds are 8 to 600 $/kVA for pad-mounts, 90 to 3,000 $/kW for gensets, 150 to 6,000 $/ton for chillers, $40,000 to $20 million per power transformer, $8,000 to $6 million per MV switchgear unit and $3,000 to $3 million per LV unit. Nothing is dropped silently. The last rule is n. A band is published only when five or more award lines sit in it; below five, `bands.csv` prints `insufficient` and blank statistics, and no article may quote it. Of 80 class-and-spec bands computed across the two windows, 51 pass. Five is a low bar, chosen so a reader sees a thin band and its buyer share rather than nothing. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) cells that merge into version 1 hold to five contributors and eight quotes. ## Coverage by Class | Class | Award lines | With unit price | Bid rows | Buyers | States | Unit | Bands published (all / recent) | Award dates | |---|---|---|---|---|---|---|---|---| | Pad-mount and pole-mount distribution transformers | 2,138 | 2,098 | 2,140 | 26 | 22 | $/kVA by phase, mount, kVA band | 15 / 13 | Jan 2019 to Aug 2026 | | Substation and power transformers | 33 | 33 | 117 | 13 | 9 | $ per unit by HV class | 1 / 1 | Apr 2016 to Sep 2026 | | MV switchgear 5 to 38 kV | 155 | 151 | 110 | 18 | 11 | $ per lineup or per unit | 3 / 3 | Dec 2016 to Aug 2026 | | LV switchgear and switchboards | 8 | 8 | 6 | 6 | 4 | $ per unit | 0 / 0 | Sep 2018 to May 2025 | | Diesel and gas generator sets | 205 | 202 | 113 | 82 | 34 | $/kW by type and kW band | 5 / 1 | Apr 2019 to Aug 2026 | | Chillers | 167 | 162 | 79 | 50 | 30 | $/ton by type and tonnage | 5 / 3 | Jul 2018 to May 2026 | | CNC machining centers | 6 | 6 | 2 | 4 | 2 | $ per unit | 1 / 0 | May 2020 to Nov 2023 | *Source: [Open Factory Public Bid-Tab Price Book 0.1](https://open-factory.com/models/bid-tab-price-book/) (n=2,712, public awards 2016 to 2026). [CSV](data/09-coverage-by-class.csv)* Pad-mounts are 79% of the lines and carry 28 of the 51 published bands. Gensets have the widest buyer base, 82 buyers in 34 states, because every federal base, park and prison buys a standby set. LV switchgear at 8 lines does not band. The six CNC rows are Georgia technical-college awards at a $71,572 median (2020 to 2023) that the [Haas VF-2 cost piece](https://open-factory.com/articles/haas-vf2-true-cost/) uses as rows. ## The Biases, Stated Plainly **Austin Energy is 1,665 of the 2,712 lines and 1,574 of the 2,138 pad-mount lines, so the all-years single-phase bands are largely one utility's price history.** The `top_buyer_share` column exists for that reason: the single-phase 10 to 50 kVA band is 95% Austin in the `all` window (n=680) and 80% in the `recent` window (n=59). The three-phase 225 to 500 kVA band is the healthiest in the Book, 48% top-buyer share across 13 buyers in the `all` window and 40% across 11 buyers since 2024, and it is the one to quote first. Any band above 0.8 is a price history, not a market. The Dover, Greenville, Fayetteville, Idaho Falls, Ocala and Douglas tabs are the cross-check. Public buyers only. No investor-owned utility, developer, EPC or contractor price is in the Book. Municipal utilities buy stock units in lots of two to forty; federal rows are obligations at award, not invoices, with ratings parsed from free text. Small utilities pay more for one or two units than a 200-unit blanket order clears at, and the Book does not adjust for it. Dates spread across an inflation. The `all` window mixes 2019 with 2026 prices in a period when the three-phase 225 to 500 kVA median went from $30.61 to $74.11 per kVA. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists the PPI series for the same years. Use the `recent` window for anything price-sensitive. Power transformers are per unit by HV class, not $/MVA, because 19 of the 33 award memos never state the MVA. The one published band, 34.5 to 99 kV class, has a $1,886,464 median across 14 units from 7 buyers (2018 to 2026) and $2,132,514 across the 6 units since 2024. The 100 kV and 200 kV classes have 4 and 3 rows and do not band, but the rows are public: Ocala's [105/140/175 MVA 230 kV autotransformer at $4,003,500 per unit](https://ocala.legistar1.com/ocala/attachments/d53573e5-ff7c-4f3a-a6dc-0bd089e89006.pdf) (August 2023, two bids), Silicon Valley Power's [336 MVA 115 kV units at $6,346,745 and 560 MVA 230 kV units at $8,098,959](https://santaclara.legistar.com/LegislationDetail.aspx?ID=21747&GUID=8FAA8943-A4EF-4E67-98E2-C5E1CF58F4C1) from HD Hyundai (October 2023, seven proposals), and Rocky Mount's [168 MVA 230 to 66 kV unit at $3,362,000](https://www.rockymountnc.gov/DocumentCenter/View/5867) (July 2025, three bids). The [transformer lead-time piece](https://open-factory.com/articles/transformer-160-weeks/) is why they ordered when they did. ## What the Bands Say: Pad-Mounts The single-phase 10 to 50 kVA median went from $46.72 per kVA in 2021 (n=128) to $99.80 in 2023 (n=67); three-phase 225 to 500 kVA from $33.35 (n=22) to $69.02 (n=27); three-phase 1,500 to 2,500 kVA from $15.61 (n=7) to $30.30 (n=12); and 2026 to date stands at $114.80, $48.19 and $29.68, which is 2.5x, 1.4x and 1.9x the 2021 level. **The step happened in 2022, when every band's median rose 53% to 84% in one year, and the 2,500 kVA class went from $16.82 to $30.30 per kVA between 2019 and 2023.** Put into the [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) calculator, the recent band prices a 2,500 kVA service at about $75,900 with a $69,200 to $86,200 interquartile range (n=27, 2024 to 2026), not a 2021 number escalated by a PPI that fits stock items better than engineered ones. PulseMac (Jiangsu Daelim) bid $64,285 with no lead time stated, Alpha Plotter (Eaglerise) $65,530.50 at 18 to 20 weeks and won all six lines for $767,188.10, five bidders sat between $71,600 and $100,612 at 36 to 54 weeks, JST was $118,000 at 30 weeks, WEG $148,650 at 48 to 50 weeks, and Technology International (Larson) $219,000 at 10 to 15 months. The median bid was $100,241, 53% above the award. Read that tab against the same utility fifteen months earlier. In January 2023 the [2,500 kVA line drew five bids from $102,000 to $259,897](https://www.faypwc.com/wp-content/uploads/2023/02/Bid-Tab-PWC2223024-Single-Phase-and-Three-Phase-Pad-Mount-Distribution-Transformers-2023.pdf), JST won at $102,000 with a 196-day lead, and the [Commission minutes of January 25, 2023](https://www.faypwc.com/wp-content/uploads/2023/02/board-minutes-20230125.pdf) confirm the award, and the tab's own footnote records that UTB Transformers and Flanders Electric, who bid ERMCO units via Maddox at $169,408 and $259,897, "stated they could not honor their bid" after opening. By March 2024 the low bid on the same stock number was $64,285 and the bidder count had gone from five to twelve. That is the 2024 pad-mount market in one public document, and the [switchgear cost piece](https://open-factory.com/articles/switchgear-50-percent/) tells the same story for lineups. The lead-time premium is visible the same way. Ocala Electric's [April 2024 critical purchase](https://ocala.legistar.com/LegislationDetail.aspx?ID=5600&GUID=FD87D20D-FC50-4606-A88B-557C7D018450) of a 2,500 kVA Prolec GE unit from Irby at $163,750 sits in the staff report next to a JST quote at $102,566 with 32 weeks and a Hitachi quote at $159,905 with 156 to 158 weeks; the utility paid $61,184 above the low quote for a unit it could get. A band cannot tell you which of the three you will be offered. The [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) exists because that answer moves the PO date more than the price does. ## MV Switchgear, Gensets, Chillers Pad-mounted 15 to 38 kV switches have a $57,415 median across 122 awards from 9 buyers (2019 to 2026, interquartile $19,575 to $91,314, 73% Austin) and $73,715 across the 24 awards since 2024 (8 buyers, 58% top share). The "other MV" band, mostly breakers and reclosers, is $113,224 across 11 units. Metal-clad lineups are off that scale. Fourteen lineups from 5 buyers in 4 states, 2016 to 2026, have a $1,579,382 median with a $930,308 to $2,137,025 interquartile range. The six since June 2024 have a $2,019,662 median, from Snohomish PUD's [$1,540,000 13.8 kV lineup](https://www.snopud.com/wp-content/uploads/2026/04/20260421-REG-FINAL_P.pdf) of April 2026 to Naperville's [$3,827,000 34.5 kV indoor lineup](https://naperville.legistar.com/LegislationDetail.aspx?ID=12613&GUID=F54CCBA8-A39B-4B5C-A90E-EE7D33725DCF) of September 2024. A lineup is not a unit, as [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/) showed on Riverside's $2.56M lineup; $/section is the fix, and it needs section counts no tab states. Stationary sets ran $840 per kW at 20 to 249 kW (n=58, 21 buyers, interquartile $516 to $1,142) and $760 in the 30 awards since 2024, $462 at 250 to 499 kW (n=18) and $335 at 500 to 999 kW (n=11, $270 to $488), while portable and trailer sets were $1,025 per kW at 20 to 249 kW (n=13, 12 buyers) and $528 at 500 to 999 kW (n=10). The recent window is thin above 250 kW: 3 stationary awards at 250 to 499 kW and 3 at 500 to 999 kW since 2024, both `insufficient`, and nothing above 1 MW bands at all, which is the market [the genset allocation piece](https://open-factory.com/articles/gensets-sold-out-2028/) covers. The [Genset TCO](https://open-factory.com/tools/genset-tco/) calculator's capital-cost line should start from the 500 to 999 kW band, n=11, rather than from a list price. Air-cooled chillers were $1,385 per ton at 10 to 99 tons (n=16, 5 buyers) and $1,369 at 100 to 299 tons (n=7). Where the description does not state the type, the medians were $1,423 at 10 to 99 tons (n=40), $1,046 at 100 to 299 tons (n=27) and $1,150 at 300 to 999 tons (n=6), and the two recent-window bands jumped to $2,308 (n=12) and $1,876 (n=13). Of the 102 chiller rows in the bands, 77 are Department of Defense, mostly base replacements of 20 to 250 ton units bought one at a time with delivery, which is why the "type unstated" bands are larger than the typed ones and the recent medians run 62 to 79% above the all-years figure. No water-cooled band passes n of five. The [chiller lead-time piece](https://open-factory.com/articles/chiller-60-week-problem/) and the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) cover the 500 to 3,000 ton machines a datacenter buys, and none of those is in a public tab we have found. ## Who Is Credited Every buyer whose document we transcribed is named in its row. The ones whose multi-vendor bid tabs make the bands worth reading are Greenville Utilities Commission (150 lines, 630 bids), Fayetteville Public Works Commission (27 lines), City of Dover Electric Department (75 lines, 31 tabs), City of Ocala and Ocala Electric Utility (184 lines), Idaho Falls Power (89 lines), Riverside Public Utilities (13 lines, four metal-clad procurements), City of Douglas Utilities in Georgia (13 lines, tabs of 12 to 15 bidders), Tacoma Power (4 preliminary results), City of High Point Electric, City of Rocky Mount Public Utilities, Wake County, Mecklenburg County, Pinellas County, Broward County, Gwinnett County, DuPage County, Dakota County, Fayette County, San Mateo County, the Port of Long Beach, Columbus Division of Water, the Florida Departments of Transportation, Health, Corrections and Agriculture, the Florida Turnpike Enterprise, the Florida Fish and Wildlife Conservation Commission, the Georgia State Financing and Investment Commission, Tallahassee State College, and Georgia Northwestern, Gwinnett and Southern Crescent technical colleges. Award memoranda and minutes without a tab came from Naperville, Mesa, Silicon Valley Power, Los Alamos, Clark Public Utilities, Snohomish County PUD, Albany Utilities, the Texas Department of Criminal Justice and nine other Texas agencies, BART, the Metropolitan Water Reclamation District of Greater Chicago and seven school boards; purchase-order lines from Austin and Los Angeles; federal awards from nine departments and GSA through USAspending. All 141 buyers and their document types are in [buyers.csv](data/10-buyers.csv). ## What Version 1 Adds Three things, in order of how much they move the bands. First, contributed quotes. The [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) takes private quotes from contractors, distributors and owners under the cell-size rules (five contributors, eight quotes, no contributor above 40%, every quote three months old or older) and merges the cells into the same bands, labelled as contributed, so the pad-mount bands stop being 61% one Texas utility and LV switchgear and CNC get their first bands. No individual quote is ever shown. Second, more portals. Columbus, Fort Collins, Colorado Springs Utilities and the Ohio and Michigan state portals hold tabs reachable only through a browser, and the Fayetteville CY2025 stock-contract tab, 12 bidders across 50 lines, is downloaded and not yet transcribed because the award is not in the minutes we could find. Every portal that opens adds a buyer, and the buyer count is what fixes the Austin share. Third, MVA capture for power transformers and section counts for switchgear, so the power transformer class moves from $ per unit by HV class to $/MVA and the metal-clad class from $ per lineup to $/section. Both need the specification, not the award memo; version 1 reads the rating schedule from the IFB where the portal posts it. ## What to Do Monday Pull `bands.csv`, find the band for the unit you are about to buy, and read three columns before the median: n, `top_buyer_share` and `date_max`. If n is under 20 or the share is above 0.8, open the awards behind it and read the buyer names. The [Quote Check](https://open-factory.com/tools/quote-check/) tool is built to do this for a spec, returning the band, the interquartile range and the nearest public awards with their links. If you are buying a 2,500 kVA pad-mount, your quote should sit near $30 per kVA on the recent window (n=27), and ask for the lead time on the same line as the price, because the Fayetteville tab shows a 3.4x price spread and a 3.5x lead-time spread on one unit on one day. If you are a public utility, post your tabulation as a PDF with unit prices and bidder names, the way Greenville, Fayetteville and Dover do. The bands, the band file, the eight chart files and the coverage table are free. Behind the short paywall are the two row-level files: **2,712 award lines and 2,567 vendor bids** with the document URL on every row, the files the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) page describes. --- # Cummins and Caterpillar Are Sold Out Through 2028 URL: https://open-factory.com/articles/gensets-sold-out-2028/ Date: 2025-09-15 Electrical / Deep dive 62 to 110 week lead times, Caterpillar $72.1B backlog, Generac $1.6B backlog, used Tier 2 iron at 18 to 46 cents, INNIO and Wartsila recips, Tier 2 vs Tier 4 routes, Genset TCO Cummins is taking generator orders "well into 2028" and Caterpillar closed June 2026 with $72.1 billion of firm backlog, $29.2 billion of it not shipping inside twelve months. Those two sentences come from [Cummins' fourth-quarter call](https://www.fool.com/earnings/call-transcripts/2026/02/05/cummins-cmi-q4-2025-earnings-call-transcript/) on February 5, 2026 and [Caterpillar's 10-Q](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm) filed August 5, 2026, and between them they set the delivery date for every 2 MW diesel genset a factory or datacenter tries to buy this year. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes published lead times of 50 to 66 weeks for a 1,500 to 2,000 kW diesel and 90 to 110 weeks above 3,500 kW, against 16 to 24 weeks before 2020. Below: what the three listed OEMs told the SEC, the lead-time table by kW class, what used iron actually costs against a new Tier 4 unit, the emergency-engine loophole that makes used iron legal, who still has open engine capacity, and the permitting route by state. Caterpillar's firm backlog sat between $27.5 billion and $28.7 billion for four straight quarters through September 2024, then added $2.4 billion in Q3 2025, $11.4 billion in Q4 2025, $11.5 billion in Q1 2026 and $9.4 billion in Q2 2026, while the slice not expected to ship within twelve months went from $6.0 billion to $29.2 billion. ## What Cummins Told Investors in February Cummins does not publish a backlog number. Asked to size it on the February 5, 2026 call, CEO Jennifer Rumsey said "we do not quantify the size of our back order" and then gave the only two datapoints that matter for a buyer: "we had record order intake in Q4 for power generation. We are taking orders now well into 2028." Five days later the [2025 10-K](https://www.sec.gov/Archives/edgar/data/26172/000002617226000009/cmi-20251231.htm) put it in filing language: "We also have strong demand for our data center products extending out six to eight quarters." Six to eight quarters from February 2026 is 78 to 104 weeks, which is the row the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes for the QSK95 class. **Cummins doubled its 95-liter capacity in 2025 and still sold the whole expansion through 2028.** Rumsey on the same call: "we completed the doubling of our capacity of the 95-liter engine and genset that we supply, which is very popular in the data center market." Data center revenue across the company was about $3.5 billion in 2025 per the call summary, and the guide for 2026 power generation revenue is up 10% to 20% "driven by continued high demand in the data center market and the successful execution of our capacity expansion, which was completed in 2025." The [Power Systems segment](https://www.sec.gov/Archives/edgar/data/26172/000002617226000026/cmi2026q28-kex99.htm) sold $2.3 billion in Q2 2026, up 19%, at a 24.5% EBITDA margin, and Q1 2026 ran a 29.5% margin on $2.0 billion. Those margins are what a sold-out factory looks like in a segment table. Cummins' power generation product line sold $770 million in Q1 2023, $1,090 million in Q4 2024 and $1,536 million in Q2 2026, a doubling in fourteen quarters, and the segment's [own tables](https://www.sec.gov/Archives/edgar/data/26172/000002617225000003/cmi2024q48-kex99.htm) show why the money grew faster than the metal: high-horsepower power generation unit shipments went from 12,300 in 2023 to 12,800 in 2024, up 4%, while power generation dollars rose 19%. The mix shifted to bigger engines and higher prices, and Cummins stopped publishing the unit table in 2025. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) marks that: no backlog in dollars, no lead times, and a unit series that ended the year the market tightened. ## Caterpillar Carries a Year of Sales in Backlog Caterpillar files the number Cummins withholds. The Order Backlog paragraph in each 10-Q reads the same way every quarter, and the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) simply lines them up: $27.9 billion at March 2024, $30.0 billion at December 2024, $39.8 billion at September 2025, $51.2 billion at December 2025, $62.7 billion at March 2026, $72.1 billion at June 2026. Every filing since Q3 2024 names Energy & Transportation, renamed Power & Energy in 2025, as the segment with "the largest increase," and the [Q1 2026 10-Q](https://www.sec.gov/Archives/edgar/data/18230/000001823026000021/cat-20260331.htm) adds the sentence a genset buyer should read twice: "The backlog for large reciprocating engines and turbine products continues to grow within Power & Energy." Firm backlog was 0.42 times trailing-twelve-month sales in Q1 2024 and 0.96 times in Q2 2026, so on a company-wide basis Caterpillar is booked for about twelve months of production, and since the growth is concentrated in one segment the recip engine lines are booked for far longer than that. **Caterpillar's large reciprocating engine backlog is 3.5 times what it was in January 2024, and the company is tripling that capacity.** CEO Joseph Creed said both on the April 30, 2026 call, as [reported by Manufacturing Dive](https://www.manufacturingdive.com/news/caterpillar-triple-power-generation-capacity-raises-2030-targets-q1-2026-earnings/): the large recip backlog "has grown more than 3.5 times since January 2024," Caterpillar "will triple its large reciprocating engine capacity from 2024 levels," and the ProPWR framework for up to 2.1 GW of generation assets over five years was "Caterpillar's sixth agreement with a customer seeking at least 1 gigawatt." The [Slot Economy](https://open-factory.com/articles/slot-economy/) piece covers what a 1 GW frame agreement does to the queue behind it; for a buyer of eight 2.5 MW units, it means six customers with 400 to 800 of them each are ahead in line. Power generation sales were $1,284 million in Q1 2023, $2,242 million in Q4 2024 and $3,238 million in Q4 2025, and the [10-K](https://www.sec.gov/Archives/edgar/data/18230/000001823026000008/cat-20251231.htm) attributes the full-year increase from $7,756 million to $10,275 million to "large reciprocating engines, primarily data center applications," the same clause that appears in the [Q2 2026 release](https://www.sec.gov/Archives/edgar/data/18230/000001823026000040/ex991toformcat2q2026earnin.htm) with turbines added. ## Generac Went From Zero to $1.6 Billion in Twelve Months Generac is the control case for what an OEM with open capacity does in this market. It entered large-megawatt datacenter gensets in Q2 2025. The [July 30, 2025 release](https://www.sec.gov/Archives/edgar/data/1474735/000143774925023858/ex_845273.htm) says the company "began building backlog for our new high-output diesel generator product offering." By [October 29, 2025](https://www.sec.gov/Archives/edgar/data/1474735/000143774925032069/ex_878327.htm) that backlog had "doubled over the last 90 days" with no dollar figure attached. On the [February 11, 2026 call](https://www.fool.com/earnings/call-transcripts/2026/02/11/generac-gnrc-q4-2025-earnings-call-transcript/) it was "$400 million," most of it shipping in 2026. On [April 29, 2026](https://www.fool.com/earnings/call-transcripts/2026/04/29/generac-gnrc-q1-2026-earnings-call-transcript/) it was "more than $700 million," excluding a non-binding notice to proceed "for approximately $600 million in 2027 deliveries with a certain hyperscale customer." On [July 29, 2026](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm): "approximately $1.6 billion as of today, which does not include any committed volumes from the second hyperscale customer," after "approximately $1 billion in additional orders" in one quarter. Generac's disclosed datacenter backlog went $400 million, $700 million, $1,600 million across the three 2026 updates, and its C&I segment sales went from $399 million in Q1 2025 to $556 million in Q2 2026, up 29% year over year. **Generac says its domestic capacity for large-megawatt gensets will pass $1 billion a year by Q4 2026, and it has already booked 1.6 years of it.** The April call also told buyers where the constraint sits: "Even if we can get great lead times on the unpackaged product, it doesn't help us if the packaging phase is constrained," which is why Generac bought Enercon in East Peoria and a second plant in Belvidere, Illinois "significantly expanding capacity for large megawatt generator packaging." Engines are one queue; the enclosure, base tank, switchgear and controls are a second queue, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) tool treats them as separate rows for exactly that reason. Generac's two hyperscaler frame agreements, one signed in June and one on June 24, 2026 with "final product specific terms for 2027 and 2028 volumes" still in negotiation, mean the third entrant's 2027 and 2028 slots are being allocated before most colocation and industrial buyers have asked for a quote. ## Published Lead Times by kW Class Nobody publishes an OEM lead time for a 2 MW genset. The two public series the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) includes are a general contractor's slot-reservation data and a dealer's stock-and-order data, and they agree with each other and with the Cummins filing. [Terrapin Construction Group](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026), June 10, 2026, "from active 2026 manufacturer slot reservations": diesel 500 to 1,000 kW at 32 to 48 weeks, 1,500 to 2,000 kW at 50 to 66, 2,500 to 3,500 kW at 62 to 82, 3,500 to 4,500 kW at 90 to 110, natural gas 1,000 to 3,000 kW at 54 to 78, ATS 24 to 40. [Central States Diesel Generators](https://csdieselgenerators.com/the-ai-data-center-boom-and-generator-shortage-lead-times-prices/), August 11, 2026: 25 to 400 kW at 12 to 26 weeks against 4 to 8 before 2020, 500 to 1,000 kW at 26 to 52 against 10 to 16, 1,250 to 3,250 kW at "52+" against 16 to 24, and a tested used unit from stock at 1 to 4 weeks. The 1,500 to 2,000 kW class that hyperscalers standardize on sits at 50 to 66 weeks, the 2,500 to 3,500 kW class at 62 to 82, and the largest units at 90 to 110, so a PO placed this month for a 3 MW unit lands between Q4 2027 and Q3 2028. | Class | Fuel | 2026 lead time | Pre-2020 | Published by | Date | |---|---|---|---|---|---| | 25 to 400 kW standby | Diesel | 12 to 26 wk | 4 to 8 wk | Central States Diesel (dealer) | Aug 2026 | | 500 to 1,000 kW | Diesel | 32 to 48 wk | n/a | Terrapin CG (GC slot reservations) | Jun 2026 | | 500 to 1,000 kW standby | Diesel | 26 to 52 wk | 10 to 16 wk | Central States Diesel (dealer) | Aug 2026 | | 1,500 to 2,000 kW | Diesel | 50 to 66 wk | n/a | Terrapin CG | Jun 2026 | | 1,250 to 3,250 kW standby | Diesel | 52+ wk | 16 to 24 wk | Central States Diesel (dealer) | Aug 2026 | | 2,500 to 3,500 kW | Diesel | 62 to 82 wk | n/a | Terrapin CG | Jun 2026 | | 3,500 to 4,500 kW | Diesel | 90 to 110 wk | n/a | Terrapin CG | Jun 2026 | | 1,000 to 3,000 kW | Natural gas | 54 to 78 wk | n/a | Terrapin CG | Jun 2026 | | High-horsepower DC gensets (QSK95 class) | Diesel | 78 to 104 wk (six to eight quarters) | n/a | Cummins 10-K FY2025 | Feb 2026 | | High-horsepower DC gensets (QSK95 class) | Diesel | orders taken well into 2028 | n/a | Cummins Q4 2025 call | Feb 2026 | | Automatic transfer switch, large | n/a | 20 to 40 wk | 6 to 12 wk | Terrapin CG; Central States Diesel | Jun to Aug 2026 | | Tested used or surplus, dealer stock | Diesel or gas | 1 to 4 wk | 1 to 4 wk | Central States Diesel (dealer) | Aug 2026 | Twelve rows, three publishers, and the pre-2020 column shows the multiple: 3x at the small end, 2.5 to 3x at 500 to 1,000 kW, and a floor of 52 weeks against a 16 to 24 week norm at 1,250 kW and above. Two things in this table matter more than the numbers. First, the contractor's ranges are tighter than the dealer's because Terrapin is quoting reserved slots, meaning the buyer paid a 10 to 30% deposit to hold a production date, which is the mechanism the [Slot Economy](https://open-factory.com/articles/slot-economy/) piece priced at $60 million for a hyperscale campus. Second, the dealer's floor for 1,250 kW and up is open-ended at "52+" because dealers are quoting allocation, not a schedule: Central States Diesel describes "allocation language that didn't exist in 2019," and that language is how a 52-week quote becomes 80 weeks without anyone breaching a contract. The [RFQ for capital equipment](https://open-factory.com/articles/rfq-capital-equipment/) template handles this with a firm ship date plus [liquidated damages at 0.5% a week](https://open-factory.com/articles/liquidated-damages/), and in this market a vendor who accepts LDs on a genset is telling you it has a real slot. ## Order the Transformer the Same Month **The genset is no longer the longest line on the electrical schedule, and that is the trap: a 66-week genset arriving before a 95-week substation transformer is a $1.4 million asset sitting on a pad with nothing to feed.** Terrapin's same June 2026 sheet puts a 5 to 25 MVA substation transformer at 65 to 95 weeks, 15 kV metal-clad switchgear at 60 to 80 weeks and a large ATS at 24 to 40, so the genset order, the [transformer](https://open-factory.com/articles/transformer-160-weeks/) order and the paralleling switchgear order have to be placed in the same month for the plant to energize in the same quarter. The buyers who are getting 2027 dates are ordering all three at land close and reserving the slot with the deposit. The buyers who wait for construction documents are ordering a genset for a building whose transformer arrives six months later. Natural gas is the one class where the queue is shorter than diesel at the same kW, 54 to 78 weeks against 62 to 82 for 2,500 to 3,500 kW diesel, and the [Recips vs Turbines vs Fuel Cells](https://open-factory.com/articles/turbines-recips-fuel-cells/) TCO shows why: gas prime-power sets are sold by engine houses whose factories were built for gas compression and CHP, not by the two diesel OEMs whose high-horsepower lines the hyperscalers have booked. ## The Used Market Is Pricing Time, Not Iron The dealer consensus, in Central States Diesel's words, is that "low-hour large gensets that once sold at 40 to 50% of new now command 60 to 75% when new delivery is a year away." That is true on the dealer's own terms and misleading for a datacenter or plant buyer, so we pulled the dated listings and checked. Four listings from one dealer's yard in Waukesha, Wisconsin, between June 23 and August 11, 2026, and the price bands from [Power Generation Enterprises' May 2, 2026 guide](https://powergenenterprises.com/industrial-generator-pricing-guide-2026-what-youll-actually-pay/) are in [used-vs-new-prices.csv](data/used-vs-new-prices.csv); asking prices, n=1 per listing, FOB dealer. | Unit | kW | Year / hours | Tier | Asking price | New reference | Ratio | Listed | |---|---|---|---|---|---|---|---| | [Used Cummins DQKC (QSK60), open skid](https://csdieselgenerators.com/megawatt-might-used-cummins-dqkc-2000-kw-diesel-generator-with-qsk60-engine-csdg-5092/) | 2,000 | 2002 / 406 h | not stated | $249,900 | New 2,000 kW Tier 4: over $1.4M | 0.18 | Aug 5 2026 | | [Used Kohler 2000REOZMD (Mitsubishi S16R)](https://csdieselgenerators.com/a-century-of-trust-used-kohler-2000reozmd-2000-kw-diesel-generator-csdg-4884/) | 2,000 | 2018 / 223 h | Tier 2 | $449,900 | New 2,000 kW Tier 4: over $1.4M | 0.32 | Aug 11 2026 | | Used 2,000 kW band (3516B/C, QSK60) | 2,000 | various | mostly Tier 2 | $280k to $650k | New 2,000 kW Tier 4: over $1.4M | 0.20 to 0.46 | May 2 2026 | | Used 1,000 kW band (3508B, QST30) | 1,000 | various | mixed | $90k to $250k | New 1,000 kW Tier 4 all-in: ~$810k | 0.11 to 0.31 | May 2 2026 | | [New Cummins DQFAD (QST30), emergency-only](https://csdieselgenerators.com/new-cummins-dqfad-qst30-g5-nr2-diesel-generator--epa-tier-2-5036.html) | 1,000 | 2024 / 0 h | Tier 2 | $269,900 | New 1,000 kW Tier 4 all-in: ~$810k | 0.33 | Jun 23 2026 | | [Used Cummins DQFAA (QST30)](https://csdieselgenerators.com/low-hour-heavyweight-used-cummins-dqfaa-750-kw-diesel-generator-with-qst30-g5-engine-csdg-4632/) | 750 | 2007 / 792 h | not stated | $135,900 | Same dealer's new Tier 2 at $270/kW | 0.67 per kW | Jul 20 2026 | A 2018 Kohler 2,000 kW Tier 2 unit with 223 hours asks $449,900, or $225/kW, against "north of $1.4M" for a new 2,000 kW CAT 3516 with Tier 4 aftertreatment, a ratio of 0.32; the 2002 Cummins QSK60 at $249,900 is 0.18; and the only row that reaches the dealers' 60 to 75% is a used 750 kW unit at $181/kW against the same dealer's new Tier 2 1,000 kW at $270/kW. **Used gensets trade at 60 to 75% of new only when the new unit is also Tier 2, which is a unit you can only run 100 hours a year.** Against the Tier 4 unit a non-emergency application must buy, used iron is 18 to 46 cents on the dollar, and the gap is the aftertreatment: Power Generation Enterprises prices a new 1,000 kW Tier 4 Final base unit at $580,000 to $720,000 "plus $130,000 to $185,000 for the aftertreatment package," puts a 2026 CAT 3512B HD Tier 4 Final at $850,000 to $1,050,000, and says a used Tier 4 Final unit "commands $35,000 to $80,000 more than the same kW Tier 2 unit, because the buyer doesn't have to navigate emissions paperwork." The 2024-model-year Cummins DQFAD in the table is the clearest case: a brand-new 1,000 kW unit at $269,900 is a third of a new Tier 4 unit because it is certified to Tier 2 and can only be sold for emergency duty. The [Used and Refurbished Market](https://open-factory.com/articles/used-refurbished-market/) piece covers what "load-bank tested, documented hours" is worth on these units. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool lists the Tier 4 aftertreatment line and the run-hour cap as separate inputs so the two routes can be compared on ten-year cost rather than sticker. ## Tier 2 Is the Loophole That Makes Used Iron Legal The reason a 2018 Tier 2 engine is still a legal datacenter generator is written into [40 CFR 60.4202](https://www.ecfr.gov/current/title-40/section-60.4202). Manufacturers must certify emergency stationary CI engines of 37 kW and up to "the Tier 2 or Tier 3 emission standards for new nonroad CI engines" from model year 2007, and engines above 3,000 HP to "the Tier 2 emission standards" from model year 2011. Non-emergency engines are different: [60.4201](https://www.ecfr.gov/current/title-40/section-60.4201) sends them to the current 40 CFR part 1039 standards for their model year, which for a new engine means Tier 4 Final, and [60.4204](https://www.ecfr.gov/current/title-40/section-60.4204) puts the same duty on the owner. A datacenter genset is an emergency engine because its owner promises it will behave like one. **The promise is 100 hours a year, and under [60.4211(f)](https://www.ecfr.gov/current/title-40/section-60.4211) only 50 of them can be anything other than testing.** The text: emergency engines may run "for maintenance checks and readiness testing" for "a maximum of 100 hours per calendar year," of which "up to 50 hours per calendar year in non-emergency situations" is allowed, and "the 50 hours per calendar year for non-emergency situations cannot be used for peak shaving or non-emergency demand response, or to generate income for a facility to an electric grid," unless the engine is dispatched by the balancing authority to avert "voltage collapse or line overloads." There is "no time limit on the use of emergency stationary ICE in emergency situations." An owner who runs a Tier 2 unit outside those limits "must meet all requirements for non-emergency engines," which is Tier 4, which is the $1.4 million unit with the 90-week lead time. Virginia had permitted about 10,500 generator units and 27 GW of diesel capacity at datacenters by the end of 2025, "over 70%" of it Tier 2 per the [Better Data Center Project report as covered by Latitude Media](https://www.latitudemedia.com/news/the-data-center-boom-is-a-diesel-generator-boom), which works out to roughly 19 GW of Tier 2 iron against 8 GW of everything else, while "over half of the newly permitted capacity is Tier 4." The national numbers in the same report explain the backlog charts above: US datacenter diesel generator capacity went from about 20 GW in 2018 to about 55 GW in 2024, and one Amazon site in Manassas runs 93 generators of 2.5 MW each. Texas is now doing the same at higher speed and under minor permits. A Floodlight analysis [reported by Latitude Media on July 13, 2026](https://www.latitudemedia.com/news/ai-data-centers-are-getting-built-thanks-to-a-permit-common-for-dry-cleaners) counted "at least 38 data centers across Texas" that "received minor permits to operate on-site power sources" since 2024, covering "more than 2,100 backup diesel generators," with one Vantage site permitted at "99.8 tons per year" of NOx "barely below the area's 100 ton-per-year threshold" that would trigger public review. The emergency route is not a Virginia habit. It is how 55 GW of engines got permitted in six years, and it is why the [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/) now includes generator permitting alongside interconnection rules for every state where a 100 MW load is being built. ## The Rental Fleet Is the Only Thing Shipping This Quarter Aggreko filed to list on the NYSE on August 24, 2026, and its [Form F-1](https://www.sec.gov/Archives/edgar/data/2134090/000110465926100313/tmq-20260824xf1.htm) is the first audited look at what the bridge-power market is worth. The fleet is "approximately 120,000 assets" and "17 GW, including 6.3 GW of diesel/HVO power generation, 2.1 GW of gas power generation," with "average physical utilization for the year ended January 3, 2026" of 54%. Data center net revenue went from $96 million in 2023 to $196 million in 2024 to $391 million in fiscal 2025, and $362 million in the first half of 2026 alone against $160 million a year earlier, on a total of $1,918 million. The filing names "two large (135 MW and 86 MW), multi-year contracts recently won with a blue-chip data center developer in the United States" and describes a bridging job of "48 times 1,200 kW (or equivalent) Tier 4 Final generators totaling 48 MW of continuous operational bridging power." Data centers were 3.8% of Aggreko's revenue in 2023, 11.4% in fiscal 2025 and 18.9% in the first half of 2026, and the sector added $202 million of revenue growth in the latest year, more than utilities, building services and events combined. **A 6.3 GW diesel rental fleet at 54% utilization is about 2.9 GW of idle Tier 4 Final engines, which is more than the entire US behind-the-meter fleet actually running.** The [Behind the Meter](https://open-factory.com/articles/behind-meter-90gw/) forecast counted about 2 GW operating against 90 GW announced. Rental is not cheap; Central States Diesel's advice is to "budget it as insurance, not as the plan," and the 48 MW Tier 4 Final bridging job in the F-1 was run for a hyperscaler waiting on grid power, not on gensets. But the arithmetic for a plant that needs 4 MW of standby in 2027 is simple: a 24-month rental of Tier 4 units while a new-build slot matures, priced through the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool against a used Tier 2 purchase, is the only path that does not depend on an OEM quoting a date. Generac's April call adds a second reason rental fleets have iron: "many of our rental customers have begun to invest in new equipment as part of a refleeting cycle," which puts the rental houses in the same OEM queue as everyone else, buying now what they will rent to you in 2027. The gas side of the rental and bridge market is where the OEMs themselves are now selling prime power, not standby. Cummins' [Q2 2026 release](https://www.sec.gov/Archives/edgar/data/26172/000002617226000026/cmi2026q28-kex99.htm) lists a June agreement with Circe Energy "to provide a series of natural gas generator sets to support a scalable, behind-the-meter, prime power microgrid solution" for an HPC datacenter in Texas, and describes "natural gas-fueled generator sets and integrated microgrid controls designed to address power grid constraints." That is a diesel standby company selling gas prime power into the same Texas minor-permit route described above, and it is the clearest sign that the 100-hour emergency engine and the 8,000-hour prime mover are converging on the same order book. A plant buyer competing for a 2 MW slot in 2027 is now behind hyperscale standby fleets, behind-the-meter gas campuses and rental refleeting, in that order. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lists the carrying cost of each month of that delay per MW. ## What to Do Monday The Cummins and Caterpillar queue is not going to shorten before 2028, because both companies have told the SEC as much: six to eight quarters of demand on the 95-liter line after a capacity doubling, and $29.2 billion of Caterpillar backlog that will not ship within twelve months. The buyer's job is to stop waiting for a quote and pick a route. For emergency standby under 100 hours a year, buy the tested used Tier 2 unit this month at 18 to 46 cents on the Tier 4 dollar and put the savings into the ATS and paralleling gear that also runs 20 to 40 weeks. For anything that will run more than 50 non-emergency hours, order the Tier 4 unit now with a firm date and LDs, and bridge with rental. Either way, run the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) with the genset row at 62 to 82 weeks and the ATS row at 24 to 40, and put the [Quote Check](https://open-factory.com/tools/quote-check/) band next to whatever the dealer says. A quote inside 40 weeks for anything above 1,500 kW is either a cancelled order, which you want, or a slot the dealer does not hold, which you do not. Behind the paywall: the ten suppliers with open or committed engine and packaging capacity, dated, including the **1.1 GW** INNIO order and the 1.25 GW Rehlko allocation that show where prime-power gas slots still exist, and the Tier 2 versus Tier 4 permitting route in Virginia, Texas and Georgia, with the hour limits and thresholds on one page. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) lists the genset rows monthly. The [Genset TCO](https://open-factory.com/tools/genset-tco/) tool prices the two routes side by side. --- # The 171-Day Capex Lead Time: ISM's Most Ignored Table URL: https://open-factory.com/articles/171-day-capex-lead-time/ Date: 2025-09-13 Procurement / Debrief Capex 171 days, production materials 84, MRO 48, electrical components in short supply 14 months, aluminum up 33 months, how to read it before the PPI moves The average manufacturing buyer now commits to capital equipment [171 days ahead of need](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), and 27% of them commit more than a year out. Those two numbers sit in the Buying Policy table on the last page of the ISM Manufacturing Report On Business, below the commodity lists, below the industry quotes, below everything the wire services quote. The PMI headline (54.6 in August) gets the coverage. The lead-time table is the part a plant buyer can actually use, because it moves two quarters before the price indices that end up in your escalation clause. This debrief covers the August table, the 19 years of history we rebuilt for the [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/), the backlog and tariff mechanics behind the days, the Supplier Deliveries index against the switchgear PPI, and, behind the wall, the three lines in the report that lead the PPI and by how many months. Capex lead time was 147 days in December 2019, peaked at 186 days in June 2022, and has held between 165 and 179 days for every month since January 2023; production materials fell from a 100-day record to 84, MRO from 51 to 48 ([data/ism-lead-times.csv](data/ism-lead-times.csv), 212 monthly releases). ## The Table Nobody Reads ISM asks its manufacturing panel one plain question each month: how far ahead are you committing for capital expenditures, for production materials, and for MRO supplies. Answers land in six buckets, hand-to-mouth (counted as 5 days), 30, 60, 90, 180 and 360 days or more, and ISM publishes the percent in each bucket plus the weighted average. In August, capex answers split [18% hand-to-mouth, 3% at 30 days, 8% at 60, 13% at 90, 31% at six months and 27% at a year or more](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html). The weighted average is 171 days. ISM notes the responses are raw, never revised and not seasonally adjusted, which is exactly why they are useful: nobody smooths them. **The 27% is the number that matters, not the 171.** A weighted average of 171 days describes no actual buyer. It blends a hand-to-mouth bolt buyer with a plant waiting on a 2,500 kVA transformer. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) publishes the class detail behind that blend: Schneider's September guide puts LV switchgear at 45 to 50 weeks and switchboards at 18 to 34 weeks, and the sizing tool for a pad-mount runs the [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) numbers against the same published lead times. The year-plus bucket in the ISM table is the electrical room, the chiller plant and the machine tool with a custom spindle. Everything else in a factory still ships inside two quarters. ISM says it has collected Buying Policy since 1987, but the archive is members-only, so we rebuilt the series from every monthly release we could recover: [May 2007 at 115 days](https://web.archive.org/web/20070527090743/http://www.ism.ws/ISMReport/MfgROB.cfm), [August 2010 at 127](https://web.archive.org/web/20100906104343/http://ism.ws/ISMReport/MFGROB.cfm), [September 2014 at 120](https://web.archive.org/web/20141018194916/http://www.ism.ws/ismreport/mfgrob.cfm), [August 2018 at 144](https://web.archive.org/web/20180917020731/https://www.instituteforsupplymanagement.org/ismreport/mfgrob.cfm?SSO=1), and every month since. The whole file, month by month with the release URL on each row, is in [data/ism-lead-times.csv](data/ism-lead-times.csv) and in the Monitor's history tab. Two things stand out. Capex commitments lengthened by roughly 30 days across the 2010s before anyone said "supply chain", and the pandemic step-up of 2021 to 2022 never reversed: three and a half years after the peak, the series is 24 days above its pre-pandemic level while materials and MRO have given most of theirs back. ## Where the 171 Days Come From The days are backlog. [Census M3](https://www.census.gov/manufacturing/m3/) tracks unfilled orders and shipments for NAICS 3353, the electrical equipment industry that makes transformers, switchgear, motors and generators, and the ratio of the two is the cleanest public measure of how many months of work the factories already hold. Unfilled orders at electrical equipment makers were [$28.4 billion against $5.0 billion of monthly shipments in July 2026](https://www.census.gov/manufacturing/m3/), 5.7 months of work, versus an average of 3.2 months in 2017 to 2019 and a peak of 7.1 months in October 2022. Shipments are up 45% from January 2019 in nominal dollars; unfilled orders are up 154%. The industry has added capacity and raised prices and still holds nearly twice the backlog it did before the boom. That is what a 171-day average and a 27% year-plus bucket look like from the seller's side. The [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) shows the same shape one company at a time from filings, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) marks which of those companies will tell a buyer a lead time at all. The second source is the ISM commodities list, which the wire stories skip entirely. Each month ISM prints which commodities respondents report up in price, down in price and in short supply, with the count of consecutive months beside each. Eight items were in short supply in August: [electronic components for the 18th consecutive month, electrical components for the 14th, memory for the 8th, and copper, printed circuit boards, steel and tungsten products for the 2nd](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), plus labor for the first. Electrical components entered the list in [July 2025](https://web.archive.org/web/20251102130113/https://www.ismworld.org/supply-management-news-and-reports/reports/ism-pmi-reports/pmi/august/) (the August 2025 report showed it at 2 months) and has not left. In the 2021 cycle the same line ran for [13 straight months by October 2021](https://web.archive.org/web/20211111061944/https://www.ismworld.org/supply-management-news-and-reports/reports/ism-report-on-business/pmi/october/), the month ISM said production materials and MRO lead times were the longest since it began collecting the data in 1987. The month-by-month record of that one line is in [data/electrical-components-short-supply-months.csv](data/electrical-components-short-supply-months.csv). Aluminum has been reported up in price for 33 straight months, copper for 14, steel for 10, steel products for 9, electronic components and hot-rolled steel for 8, and 24 commodities were on the up list against two on the down list. A Machinery respondent put it directly: ["Prices continue to rise on all goods. Suppliers are noting that energy, steel and labor costs are increasing very quickly."](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html) The 33 months of aluminum start in December 2023, well before the current tariff schedule, and the tariffs then stacked on top: Section 232 steel and aluminum duties rose [from 25% to 50% ad valorem effective June 4, 2025 (Proclamation of June 3, 2025, 90 FR 24199)](https://www.federalregister.gov/documents/2025/06/09/2025-10524/adjusting-imports-of-aluminum-and-steel-into-the-united-states), and semi-finished copper and intensive copper derivatives took a [50% duty effective August 1, 2025 (Proclamation of July 30, 2025, 90 FR 37727)](https://www.federalregister.gov/documents/2025/08/05/2025-14893/adjusting-imports-of-copper-into-the-united-states). Copper's 14 months on the up list begin in July 2025, the month before that proclamation took effect. The [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) lists both instruments by HTS line, and the [Copper and Wire Price Sheet](https://open-factory.com/tools/copper-wire-price-sheet/) turns the COMEX move into $/ft for a bid. ## Supplier Deliveries and the Switchgear PPI The headline table has one more line buyers should read before the PMI: Supplier Deliveries. It is the [only ISM index that is inverted](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), above 50 means slower deliveries, and it is the one that speaks directly to lead time. Supplier Deliveries registered [59.3 in August, slowing for the ninth month in a row](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), against 51.3 in August 2025 and a 2023 low of 43.5. On the right panel, the switchgear PPI (WPU1175) was [up 9.3% year over year in July 2026 and the transformer PPI (WPU117409) up 8.7%](https://data.bls.gov/timeseries/WPU1175), after both ran at 20 to 43% in 2022. The two episodes on the chart behave differently by product. In 2020 to 2022, Supplier Deliveries went above 55 in February 2020, stayed there until September 2022 and peaked at 78.8 in May 2021. The transformer PPI passed 10% year over year in May 2021, the switchgear PPI not until January 2022. In 2025, Supplier Deliveries crossed 55 in April and the switchgear PPI stepped from 5.8% to 9.0% in the same month, then ran to 13.4% by February 2026, while transformers stayed under 6% until July 2026. Transformers move on copper and GOES steel almost immediately; switchgear moves on the backlog, two to three quarters later. The ISM Prices index (71.1 in August, second month at that level) has the same relationship with more lag, which is the subject of the paid section. Our piece on [why switchgear costs 50% more than in 2021](https://open-factory.com/articles/switchgear-50-percent/) decomposes the price move into copper, GOES steel and OEM margin. This one is about the timing signal that precedes it. Why the lag exists is mundane. The PPI measures prices on shipments, and a switchgear lineup shipped in July 2026 was quoted 45 to 50 weeks earlier under Schneider's guide, so July's PPI reflects the pricing of the summer of 2025. ISM's respondents answer about the month they are in. Whatever ISM's panel reports about delivery speed and input prices today is what the PPI will report about shipments in two to four quarters. The [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) uses this lag to carry each electrical line item at the lead time it will actually ship at, not the price on the day the budget was set, and the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) turns the same published weeks into an energization month. ## The Three Categories Are Three Different Problems Production materials at 84 days and MRO at 48 are back near their long-run trend: materials averaged 64 days in 2018 and 2019, MRO 33, and both have settled 20 and 15 days above that. The gap is inventory policy, not scarcity. Respondents kept more weeks of cover after 2021 and never went back, and Customers' Inventories at [42.8, "too low" for the 23rd month](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), says their customers are pulling that cover down again. The long file makes the same point in a different way. In 2010 and 2011 the three lines averaged 111, 54 and 24 days; in 2018 and 2019, 146, 67 and 34 ([data/ism-lead-times.csv](data/ism-lead-times.csv)). Capex commitments grew 35 days across a decade with no shortage, no tariff and no pandemic, because the equipment got more engineered and the factories that build it got fewer. The pandemic added 40 more days at the peak and left 24 of them behind. Anyone budgeting a 2027 project off a 2019 lead-time memory is 24 days late before the RFQ goes out, and off a 2012 memory, two months late. Capex is a different problem. A 171-day average with 27% at a year or more means the queue is the product. Slots at Hitachi Energy, Siemens Energy, Eaton and Powell trade against deposits, and the [$60 million deposit economy](https://open-factory.com/articles/slot-economy/) that grew up around datacenter electrical gear is now the environment a factory buyer bids into. The [Wood Mackenzie 128 to 160 week transformer figures](https://open-factory.com/articles/transformer-160-weeks/) describe the top of the year-plus bucket. The ISM table describes how much of manufacturing has fallen into it. The [Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) shows what public buyers have been paying for the gear at the front of that queue, and [Quote Check](https://open-factory.com/tools/quote-check/) puts a quote in hand against both the price band and the published lead time. ## What to Do Monday Read the table on the first business day of the month, before the PPI (mid-month) and before the Census M3 full report (the July M3 report [landed September 2, the August one lands October 2](https://www.census.gov/manufacturing/m3/release_schedule.html)). Four actions follow from the August numbers. First, any capex line in the 2027 budget that touches the electrical room needs a PO or a slot reservation this quarter, because 171 days is the average and the electrical room is in the 27%. Run the schedule through the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) with the Monitor's September published weeks and see which month the energization actually lands. Second, write the escalation clause against a named index and a named month: WPU1175 or WPU117409, base month equals quote month. With the PPI at 9% and the ISM Prices index at 71, an unindexed fixed price is a 9% gift to the vendor or a change-order fight, and the [RFQ guide](https://open-factory.com/articles/rfq-capital-equipment/) has the clause text. Third, put the tariff instrument in the quote: the 50% steel and aluminum rate and the 50% copper rate, by proclamation and effective date, so a pass-through claim mid-contract can be checked against the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) rather than argued. Fourth, add the ISM Buying Policy row to the monthly procurement report next to the PMI, with the three-month change beside it. It costs nothing, it arrives on the first business day, and it is the only free monthly number that describes your own queue rather than someone else's output. The paid section tests the report as a leading indicator. Across 2020 to 2026, **the ISM Prices index led the transformer PPI's year-over-year rate by nine months with a correlation of 0.84**, and two other lines in the same free release lead it by less with fewer false starts. The full lag table, the thresholds we would trade on, and the current readings are below and in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). --- # Book-to-Bill League Table, Q2 2026: Every Electrical OEM URL: https://open-factory.com/articles/book-to-bill-league/ Date: 2025-09-10 Electrical / Debrief Eaton 1.2 to 1.3, Siemens Energy Grid 1.48 and EUR 51B, GE Vernova $44.6B, ABB +59%, Powell $2.4B, Hitachi $63.6B, the read-across for buyers Powell Industries booked $934 million of switchgear orders in the quarter to June 30 2026 and shipped $312 million, a book-to-bill of 3.0. That is the top of the table. The bottom of the table is Eaton at 1.2, and Eaton is the only one of eight ratio-reporting OEMs that measures it over twelve months rather than one quarter. Every electrical, power and cooling OEM that tells you its orders is booking more than it ships, and ten of the eighteen we track print no ratio at all. This debrief builds the first Open Factory [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/), version 0.1, from 10-Qs, 10-Ks and earnings releases only, and covers the eight ratios, the $300 billion backlog stack, the quarter book-to-bill peaked, what Vertiv's deposits say now that its backlog is gone, and behind the wall the read-across for buyers, OEM by OEM. Eight ratios, all above 1.0: Powell 3.0, Hitachi Energy 2.1, GE Vernova Electrification 1.75, Siemens Energy Grid Technologies 1.48, Caterpillar 1.46 implied at company level, ABB Electrification 1.39, Trane 1.23 and Eaton 1.2. ## The League Table The table is the model. Each row is one filing or release for the calendar quarter ended June 30 2026, with fiscal quarters mapped (Siemens Energy Q3 FY2026, Hitachi Q1 FY2026, Powell and Johnson Controls Q3 FY2026, Modine Q1 FY2027). Where a document has no figure the cell says n/d, because a blank cell would let the reader assume we forgot to look. **Of eighteen OEMs, eight print a book-to-bill or a dollar backlog for the quarter, one prints percentages only, one prints a data center backlog only, and eight print nothing.** The full rows, with a source URL on every line, are in the CSV under the table and in the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) model page. | OEM | Segment | Orders Q2 | Revenue Q2 | Backlog Jun 30 | Book-to-bill | Backlog YoY | |---|---|---|---|---|---|---| | Powell Industries | Company | $934M | $312M | $2.4B | 3.0 | +69% | | Hitachi Energy | Energy segment | ¥1,906bn | ¥912bn | ¥10.3tn ($63.6B) | 2.1 | +12% vs Mar | | GE Vernova | Electrification | $6.35B | $3.64B | $40.6B | 1.7 | +69% | | Siemens Energy | Grid Technologies | €5.37bn | €3.62bn | €51bn | 1.48 | +34% | | ABB | Electrification | $7.23B | $5.20B | $13.7B | 1.39 | +59% | | Trane Technologies | Enterprise | $7.82B | $6.35B | $12.1B | 1.23 | +70% | | Eaton | Electrical | n/d (+41% R12) | $3.95B (Americas) | n/d | 1.2 (R12) | +43% | | Caterpillar | Company / Power & Energy | n/d | $8.24B (P&E) | $72.1B | ~1.5 implied | +92% | | Johnson Controls | Solutions and Services | n/d (+27% org.) | $6.6B | $21.0B | n/d | +32% | | Vertiv | Company | n/d (Q4: +252%) | $3.27B | n/d (Q4: $15.0B) | n/d (Q4: 2.9) | n/d (Q4: +109%) | | Carrier | Company | n/d (~+40%) | $6.35B | n/d | n/d | n/d | | nVent | Company | n/d | $1.5B | n/d (Q1: $2.6B) | n/d | n/d | | Generac | C&I | n/d | $556M (C&I) | ~$1.6B (data center) | n/d | n/d | | Schneider Electric | Group | n/d | €11.46bn | "record", no figure | n/d | n/d | | Cummins | Power Systems | n/d | $2.3B | n/d | n/d | n/d | | Modine | Data Centers | n/d | $349M (DC) | n/d | n/d | n/d | | Hubbell | Company | n/d | $1.71B | n/d | n/d | n/d | | Atkore | Company | n/d | $795M | n/d | n/d | n/d | Data: [data/book-to-bill-q2-2026.csv](data/book-to-bill-q2-2026.csv) (50 rows, Q4 2025 to Q2 2026, source URL per row) and [data/05-league-table.csv](data/05-league-table.csv). Reading the ratios in order. Powell's 3.0 is [$934 million of orders against $311.7 million of revenue](https://www.sec.gov/Archives/edgar/data/80420/000008042026000103/ex991-powlq3xfy2026earning.htm), with one behind-the-meter data center order above $400 million inside it, which is a sixth of the $2.4 billion backlog on its own. Hitachi Energy's 2.1 is [¥1,906 billion of orders against ¥912 billion of revenue](https://www.hitachi.com/content/dam/hitachi/global/en/press/files/2026/07/260729/2026_1Qpre.pdf), orders up 87% on several large HVDC awards in Europe. GE Vernova's Electrification segment [booked $6,347 million against $3,637 million of revenue](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm), a ratio the release rounds to 1.7. Siemens Energy Grid Technologies [printed 1.48 on €5,367 million of orders](https://assets.siemens-energy.com/dam/d0147174-31a9-4a78-b062-b49d00428fc4/earnings-release-q3-fy2026-en-pdf_Original%20file.pdf) and named transformers, including data center projects, as the largest contributor. ABB Electrification [took $7,231 million of orders on $5,200 million of revenue](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf), its first quarter above $7 billion, with Americas orders up 114%. Trane's 1.23 comes from [$7,818 million of bookings on $6,353.5 million of revenue](https://www.sec.gov/Archives/edgar/data/1466258/000162828026050743/exhibit991-q22026earningsr.htm), with Americas Commercial HVAC bookings up 50% and applied equipment bookings up 130%. Eaton's 1.2 is the [rolling twelve-month ratio for its Electrical businesses](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm); the in-quarter Americas figure of 1.3 was given on the call, [as reported by Yahoo Finance](https://finance.yahoo.com/markets/stocks/articles/eaton-corp-plc-etn-q2-230128644.html), not in the release. Caterpillar prints no orders, but its 10-Q says [backlog rose $9.4 billion in the quarter on $20.5 billion of sales](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm), which implies orders of roughly $29.9 billion and a company-wide ratio near 1.46, with the largest increase in Power & Energy. The [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) uses these same backlog rows as its filings input, so a buyer checking a quoted transformer or switchgear delivery against the published schedule is looking at the same numbers. ## Who Prints a Number and Who Does Not The disclosure gap is the second finding, and for a buyer it matters as much as the ratios. **Eight of the eighteen largest listed suppliers of electrical, power and cooling equipment tell the market nothing about orders or backlog in dollars.** Schneider Electric reported [Q2 revenue of €11,459 million, up 16.5% organic, and "a record backlog"](https://www.se.com/ww/en/assets/pdf/release-hy-results-2026) with no number attached. Carrier reported [total orders up about 40%, Commercial HVAC up about 65% and data center orders up more than 300%](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000030/a99-q22026earningsexhibit.htm), all percentages on an undisclosed base. Cummins' Power Systems segment [grew 19% to $2.3 billion on data center standby demand](https://www.sec.gov/Archives/edgar/data/26172/000002617226000026/cmi2026q28-kex99.htm) and its 10-Q has no backlog figure at all. The percentages are not useless, they are just not comparable. Carrier's April release said its [backlog "fully covers expected 2026 data center sales"](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000023/a99-q12026earningsexhibit.htm) after data center orders rose more than 500% in the March quarter, which tells a buyer that a 2026 Carrier chiller slot is gone without telling them what a 2027 one costs in weeks. Johnson Controls prints the dollars but [restated its backlog this year to include equipment-only sales for longer-cycle projects](https://www.sec.gov/Archives/edgar/data/833444/000083344426000083/q3ex991xq3fy26earningsrele.htm), so its $21.0 billion is not the same series as the $18.2 billion it printed in February without the footnote. The table keeps each OEM's own definition and says so in the basis column of the CSV. A league table that silently normalized them would be worse than one that shows the seams. Two OEMs went quiet this year. Vertiv's February release printed [a book-to-bill of about 2.9x and a backlog of $15.0 billion, up 109%](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm); its [April](https://www.sec.gov/Archives/edgar/data/1674101/000162828026026379/q12026exhibit991vrt04222026.htm) and [July](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050323/q22026exhibit991vrt07292026.htm) releases contain the words "orders" and "backlog" only in the forward-looking-statements boilerplate. nVent's May release said [orders were a record and backlog had increased to $2.6 billion](https://www.sec.gov/Archives/edgar/data/1720635/000162828026029098/q12026nvtpressrelease.htm); its [July release](https://www.sec.gov/Archives/edgar/data/1720635/000162828026051203/q22026nvtpressrelease.htm) reports sales up 47% organic and no backlog. Modine, Hubbell and Atkore have never printed one; Modine's [Data Centers segment grew 90% to $348.6 million](https://www.sec.gov/Archives/edgar/data/67347/000110465926088230/mod-20260729xex99d1.htm) while its release says supply chain constraints limited production. The [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) scores exactly this, backlog in dollars being one of its five observable facts, and the eight n/d rows above are the eight zeros in that column. ## The Backlog Stack Is About $300 Billion Add up every backlog that is disclosed and the eleven figures come to about $303 billion of equipment sold and not yet shipped. Caterpillar alone is [$72.1 billion, of which $29.2 billion is not expected to ship within twelve months](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm). Hitachi Energy is [¥10.3 trillion, which Hitachi itself translates to $63.6 billion](https://www.hitachi.com/content/dam/hitachi/global/en/press/files/2026/07/260729/2026_1Qpre.pdf). Siemens Energy Grid Technologies is €51 billion, or $58.2 billion at the [June 30 2026 rate of 1.1417](https://fred.stlouisfed.org/graph/fredgraph.csv?id=DEXUSEU). GE Vernova's Electrification equipment backlog is [$40.6 billion, up $16.6 billion in a year including $5 billion from consolidating Prolec GE](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm). The three grid-equipment makers at the top, Caterpillar excepted, hold $162 billion between them, and the seven below them hold $68 billion. The three at the top are also the three that make large power transformers, which is why the [transformer lead time sits at 128 to 160 weeks](https://open-factory.com/articles/transformer-160-weeks/) and why the Monitor's transformer row has not moved down in two years. The composition matters as much as the total: Caterpillar says [$29.2 billion of its $72.1 billion, 40%, will not ship within twelve months](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm); Powell's 10-Q counts [about $800 million of data center awards](https://www.sec.gov/Archives/edgar/data/80420/000008042026000107/powl-20260630.htm) in nine months, a third of its $2.4 billion; Eaton's [Electrical Global backlog is up 103%](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm) largely because Boyd Thermal, acquired in March for $9.55 billion, now sits inside it. **Every OEM that prints a backlog grew it between 32% and 109% in twelve months, and none grew it less than the rate at which it grew revenue.** Caterpillar's backlog is [up 92% from $37.5 billion at June 30 2025](https://www.sec.gov/Archives/edgar/data/18230/000001823025000040/cat-20250630.htm). Trane's is [$12.1 billion, up 70%, with Americas Commercial HVAC up 90%](https://www.sec.gov/Archives/edgar/data/1466258/000162828026050743/exhibit991-q22026earningsr.htm). Powell's is [up 69% from $1.4 billion](https://www.sec.gov/Archives/edgar/data/80420/000008042026000103/ex991-powlq3xfy2026earning.htm). ABB Electrification's is [up 59% on a comparable basis from $8,685 million](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf). Eaton's total Electrical backlog is [up 43%](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm), Siemens Grid Technologies [from €38 billion to €51 billion, up 34%](https://assets.siemens-energy.com/dam/3e846440-66dd-4a56-be75-b49d004e6741/2026-08-05_Q3_Analyst_presentation-pdf_Original%20file.pdf), and Johnson Controls [from a restated base to $21.0 billion, up 32% organic](https://www.sec.gov/Archives/edgar/data/833444/000083344426000083/q3ex991xq3fy26earningsrele.htm). Nine OEMs, nine positive bars, from Johnson Controls at +32% to Vertiv's final disclosure at +109%. ## Book-to-Bill Peaked in the First Quarter The ratios are high, and at the two largest grid-equipment makers they are already coming down. GE Vernova Electrification booked [$7,112 million on $2,959 million of revenue in Q1 2026, a ratio near 2.4](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000063/gevpressrelease1q26.htm), then $6,347 million on $3,637 million in Q2, near 1.75. Siemens Energy Grid Technologies went [1.95 in the December quarter](https://assets.siemens-energy.com/dam/eb218e11-1294-4e27-a750-b3ee0059367f/Earnings-Release-Q1-FY26-EN-pdf_Original%20file.pdf), [2.28 in the March quarter](https://assets.siemens-energy.com/dam/fb695b3d-e910-4f47-81b5-b4480034f6a1/Earnings-Release-Q2-FY26-EN-pdf_Original%20file.pdf), 1.48 in the June quarter. In both cases orders fell sequentially while revenue rose; Siemens says its revenue growth "was supported by expanded production capacities". **A falling book-to-bill on rising backlog is the signature of a factory that has started to catch up, not of demand rolling over, and it is the first such signature in this cycle.** GE Vernova fell from 2.5 to 2.4 to 1.75 across three quarters, Siemens rose to 2.28 then dropped to 1.48, Trane went 1.12 to 1.35 to 1.23, and Powell held 1.7 for two quarters before the $400 million order took it to 3.0. The backlog dollars tell the same story from the other side. Caterpillar's firm backlog went [$51.2 billion at December 31](https://www.sec.gov/Archives/edgar/data/18230/000001823026000008/cat-20251231.htm), [$62.7 billion at March 31](https://www.sec.gov/Archives/edgar/data/18230/000001823026000021/cat-20260331.htm), $72.1 billion at June 30, additions of $11.5 billion and then $9.4 billion. GE Vernova Electrification added $8.1 billion in Q1 and $2.0 billion in Q2. Johnson Controls added $1.8 billion then $1.0 billion. The sequential additions are shrinking everywhere except Powell and Trane, which is what a ratio above 1.0 but falling toward it looks like in dollars. Five OEMs, fifteen quarter-end bars, all rising; Caterpillar from $51.2 billion to $72.1 billion, Powell from $1.6 billion to $2.4 billion. Where the demand is coming from is not in dispute. GE Vernova's [data center orders passed $5 billion year to date, more than double all of 2025](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm), having been $2.4 billion in Q1. Powell's 10-Q counts [about $800 million of data center awards in nine months](https://www.sec.gov/Archives/edgar/data/80420/000008042026000107/powl-20260630.htm), a third of its backlog. Generac's data center backlog reached [about $1.6 billion, after roughly $1 billion of new orders since the prior update, plus nearly $700 million of committed 2027 volume](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm) with its first hyperscale customer. The [genset allocation problem](https://open-factory.com/articles/gensets-sold-out-2028/) and the [Genset TCO tool](https://open-factory.com/tools/genset-tco/) start from those rows. ## Vertiv's Deposits Say What Its Backlog Used To Vertiv stopped printing orders and backlog after February, but a balance sheet cannot go quiet. Current deferred revenue, the line where customer deposits and advance billings sit, was [$888.8 million at June 30 2024 and $3,633.7 million at June 30 2026](https://data.sec.gov/api/xbrl/companyfacts/CIK0001674101.json), 4.1 times higher. It rose [$1,171.5 million in the June 2026 quarter alone against $148.1 million in the same quarter of 2025](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050323/q22026exhibit991vrt07292026.htm), on net sales of $3,274 million. **Vertiv's customers handed it $1.8 billion of new deposits in the first half of 2026, more than ten times the $171.5 million they handed it in the first half of 2025.** Nine quarter-ends: $889 million, $1,016 million, $1,063 million, $1,095 million, $1,257 million, $1,132 million, $1,815 million, $2,462 million, $3,634 million. Our [estimate of the undisclosed Vertiv backlog](https://open-factory.com/articles/vertiv-backlog-estimate/) uses this line, the Q4 2025 print, and the 2026 guidance of $14.0 billion of sales to reconstruct a range. For a buyer the deferred revenue line matters for a simpler reason: it is the deposit money. At the deposit rates that now sit in [slot reservation agreements](https://open-factory.com/articles/slot-economy/), 10 to 30% of contract value at signing, a $1.8 billion half-year increase implies a large multiple of that in new contract value, and every dollar of it is a customer who has already paid to hold a place in Vertiv's 2027 schedule ahead of you. The Q2 release also says revenue "reflected minor timing shifts, primarily due to temporary supply chain congestion", the first mention of supply chain congestion in a Vertiv earnings release since at least the February 2026 print, and the sentence the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) UPS and CDU rows will be tested against when the Q3 filings land. ## What to Do Monday Three things follow from the table for anyone with a capital equipment PO to place this quarter. First, treat any OEM with a quarterly book-to-bill above 1.4 as allocating, not selling, and write the RFQ that way: ask for the delivery slot before the price, and put the [liquidated damages clause](https://open-factory.com/articles/liquidated-damages/) in the first draft, because the vendor's paper will cap it at the margin. That list today is Powell, Hitachi Energy, GE Vernova and Siemens Energy Grid Technologies. Check the quoted week against the [Lead-Time Schedule](https://open-factory.com/tools/lead-time-schedule/) before you accept it. Second, treat the falling ratios as the first opening. GE Vernova and Siemens Energy both shipped more than they booked relative to the prior quarter, and both say capacity is expanding. A buyer who lost a 2026 transformer slot should re-quote now, with the [Transformer Sizing and Cost tool](https://open-factory.com/tools/transformer-sizing-cost/) as the reference for what the unit should cost per kVA, and with the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) bands as the check on the number that comes back. Third, discount the OEMs that will not print a backlog. When Schneider says "record backlog" and Carrier says orders are up 40%, the buyer has no way to know whether the queue in front of them is six months or eighteen. Put the question in the RFQ: current backlog in dollars and in weeks for the product line quoted. A vendor that answers on a call but not in a filing can answer on a bid form; run the quote through [Quote Check](https://open-factory.com/tools/quote-check/) and the switchgear pricing through the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) to see whether the schedule line and the price line agree with each other. Behind the paywall: the read-across for buyers OEM by OEM, who will take price this year, who has slots opening, and who is over-earning on a backlog it cannot ship; the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) says that **ABB's Electrification margin reached 24.9%** in a quarter in which it says pricing has not yet caught up with input costs, and that is the OEM to watch on price. --- # Introducing the Open Factory Lead-Time Monitor URL: https://open-factory.com/articles/lead-time-monitor-launch/ Date: 2025-09-10 Electrical / Model launch Eleven equipment classes, every cell a published lead time with its OEM and date, Schneider LV switchgear at 50 weeks, Census M3 at 5.7 months, ISM supplier deliveries 59.3, backlog from seven filings, monthly on the first business day, what version 1 adds A low-voltage switchgear section ordered from Schneider Electric today is quoted at [50 weeks from release to manufacturing](https://www.se.com/us/en/download/document/Monthly_Lead/), per the September 2026 edition of Schneider's Lead Times Guide, published September 2. That number is one of exactly four equipment classes, out of the eleven a factory or datacenter has to buy, for which any OEM in the United States publishes a dated lead time. The other seven, including every class over 60 weeks, are known only through consultant surveys and contractor blogs. The [Open Factory Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) launches today as version 0.1 to put those eleven classes in one table, with the OEM and the date next to every cell, and to say plainly which cells are an OEM's own number and which are somebody else's. This piece gives away the launch table, walks through every input and how a cell is built, shows the three context series that move before quoted lead times do, tells each OEM what one disclosure would improve its row, and ends with what version 1 adds. The eleven ranges run from 4 to 32 weeks for drives at the bottom to 128 to 160-plus weeks for large power transformers at the top, and the four OEM-published classes (LV switchgear and switchboards at 18 to 50 weeks, MV switchgear at 15 to 40-plus, MCC and busway at 6 to 40, drives at 4 to 32) all sit below the 50-week line while the seven third-party classes all reach above it. ## What the Monitor Is, and Is Not **Version 0.1 is a compilation of public series, and every cell says where it came from.** The Monitor covers large power transformers above 100 MVA, medium power transformers of 5 to 50 MVA, pad-mount distribution transformers, MV switchgear, LV switchgear and switchboards, MCC and busway, UPS, diesel gensets, chillers, coolant distribution units, and PLC CPUs or drives. A cell holds a published lead time, the name of who published it, the document, and the date. Where the publisher is the OEM, the cell says so. Where the publisher is Wood Mackenzie, Terrapin Construction Group, Origin Partners or VAWN, the cell includes their name and the figure is theirs, not ours. The full launch table is in [lead-time-monitor-v0.1.csv](data/lead-time-monitor-v0.1.csv), 49 rows, one URL per row. Three context series sit next to the class table: Census M3 unfilled orders over shipments for NAICS 3353, the ISM Supplier Deliveries Index with ISM's capital-expenditure commitment lead time, and the BLS producer price indices for transformers and switchgear. A fourth leg is backlog as stated in filings by Powell, Eaton, Vertiv, GE Vernova, Caterpillar, Cummins and Generac, which the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/) tracks quarterly and the Monitor cites. What the Monitor is not, in version 0.1, is a survey. We have not called distributors, we hold no quotes, and no number in this piece rests on a panel. The methodology page includes the version number for that reason, and the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/) grades the vendors on whether they publish at all. When a cell reads "Terrapin, 40 to 65 weeks, June 2026", that is what a contractor wrote on its website about its own slot reservations, and the reader should weigh it as such. ## The Only OEM That Publishes a Schedule Schneider Electric publishes a Lead Times Guide on or about the second of every month, as a PDF, with a disclaimer that the figures are "for informational purposes only and should not be relied upon for actual orders". The September 2026 edition, published September 2, has [124 product rows](https://www.se.com/us/en/download/document/Monthly_Lead/), 104 of them with a made-to-order lead time counted from release to manufacturing. We parse it, keep every edition, and it becomes the OEM leg of the Monitor. No other US electrical OEM does this in public. Eaton, ABB and Vertiv publish nothing; Siemens publishes something, and we will come to what. Read across the families: LV switchgear (the LVDO line) is 45 to 50 weeks across six rows, busway is 24 weeks for I-Line II and 40 for plug-in units, MV switchgear runs 15 to 25 weeks for HVL load-interrupter gear and more than 40 for GHA and MCLD, switchboards are 18 to 34 weeks by frame, motor control centers 6 to 27 weeks by section count, drives 4 to 32 weeks (12 to 16 for the transactional Altivar range), and LV dry-type transformers 4 to 9 weeks across 13 rows. The one MV paralleling line, ASCO Power Control Systems at medium voltage, is 50 weeks. The guide colour-codes change. Orange means the lead time got longer since the prior edition, green means shorter, and Schneider prints the direction but not the previous value. We read the colours off the September pages. [CSV: data/03-schneider-flags-sep-2026.csv](data/03-schneider-flags-sep-2026.csv) Twelve rows moved out and seven moved in: all six LV switchgear rows went longer to 45 and 50 weeks, the SureSeT metal-clad line went to 18 to 30, and four breaker rows including the MTZ frames at 24 weeks went longer, while HVL and HVLCC medium-voltage gear came in to 15 to 25 weeks and three ASCO transfer-switch frames shortened. That is the entire month-over-month record the Monitor has today. Our history file starts with this edition and grows by one block a month; by the [switchgear cost piece](https://open-factory.com/articles/switchgear-50-percent/) we will have a run of editions to plot. The second OEM schedule in circulation needs a date on it. VAWN's lead-time index page cites "Siemens CompAS" for [low-voltage switchgear at 54 weeks, switchboards at 32 to 41 and busduct at 27](https://www.usevawn.com/resources/electrical-equipment-lead-times/), and the electrical trade press repeats those as Siemens' published LV schedule. The page they point to is Siemens' CompAS "Lead Time Adjustments" notice, and the document behind it is a [letter to distributors dated April 25, 2022](https://resources.compas.siemens-info.com/HomePage/Data/Documents/Leadtime_adjustments_external_4_25.pdf), which set 54 weeks for LV switchgear "for all orders not released by May 9th 2022" and warned of a further 7-week extension on June 6, 2022. The Monitor lists the Siemens rows with that date. A 2022 letter is a data point about 2022, and the honest current OEM figure for LV switchgear is Schneider's 50 weeks. Rockwell Automation is the case at the other end. Its public supply-chain page says some product lines have "returned to pre-pandemic lead times" and others "remain constrained", and tells buyers to [work with your distributor](https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html) for any actual figure. The Monitor's PLC-CPU cell is therefore empty and says so, and the drives cell is Schneider's. The [ControlLogix lead-time cycle](https://open-factory.com/articles/controllogix-36-weeks-to-stock/) is the story of why that matters. ## Where the Longest Numbers Come From For seven classes there is no OEM schedule, and the Monitor shows the best-documented third-party figure with its author's name on it. Wood Mackenzie's T&D supply-chain survey is the most cited, and its October 15, 2025 note put generator step-up transformers at [143 weeks on average and power transformers at 128 weeks in Q2 2025](https://www.woodmac.com/news/opinion/mind-the-gap-tackling-supply-chain-challenges-in-the-electric-td-sector/), with MV switchgear at 44 and pad-mount transformers still rising. VAWN reports Wood Mackenzie's 2026 read as above 160 weeks for substation units. That figure comes to us second-hand and is labelled so. Siemens Energy's own press release on its Charlotte expansion said US large power transformer lead times run [up to five years](https://www.siemens-energy.com/us/en/home/press-releases/siemens_energy_addresses_shortage_US_powertransformers_invests_new_factory.html), which is an OEM statement about the market, not a schedule for its own plant. The [transformer piece](https://open-factory.com/articles/transformer-160-weeks/) works through who gets delivered first. Terrapin Construction Group, a general contractor, published its own slot-reservation data on June 10, 2026: [pad-mount 0 to 5 MVA at 40 to 65 weeks, substation 5 to 50 MVA at 65 to 110, diesel gensets 1,500 to 4,500 kW at 50 to 110, UPS at 30 to 60, MV switchgear 52 to 104 weeks by voltage class](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026). Origin Partners, a power consultancy, updated its tracker on May 22, 2026 with [chillers at 40 to 70 weeks, CDUs at 30 to 50, metal-clad MV switchgear at 80 to 110 and new-build GSUs at 130 to 160](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026). Both are the authors' own procurement experience, not surveys, and the Monitor says so in the source column. **The two trackers disagree by more than 2x on diesel gensets, and the Monitor prints the conflict rather than picking one.** Terrapin has a 1,500 to 2,000 kW diesel at 50 to 66 weeks; Origin has a 1.5 to 3 MW standby diesel at 26 to 40. Neither is an OEM figure. What the OEMs have said is qualitative: Cummins' CEO told the Q4 2025 call on February 5, 2026 that the company is [taking power-generation orders "well into 2028"](https://www.fool.com/earnings/call-transcripts/2026/02/05/cummins-cmi-q4-2025-earnings-call-transcript/) and does not quantify its backlog; Caterpillar's 10-Q for June 30, 2026 shows $72.1 billion of firm backlog, [$29.2 billion of it beyond twelve months](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm). The [genset piece](https://open-factory.com/articles/gensets-sold-out-2028/) and the [Genset TCO](https://open-factory.com/tools/genset-tco/) tool include the class detail. Chillers and CDUs are the same shape. Trane, Carrier and Johnson Controls publish no lead times. The only OEM datapoint is a delivery window in a press release, such as Munters' SEK 2.1 billion order for CRAHs, CDUs and chillers with [deliveries from Q4 2026 through Q1 2028](https://www.munters.com/en-us/news-media/press-releases/2025/munters-wins-record-orders-for-data-center-equipment-with-a-total-value-of-2.1-bsek). The [chiller piece](https://open-factory.com/articles/chiller-60-week-problem/) treats that class in full, and the [Datacenter MEP Cost Table](https://open-factory.com/models/mep-cost-table/) lists each line's lead time from the Monitor. [CSV: data/07-coverage-by-class.csv](data/07-coverage-by-class.csv) Four rows are OEM schedules (MV switchgear, LV switchgear and switchboards, MCC and busway, drives), one row is half covered (PLC CPUs have no figure, drives do), and the six remaining rows lean on Wood Mackenzie, Terrapin, Origin or a trade association figure republished by VAWN, which is the 30-week NEMA number for distribution transformers. The launch table, compressed to one row per class: | Class | Published lead time | Who published it | Date | |---|---|---|---| | Large power transformers (>100 MVA) | 128 wk avg (power), 143 wk avg (GSU); >160 wk 2026 read | Wood Mackenzie survey; 2026 read via VAWN | 2025-10-15; 2026-08-13 | | Medium power transformers (5 to 50 MVA) | 65 to 110 wk | Terrapin Construction Group | 2026-06-10 | | Pad-mount distribution transformers | 40 to 65 wk; 30 wk | Terrapin; NEMA via VAWN | 2026-06-10; 2026-08-13 | | MV switchgear | 15 to >40 wk | Schneider Electric Lead Times Guide | 2026-09-02 | | LV switchgear and switchboards | 18 to 50 wk | Schneider Electric Lead Times Guide | 2026-09-02 | | MCC and busway | 6 to 40 wk | Schneider Electric Lead Times Guide | 2026-09-02 | | UPS | 30 to 60 wk | Terrapin Construction Group | 2026-06-10 | | Diesel gensets | 50 to 110 wk; 26 to 40 wk (conflict) | Terrapin; Origin Partners | 2026-06-10; 2026-05-22 | | Chillers | 40 to 70 wk | Origin Partners | 2026-05-22 | | CDUs | 30 to 50 wk | Origin Partners | 2026-05-22 | | PLC CPUs or drives | drives 4 to 32 wk; CPUs none | Schneider Electric; Rockwell publishes none | 2026-09-02 | [CSV: data/lead-time-monitor-v0.1.csv](data/lead-time-monitor-v0.1.csv) (49 rows, every row with document and URL) ## Three Series That Move Before the Quotes A published lead time is a lagging number: the OEM changes the guide after the backlog has already grown. Three public series move first, and the Monitor includes them as context on the same page. Census M3 for NAICS 3353, which covers transformers, switchgear, motors, generators and relays, shows [$28.4 billion of unfilled orders against $5.0 billion of monthly shipments in July 2026](https://www.census.gov/manufacturing/m3/), a ratio of 5.7 months, against a 2017 to 2019 average of 3.2 and a peak of 7.07 in October 2022. The ratio has not been below 5.4 months since December 2021. Read it as the whole industry's backlog in months of output: when it rises, every OEM's guide follows within a few editions, and when it falls, the guide lags on the way down too. The Census publishes about the first business day of each month, so the Monitor's monthly update lands the same week. The BLS producer price index for [power and distribution transformers stood at 381.8 in July 2026](https://data.bls.gov/timeseries/WPU117409), 91% above January 2020 and 4.6% above June, the largest single month since 2022; [switchgear and switchboards were at 419.3](https://data.bls.gov/timeseries/WPU1175), 78% above January 2020. Price is the other face of the same queue. When a factory is sold out, the list price moves before the guide does, which is why the [switchgear cost decomposition](https://open-factory.com/articles/switchgear-50-percent/) and the Monitor are two views of one table. ISM's Supplier Deliveries Index [read 59.3 in August 2026](https://www.prnewswire.com/news-releases/manufacturing-pmi-at-54-6-august-2026-ism-manufacturing-pmi-report-302865127.html), its ninth month above 50 (the index is inverted, so above 50 means slower deliveries), and the same release put the average commitment lead time for capital expenditures at 171 days, production materials at 84 and MRO supplies at 48. Electrical components have appeared on ISM's "commodities in short supply" list for 14 consecutive months. The capex lead-time table is the most ignored number in the report, and it gets [its own piece](https://open-factory.com/articles/171-day-capex-lead-time/). We rebuilt sixteen months of both series from the monthly releases because ISM's own archive is members-only. The rows and the release URLs are in [data/06-ism-supplier-deliveries-capex.csv](data/06-ism-supplier-deliveries-capex.csv). ## The Backlog Leg The fourth input is what OEMs tell the SEC. A backlog figure with a twelve-month split is the closest thing to a published lead time that most of these companies produce. [CSV: data/08-backlog-from-filings.csv](data/08-backlog-from-filings.csv) Powell Industries reported [$2.4 billion of backlog at June 30, 2026, with about $1.3 billion expected to convert within twelve months](https://www.sec.gov/Archives/edgar/data/0000080420/000008042026000107/powl-20260630.htm), which for a switchgear house implies more than a year of work already sold. Caterpillar's $72.1 billion carries $29.2 billion beyond twelve months. GE Vernova's Electrification segment [booked $6.3 billion of orders against a book-to-bill of about 1.7](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm) in the June quarter. Eaton [publishes no dollar backlog](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm), only that Electrical backlog grew 43% year over year with a rolling book-to-bill of 1.2. Vertiv's [$15.0 billion at December 31, 2025](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm) is the last figure it published before dropping the disclosure, which is why we [estimate it](https://open-factory.com/articles/vertiv-backlog-estimate/). Generac's [datacenter backlog is about $1.6 billion](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm) as of July 29, 2026. The [Q2 2026 league table](https://open-factory.com/articles/book-to-bill-league/) ranks all of them. ## How a Cell Is Built The rules, in the order we apply them. First, an OEM's own dated schedule beats everything: if Schneider prints 50 weeks for LVDO on September 2, that is the cell, and the Monitor does not average it with a survey. Second, when no OEM schedule exists, the cell shows the most recent third-party figure with a named author, a document and a date, and the author's name is part of the cell. Third, when two third parties conflict, both appear. Fourth, an OEM statement in a filing or press release ("up to five years", "orders well into 2028") is recorded as a statement, never converted into a week count for the chart. Fifth, a republished figure (Siemens via VAWN, NEMA via VAWN) carries both names and the date of the underlying document where we can find it, which is how a 2022 letter gets its 2022 date back. Refresh cadence follows the sources: Schneider on or about the second of the month, Census M3 and ISM on the first business day, BLS PPI mid-month with four months of preliminary values, filings quarterly. The Monitor publishes on the first business day of each month with whatever the sources have released, and every edition's rows are kept so the history accumulates. Our own compilation, the Schneider parse and the ISM rebuild are in the CSVs linked above. Nothing in the charts comes from anywhere else. ## What OEMs Could Publish Tomorrow **The Monitor gets better in proportion to what the OEMs put on a public page, and today only one of them does it.** Schneider's guide is the model. It is a PDF, it is disclaimed, and it is still the only dated, product-level lead-time schedule any US electrical OEM publishes. Three additions would make it the industry reference: the prior month's value next to the colour flag, so the change is a number and not an adjective; UPS and MV transformers, which the guide omits although Schneider sells both; and a stable URL per edition, so a buyer can cite the June guide in a June PO. Siemens should date its CompAS page. The 54-week LV switchgear figure is circulating in 2026 as current because the page has no date and the letter behind it is from April 2022. A monthly refresh with a publication date would make Siemens the second OEM with a live row. Eaton, ABB and Vertiv publish nothing, and each of them tells the SEC a backlog growth rate or a book-to-bill every quarter. A dollar backlog with a twelve-month split, which Powell and Caterpillar already provide, is the minimum that lets a buyer convert a filing into a delivery year. Caterpillar, Cummins and Generac could publish lead time by kW band the way their dealers already quote it. Trane, Carrier and Johnson Controls could publish lead time by tonnage band. Rockwell could publish the ControlLogix schedule its distributors hold. Every one of these disclosures is graded on the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/), and every company named here receives this text ten business days before publication. ## What to Do Monday Pull the [launch CSV](data/lead-time-monitor-v0.1.csv) and put the source column into your schedule risk register, not just the week count. A 50-week Schneider figure and a 50-to-66-week Terrapin figure are different kinds of number: one is the OEM's own planning guide for the product you are buying, the other is a contractor's experience of a class. Where your class is grey on the first chart (transformers, gensets, UPS, chillers, CDUs), the published number is not the OEM's and your PO should say what the OEM's is: write the quoted lead time, the release-to-ship convention and the liquidated damages into the order, and use the [Lead-Time-Adjusted Schedule](https://open-factory.com/tools/lead-time-schedule/) to see what a 12-week slip on that one line does to energization. Check every quote against the OEM's published figure before you sign. If a distributor quotes 60 weeks on Schneider LVDO when the September guide says 50, the [Quote Check](https://open-factory.com/tools/quote-check/) tool will show the gap, and the ten weeks are either the distributor's queue or a spec that is not in the guide's standard row. For a pad-mount or substation unit, run the [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) tool to see whether a size step changes which queue you are in, since the Terrapin bands jump from 40 to 65 weeks at 5 MVA to 65 to 95 above it. Watch the Census ratio and the ISM index on the first business day. If the 5.7 months rises again and supplier deliveries stay above 58, expect the next Schneider edition to carry more orange than green. Run the [Tariff Exposure Calculator](https://open-factory.com/tools/tariff-exposure-calculator/) on the same BoM, because price and lead time move together: the transformer PPI rose 4.6% in July 2026 alone, and a line that reprices is a line whose queue is full. ## What Version 1 Adds Version 0.1 is the public record, compiled and dated. Version 1 adds the panel: a distributor and EPC panel reporting, by OEM and class, the lead time quoted at order and the lead time actually delivered, so the Monitor can show quoted versus shipped for Schneider, Eaton, ABB, Siemens, Vertiv, Caterpillar and Cummins side by side. Panel cells will publish only when a cell has five or more contributors and eight or more quotes with no contributor above 40%, the same rules that govern the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/), and contributors stay anonymous. Until those cells exist, no article of ours will print a panel number, and the model page will say version 0.x. Two more legs are already in the pipeline. The [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) is pulling award dates and delivery dates from utility and municipal procurement records, which gives a second, independent measure of transformer and switchgear lead times from public documents rather than from anyone's survey. And the Schneider history file, one block a month from September 2026, becomes the first public month-over-month series of an OEM's own lead-time changes by product line. The headline table stays free. Class detail with every source row is Pro, and the history and API are institutional, with the methodology on the [model page](https://open-factory.com/models/lead-time-monitor/) and the sources named on every chart. --- # Model: Book-to-Bill League Table URL: https://open-factory.com/models/book-to-bill-table/ Status: Compiling data Headline: 3.0x Powell Industries book-to-bill, quarter ended June 30 2026: $934M of orders on $312M of revenue Quarterly orders, backlog and book-to-bill for every listed electrical, power and cooling OEM, from filings only, with the read-across for buyers: which classes are getting longer and which are clearing. ## What the headline is **3.0x** is Powell Industries' reported book-to-bill for the quarter ended June 30 2026: orders of $934 million against revenue of $311.7 million, backlog $2.4 billion, up 69% on a year earlier, in the [Q3 FY2026 earnings release](https://www.sec.gov/Archives/edgar/data/80420/000008042026000103/ex991-powlq3xfy2026earning.htm) filed 3 August 2026. The [10-Q](https://www.sec.gov/Archives/edgar/data/0000080420/000008042026000107/powl-20260630.htm) says $1.3 billion of that backlog ships within twelve months. The next seven: Hitachi Energy 2.09 (derived, yen), GE Vernova Electrification 1.75, Siemens Energy Grid Technologies 1.48, Caterpillar about 1.46 (implied from the backlog change), ABB Electrification 1.39, Trane 1.23, Eaton Electrical 1.2 (rolling twelve months). The rows are in [book-to-bill-q2-2026.csv](https://open-factory.com/articles/book-to-bill-league/). The headline is the highest ratio in the quarter with its basis stated. A derived ratio can lead only if the orders and revenue figures are both reported. ## What it is compiled from Version 0.1 is filings only. Nothing is asked of any company. - **Reported ratios:** [Powell](https://www.sec.gov/Archives/edgar/data/80420/000008042026000103/ex991-powlq3xfy2026earning.htm) (quarter), [Siemens Energy](https://assets.siemens-energy.com/dam/d0147174-31a9-4a78-b062-b49d00428fc4/earnings-release-q3-fy2026-en-pdf_Original%20file.pdf) (quarter), [Trane](https://www.sec.gov/Archives/edgar/data/1466258/000162828026050743/exhibit991-q22026earningsr.htm) (quarter, "123%"), [Eaton](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm) (rolling twelve months). - **Derived ratios:** orders divided by revenue where both are reported: [GE Vernova](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm) ($6,347M / $3,637M), [ABB](https://resources.news.e.abb.com/attachments/published/137496/en-US/9123465FF532/ABB_Q2_2026_Press_release_English.pdf) ($7,231M / $5,200M), [Hitachi](https://www.hitachi.com/content/dam/hitachi/global/en/press/files/2026/07/260729/2026_1Qpre.pdf) (1,906.3 / 911.9 billion yen). - **Implied ratio:** [Caterpillar](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm), from the change in company backlog plus sales over sales, labelled implied. - **Backlog in dollars:** Caterpillar $72.1B, Hitachi Energy $63.6B (10.3 trillion yen at Hitachi's translation), Siemens Energy GT $58.2B (EUR 51bn at 1.1417 [DEXUSEU](https://fred.stlouisfed.org/series/DEXUSEU), June 30 2026), GE Vernova Electrification $40.6B, [Johnson Controls](https://www.sec.gov/Archives/edgar/data/833444/000083344426000083/q3ex991xq3fy26earningsrele.htm) $21.0B, ABB Electrification $13.7B, Trane $12.1B, Powell $2.4B, [Generac](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm) about $1.6B (data center products only). - **Not disclosed, and said so:** [Vertiv](https://www.sec.gov/Archives/edgar/data/1674101/000162828026050323/q22026exhibit991vrt07292026.htm) (deferred revenue $3,633.7M is tracked instead, from [XBRL companyfacts](https://data.sec.gov/api/xbrl/companyfacts/CIK0001674101.json)), [Schneider Electric](https://www.se.com/ww/en/assets/pdf/release-hy-results-2026) ("record backlog", no figure), [Carrier](https://www.sec.gov/Archives/edgar/data/1783180/000178318026000030/a99-q22026earningsexhibit.htm) (orders growth only), [Cummins](https://www.sec.gov/Archives/edgar/data/26172/000002617226000026/cmi2026q28-kex99.htm), [nVent](https://www.sec.gov/Archives/edgar/data/1720635/000162828026051203/q22026nvtpressrelease.htm) (Q1 only), [Modine](https://www.sec.gov/Archives/edgar/data/67347/000110465926088230/mod-20260729xex99d1.htm), [Hubbell](https://www.sec.gov/Archives/edgar/data/48898/000162828026049934/exhibit991_07282026.htm), [Atkore](https://www.sec.gov/Archives/edgar/data/1666138/000162828026051510/atkr3q26exhibit991.htm). ## What is free and what is paid - **Free:** the headline ratio, this page, and the ranked chart in [Book-to-Bill League Table, Q2 2026](https://open-factory.com/articles/book-to-bill-league/). - **Pro:** the eighteen-row table each quarter with orders, revenue, backlog, basis, backlog change and datacenter commentary, plus the quarter-by-quarter series back to Q3 2025. - **Institutional:** the workbook with every filing linked, the dashboard, the API, quarterly updates for a year, an onboarding call and ad hoc calls. ## Articles built on it [Book-to-Bill League Table, Q2 2026: Every Electrical OEM](https://open-factory.com/articles/book-to-bill-league/), [Vertiv Stopped Reporting Its Backlog. Here Is Our Estimate](https://open-factory.com/articles/vertiv-backlog-estimate/), [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/) and [The $60 Million Deposit](https://open-factory.com/articles/slot-economy/). The disclosure behaviour of the same eighteen companies is scored on the [OEM Disclosure Scorecard](https://open-factory.com/models/oem-disclosure-scorecard/). ## Version history - **0.1 (September 2026).** Eighteen OEMs, calendar quarters Q4 2025 to Q2 2026. Euro and yen conversions at quarter-end FRED rates or the company's own translation, stated per row. ## What version 1 adds Private OEMs from interviews (with the count printed only when true), segment-level ratios where an OEM reports them on the call but not in the release, and a lead-time read-across column that maps each OEM's backlog months to the classes in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). ## Known gaps Johnson Controls reports orders as growth only, so no ratio is derivable. Eaton's ratio is not comparable to the quarterly ones and is ranked separately. Generac's backlog is data center products only and excludes its second hyperscaler agreement. Hitachi's yen backlog is translated at Hitachi's own rate. --- # Model: Datacenter MEP Cost Table URL: https://open-factory.com/models/mep-cost-table/ Status: Compiling data Headline: $9.8M per MW all-in datacenter build cost, Phoenix and Columbus, Turner & Townsend Data Centre Construction Cost Index 2025 Capex per MW by component on the Turner & Townsend and JLL baselines, air-cooled and liquid-cooled, with each line's lead time from the Lead-Time Monitor and the carrying cost of delay. ## What the headline is **$9.8 million per MW** is the all-in construction cost for a datacenter in Phoenix or Columbus in Turner & Townsend's [Data Centre Construction Cost Index 2025](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends), printed there as $9.8 per watt. The same index puts Atlanta at $9.9, Silicon Valley at $13.3 and Tokyo at $15.2, and allocates a US air-cooled build 54% to electrical including equipment, 22% to mechanical including equipment, 14% to core, shell and architectural and 10% to GC fees. A liquid-cooled build shifts to 48% electrical and 33% mechanical at a 7 to 10% premium. The allocation rows are in [06-mep-cost-shares.csv](https://open-factory.com/articles/chiller-60-week-problem/). The headline is the Turner & Townsend figure for the lowest-cost US Tier 1 market and moves when the index is republished. ## What it is compiled from Version 0.1 is published baselines and arithmetic. No contributed quotes. - **Baselines.** [Turner & Townsend DCCI 2025](https://reports.turnerandtownsend.com/data-centre-construction-cost-index-2025/data-centre-cost-trends) for $/W by market and the category allocation. [JLL Global Data Center Outlook 2026](https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-reports/26-research-global-data-center-outlook-new.pdf) and CBRE North America reports for market context, cited "as reported by" the outlet where the report is paywalled. - **Component split.** The electrical and mechanical shares are broken into lines using vendor disclosures where they exist (Vertiv's [2025 annual report](https://www.sec.gov/Archives/edgar/data/0001674101/000119312526174909/d27762dars.pdf) for thermal versus power mix, [Morgan Stanley's GB300 cooling BoM as reported by Tom's Hardware](https://finance.yahoo.com/news/cooling-system-single-nvidia-blackwell-154500100.html), $49,860 per rack) and trade estimates marked as such (CDUs $500 to $800 per kW installed, [Energy Solutions](https://energy-solutions.co/articles/sub/data-center-cooling-liquid-immersion-vs-air), verify). - **Lead times.** Each line carries its class from the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/): LV switchgear 50 weeks (Schneider, September 2026), water-cooled chillers over 1,000 tons 50 to 70 weeks and CDUs 30 to 50 weeks ([Origin Partners, May 2026](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026)), large power transformers above 160 weeks (Wood Mackenzie 2026, as reported by [pv magazine](https://pv-magazine-usa.com/2026/05/11/u-s-transformer-market-faces-severe-supply-constraints-as-lead-times-extend-to-four-years/)). - **Carrying cost.** [FRED GS10](https://fred.stlouisfed.org/series/GS10), 4.80% on 8 September 2026. Every line prints its source and whether it is a published figure, a derived split or a trade estimate. ## What is free and what is paid - **Free:** the headline, this page, and the cost-share chart in [Chillers Are the 60-Week Problem](https://open-factory.com/articles/chiller-60-week-problem/). - **Pro:** the component table per MW for air-cooled and liquid-cooled builds, by market, with lead times and carrying cost. - **Institutional:** the workbook with the split arithmetic exposed, the dashboard, quarterly updates for a year, an onboarding call and ad hoc calls. ## Articles built on it [Chillers Are the 60-Week Problem](https://open-factory.com/articles/chiller-60-week-problem/) and [The $60 Million Deposit: How Equipment Slots Became Datacenter Currency](https://open-factory.com/articles/slot-economy/). Forward assumptions on density, 800 VDC and cooling type sit on the [Datacenter Electrical Roadmap](https://open-factory.com/roadmaps/datacenter-electrical-2026-2032/) and the [Cooling Roadmap](https://open-factory.com/roadmaps/cooling-roadmap/). ## Version history - **0.1 (March 2026).** Turner & Townsend 2025 index, JLL 2026 outlook, Lead-Time Monitor September 2026 edition for lead times. ## What version 1 adds Bid-tab awards for chillers, gensets and switchgear from the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) replace the derived splits line by line as cells fill, and contributed MEP quotes from the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) enter under the cell-size rules. Each replaced line will say which version replaced it. ## Known gaps The component split below the electrical and mechanical totals is derived, not published, and is marked derived. CDU and chiller unit prices are trade estimates until awards accumulate. Turner & Townsend's index is annual; between editions the table ages and says so. --- # Model: Large-Load Cost Table URL: https://open-factory.com/models/large-load-cost-table/ Status: Compiling data Headline: 2033 realistic energization year for a new 50 MW load in Northern Virginia, Dominion Energy, from a 2026 application What a 50 MW connection costs and takes in 15 US markets: study fees, security per MW, minimum bills, contract terms, exit fees and realistic energization year, from tariffs and dockets. ## What the headline is **2033** is the year a 50 MW load applying to Dominion Energy Virginia in 2026 can realistically expect to energize. The basis is JLL's [Global Data Center Outlook 2026](https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-reports/26-research-global-data-center-outlook-new.pdf) (p.18), which puts Northern Virginia at about seven years to power for a new 50 MW site, and Dominion's own load pipeline of about 70 GW against a 24.6 GW system peak as reported by [Utility Dive](https://www.utilitydive.com/news/states-ferc-large-load-tariff-doe-christie/816672/) in February 2026. The shortest markets in the table (Houston, Northern Indiana, Salt Lake City) are 2028 to 2029. The full column with its basis is in [energization-years.csv](https://open-factory.com/articles/what-50mw-costs-15-markets/). The headline is the longest realistic energization year among the 15 markets. It moves when a utility publishes a schedule that changes it. ## What it is compiled from Version 0.1 is tariffs, orders and ISO documents. No developer interviews yet. - **Virginia.** Dominion Schedule GS-5 approved 25 November 2025 ([PUR-2025-00058](https://www.scc.virginia.gov/pages/Case-Information)): 25 MW threshold, 14-year term with a four-year ramp, 85% minimum for distribution and transmission and 60% for generation, $1.5 million per MW collateral. Mandatory CIAC for direct-connect transmission ordered 31 July 2026 ([PUR-2026-00056 final order](https://www.scc.virginia.gov/docketsearch/DOCS/8dt001!.PDF)). Terms as summarised in the [LBNL and Brattle brief](https://eta-publications.lbl.gov/sites/default/files/2026-08/rate_designs_large_loads_2026.08.10.pdf), pp.4 to 7. - **Ohio.** AEP Ohio Schedule DCT, [approved 9 July 2025](https://www.aep.com/news/stories/view/10327/): $50,000 study fee for 50 to 100 MW, 85% minimum, ramp plus eight years, 36-month exit fee, collateral of 50% of term minimum charges. [17,861 MW under contract](https://www.aepohio.com/company/news/view?releaseID=10753) at February 2026 against an 8,000 to 10,500 MW peak. [Tariff page](https://www.aepohio.com/company/about/rates/data-center-tariff/). - **Georgia.** PSC rule of [23 January 2025](https://psc.ga.gov/site/assets/files/8617/media_advisory_data_centers_rule_1-23-2025.pdf): 100 MW and above, contracts up to 15 years, upstream cost recovery. - **Texas.** SB6 (signed 20 June 2025, 75 MW threshold); PUCT proposed [16 TAC 25.194](https://www.gtlaw.com/en/insights/2026/3/texas-senate-bill-6-update-what-data-centers-large-load-customers-should-know-about-proposed-interconnection-standards) (12 March 2026): $50,000 per MW security plus $50,000 per MW fee; ERCOT [Batch Zero approved 18 June 2026](https://www.ercot.com/news/release/06182026-puct-approves-ercots) with 438 GW tracked; the Governor's [3 August 2026 audit directive](https://gov.texas.gov/uploads/files/press/Thomas_Gleeson_Pablo_Vegas_Data_Centers_Directive_Letter_to_PUCT_ERCOT_August_2026_.pdf) at 474 GW. - **Oregon, Colorado, Indiana, Carolinas, California.** [PUC Order 26-154](https://apps.puc.state.or.us/orders/2026ords/26-154.pdf) (PGE Schedule 96, 10 to 30 years); Xcel's [April 2026 filing](https://coloradosun.com/2026/04/02/data-centers-xcel-energy-colorado-rate-tariff-consumers/); I&M's [Cause 46097 settlement](https://www.aep.com/news/stories/view/10037/); Duke's [HLF proposal](https://www.latitudemedia.com/news/duke-energys-light-touch-large-load-tariff-could-be-a-problem/); CPUC [D.25-07-039 and PG&E Rule 30](https://www.pge.com/tariffs/assets/pdf/tariffbook/ELEC_RULES_30.pdf). - **Capacity and rates.** PJM BRA clearing prices from the [2028/29 results report](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2028-2029/2028-2029-bra-results-report.pdf); EIA [Table 5.6.A](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_6_a) industrial rates (Texas 6.58 cents per kWh, Virginia 9.31, Ohio 9.89, June 2026). Every cell prints its verification status: VERIFIED against the primary document, or SECONDARY with the outlet named. ## What is free and what is paid - **Free:** the 15-market table as published in [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/), this page, and the [Large-Load Interconnection Cost](https://open-factory.com/tools/interconnection-cost/) calculator with default inputs. - **Pro:** the table's history as dockets move, the 20 MW and 100 MW scale factors, and the minimum-bill cash schedules by market. - **Institutional:** the workbook and dashboard, docket alerts when a cell changes, custom markets, quarterly updates for a year, an onboarding call and ad hoc calls. ## Articles built on it [What 50 MW Costs in 15 US Markets](https://open-factory.com/articles/what-50mw-costs-15-markets/), [The 2 MW Plant Service Upgrade Nobody Budgets Right](https://open-factory.com/articles/2mw-plant-service-upgrade/), [A Factory Owner's Guide to the PJM Capacity Charge](https://open-factory.com/articles/pjm-capacity-factory-bill/) and [Datacenter Construction Is Eating Your Electricians](https://open-factory.com/articles/electricians-wages-by-metro/). The rule calendar lives on the [Large-Load Policy Tracker](https://open-factory.com/roadmaps/large-load-policy-tracker/). ## Version history - **0.1 (August 2026).** Fifteen markets. Tariffs and orders through 20 August 2026 (PUCT good-cause exception for ERCOT). Study fees and exit-fee amounts not published for Dominion; cells say so. ## What version 1 adds Developer interviews on actual study outcomes and actual energization dates against the tariff's promise, and the utility's study-queue position where a utility publishes one. Interview counts print here only when true. ## Known gaps Dominion's study fee and exit-fee amounts were not in the sources opened. CenterPoint and ComEd security figures are implied averages from company statements (about $64,000 and $250,000 per MW), not tariff sheets, and are marked as such. Texas's $50,000 per MW security is a proposed rule and does not apply at 50 MW. The table carries the 75 MW case beside it. --- # Model: Lead-Time Monitor URL: https://open-factory.com/models/lead-time-monitor/ Status: Compiling data Headline: 50 weeks LV switchgear, published lead time, Schneider Electric Lead Times Guide, September 2026 edition Published lead times for eleven equipment classes compiled from OEM lead-time guides, Census M3 unfilled orders over shipments, ISM supplier deliveries and backlog disclosures in filings. Every cell carries its OEM and date. ## What the headline is **50 weeks** is the made-to-order lead time Schneider Electric prints for its low-voltage switchgear families (LVDO A, B, C, D and F) in the [September 2026 Lead Times Guide](https://www.se.com/us/en/download/document/Monthly_Lead/), published 2 September 2026, with each of those six rows flagged as extended against the August edition. It is the longest lead time any OEM publishes with a date on it. The rows are in [03-schneider-flags-sep-2026.csv](https://open-factory.com/data/lead-time-monitor-launch/03-schneider-flags-sep-2026.csv) beside the launch piece. The headline is always the longest OEM-published figure, never a survey number. When Siemens, Eaton or ABB publish a longer dated schedule, the headline moves to it and says so. ## What it is compiled from Version 0.1 is public records only. No panel, no survey, no calls. - **OEM-published schedules.** [Schneider Electric Lead Times Guide](https://www.se.com/us/en/download/document/Monthly_Lead/), monthly, 124 product rows parsed to CSV (switchboards 18 to 34 weeks, LV switchgear 45 to 50, MV 15 to more than 40, MCC 6 to 27, busway 24 to 40, ATS 6 to 27, drives 4 to 32, September 2026). Siemens LV schedules as republished by [VAWN](https://www.usevawn.com/resources/electrical-equipment-lead-times/) (LV switchgear 54 weeks, switchboards 32 to 41, busduct 27, August 2026). Rockwell publishes [no product-level figure](https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html); that cell reads "none published". - **Backlog leg.** [Census M3](https://www.census.gov/manufacturing/m3/) unfilled orders over monthly shipments for NAICS 3353 (Census code 35C): 5.67 months in July 2026 against about 3.2 in 2017 to 2019. Also turbines and generators (TGP), 7.35 months. - **Deliveries leg.** [ISM Manufacturing Report On Business](https://www.ismworld.org/supply-management-news-and-reports/reports/ism-report-on-business/): supplier deliveries index 59.3 and the capital expenditure lead time of 171 days, August 2026. - **Price leg.** BLS PPI [WPU117409](https://data.bls.gov/timeseries/WPU117409) (transformers, 381.8 in July 2026) and [WPU1175](https://data.bls.gov/timeseries/WPU1175) (switchgear, 419.3), rebased to January 2020. - **Filings leg.** Backlog in dollars from [Powell 10-Q](https://www.sec.gov/Archives/edgar/data/0000080420/000008042026000107/powl-20260630.htm) ($2.4B, $1.3B shipping within 12 months, June 30 2026), [Caterpillar 10-Q](https://www.sec.gov/Archives/edgar/data/18230/000001823026000046/cat-20260630.htm) ($72.1B, $29.2B beyond 12 months), [GE Vernova 8-K](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm), [Eaton 8-K](https://www.sec.gov/Archives/edgar/data/1551182/000155118226000027/etn06302026exhibit99.htm), [Vertiv 8-K](https://www.sec.gov/Archives/edgar/data/1674101/000167410126000006/exhibit991vrt02112026.htm) ($15.0B, December 2025, not disclosed since), [Generac 8-K](https://www.sec.gov/Archives/edgar/data/1474735/000143774926024731/ex_994826.htm), and the [Cummins Q4 2025 call](https://www.fool.com/earnings/call-transcripts/2026/02/05/cummins-cmi-q4-2025-earnings-call-transcript/) ("orders well into 2028"). - **Third-party surveys, labelled as theirs.** Wood Mackenzie (128 weeks power transformers, 144 GSU, 44 MV switchgear, Q2 2025; above 160 weeks in 2026) as reported by [POWER](https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/) and [pv magazine](https://pv-magazine-usa.com/2026/05/11/u-s-transformer-market-faces-severe-supply-constraints-as-lead-times-extend-to-four-years/); [Terrapin Construction Group](https://terrapincg.com/news/switchgear-transformer-generator-lead-times-2026) (June 2026); [Origin Partners](https://originpartners.io/blog/equipment-lead-times-tracker-q2-2026) (May 2026). Where two disagree (gensets: Terrapin 50 to 110 weeks, Origin 26 to 40), both are printed as a conflict. Every cell includes the document, the OEM and the date. Nothing is estimated where a published figure exists. ## What is free and what is paid - **Free:** the headline number, this page, the eleven-class range chart in [Introducing the Open Factory Lead-Time Monitor](https://open-factory.com/articles/lead-time-monitor-launch/), and the monthly index letter. - **Pro ($40 a month or $400 a year):** every class and every source, monthly, with the Schneider edition-to-edition flags and the M3, ISM and PPI series. - **Institutional (quote):** the workbook and dashboard, the full edition history (the Schneider history file starts with September 2026), the API, quarterly updates for a year, an onboarding call and ad hoc analyst calls. ## Articles built on it [Transformer Lead Times Hit 143 Weeks](https://open-factory.com/articles/transformer-160-weeks/), [Why Your Switchgear Costs 50% More Than in 2021](https://open-factory.com/articles/switchgear-50-percent/), [Cummins and Caterpillar Are Sold Out Through 2028](https://open-factory.com/articles/gensets-sold-out-2028/), [Chillers Are the 60-Week Problem](https://open-factory.com/articles/chiller-60-week-problem/), [Vertiv Stopped Reporting Its Backlog](https://open-factory.com/articles/vertiv-backlog-estimate/) and [One Steel Mill Gates Every US Transformer](https://open-factory.com/articles/butler-works-goes/) all draw their lead-time and backlog figures from this table. Charts read `Source: Open Factory Lead-Time Monitor`. ## Version history - **0.1 (September 2026).** Eleven classes. Schneider September 2026 edition, M3 through July 2026, PPI through July 2026, ISM through August 2026, filings through Q2 2026. Public records only. ## What version 1 adds A distributor and EPC panel reporting quoted versus shipped lead time by OEM and class, so the Monitor shows what a buyer was told and what arrived. Panel size and response rate will print on this page only when they are true, per the [methodology](https://open-factory.com/methodology/). Also planned: Eaton and ABB dated schedules if they publish them, and a chiller class fed by the OEM order releases Munters, Trane and Carrier already issue. ## Known gaps PLC CPUs have no published lead time from any OEM; the cell says so. Large power transformers have no OEM schedule. The class is carried on Wood Mackenzie's survey and the Siemens Energy statement of "up to five years" (February 2024). STS and PDUs have no public series at all. The Schneider guide is the OEM's planning document and says it is "not to be relied upon for actual orders". We print it as published and say whose it is. --- # Model: OEM Disclosure Scorecard URL: https://open-factory.com/models/oem-disclosure-scorecard/ Status: Compiling data Headline: 12 of 90 Cells scored Yes across 18 OEMs and five disclosure criteria, filings and OEM sites checked September 2026 Does the OEM publish lead times, list prices, backlog in dollars, capacity in units and delivery performance? Scored from observable facts, source beside every cell, right of reply before publication. ## What the headline is **10 of 18** listed electrical, power and cooling OEMs put a dollar figure on their backlog in the Q2 2026 filing season: Powell ($2.4B), Caterpillar ($72.1B), GE Vernova Electrification ($40.6B), Siemens Energy Grid Technologies (EUR 51bn), Hitachi Energy (10.3 trillion yen), ABB Electrification ($13.7B), Trane ($12.1B), Johnson Controls ($21.0B), Eaton (about $24.1B, in the 10-Q text rather than the release) and Generac (about $1.6B, data center products only). Eight did not: Vertiv (stopped after Q4 2025), Schneider Electric ("record backlog", no number), Cummins, Carrier, nVent (Q1 only), Modine, Hubbell and Atkore. The rows, with the filing for each, are in [book-to-bill-q2-2026.csv](https://open-factory.com/articles/book-to-bill-league/). The headline is a count of filings, not a judgment. It changes each filing season. ## What it is compiled from Version 0.1 is observation only. Five cells per OEM: 1. **Publishes a dated lead-time schedule.** Yes: [Schneider Electric](https://www.se.com/us/en/download/document/Monthly_Lead/) (monthly, 124 rows). Partial: Siemens, whose LV bulletins circulate through distributors and are republished by [VAWN](https://www.usevawn.com/resources/electrical-equipment-lead-times/). No: [Rockwell](https://www.rockwellautomation.com/en-us/company/about-us/legal-notices/supply-chain-lead-times.html) and every other OEM in the set. 2. **Publishes list prices.** Checked against the OEM's own site. In the adjacent machine-tool set, Haas is the only builder with a [public price list](https://www.haascnc.com/shop/category/pricelist.html) (VF-2 $70,995, verified 31 August 2026); no OEM in the electrical set publishes switchgear, transformer or UPS list prices. 3. **Reports backlog in dollars.** From the Q2 2026 filings listed above and in the [Book-to-Bill League Table](https://open-factory.com/models/book-to-bill-table/). 4. **Reports capacity in units.** Yes: Siemens Energy's Charlotte plant, 24 large power transformers a year rising to 57 ([press release, February 2024](https://www.siemens-energy.com/us/en/home/press-releases/siemens_energy_addresses_shortage_US_powertransformers_invests_new_factory.html)). Partial: GE Vernova gas turbines in GW ([Q2 2026 8-K](https://www.sec.gov/Archives/edgar/data/1996810/000199681026000147/gevpressrelease2q26.htm)). Most expansions are announced in dollars and jobs only ([Eaton, Jonesville SC, $340M](https://www.utilitydive.com/news/eaton-transformer-production-shortage-investment/739932/); [Hitachi Energy, $1B](https://www.utilitydive.com/news/hitachi-unveils-1b-grid-manufacturing-investment-including-virginia-trans/759219/)). 5. **Reports delivery performance against quote.** No OEM in the set publishes on-time delivery. Every cell reads no; this is the column version 1 fills from the panel. ## What is free and what is paid Everything on this model is free: the headline, the five-column table, the source beside every cell and the OEM replies. It is the rating that costs nothing and that the rated parties will check. ## Articles built on it The full scorecard piece is in preparation. The backlog column is already published through [Book-to-Bill League Table, Q2 2026](https://open-factory.com/articles/book-to-bill-league/) and [Vertiv Stopped Reporting Its Backlog](https://open-factory.com/articles/vertiv-backlog-estimate/). The lead-time column through [Introducing the Open Factory Lead-Time Monitor](https://open-factory.com/articles/lead-time-monitor-launch/). ## Version history - **0.1 (September 2026).** Eighteen OEMs, five observable cells, Q2 2026 filings and OEM sites as of 9 September 2026. Right-of-reply letters go out with the first full scorecard piece. ## What version 1 adds A sixth column, delivered versus quoted lead time by OEM, from the distributor and EPC panel behind the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/) version 1. It prints only from cells that meet the [methodology](https://open-factory.com/methodology/) size rules, and every OEM sees its number before it runs. ## Known gaps List-price checks cover public web pages only; dealer price books behind logins are not counted either way. Hitachi's backlog is a segment figure from a Japanese parent's presentation and is scored yes. Generac's dollar figure covers one product line and is scored partial. --- # Model: Public Bid-Tab Price Book URL: https://open-factory.com/models/bid-tab-price-book/ Status: Compiling data Headline: $69/kVA Pad-mount 3-phase 225 to 500 kVA, median of 60 public awards, 11 buyers in 11 states, 2024 to 2026 Transaction prices from public utility, municipal, state and federal procurement awards for transformers, MV switchgear, gensets and chillers, normalized to $/kVA, $/section, $/kW and $/ton bands with n, region and award date. ## Where it stands Version 0.1 is built: **2,712 award lines** with a unit price and a document URL from 141 public buyers in 44 states, 2016 to 2026, plus 2,567 vendor bid rows from bid tabulations, reduced to **51 published bands** (n of 5 or more) across pad-mount and substation transformers, MV switchgear, gensets and chillers, in an all-years and a 2024-onward window. Plausibility bounds moved 166 rows to an excluded file with the reason on each. Austin Energy supplies 1,665 of the pad-mount rows, so every band prints its top-buyer share. The method, the portals and the biases are in [Introducing the Public Bid-Tab Price Book](https://open-factory.com/articles/bid-tab-price-book-launch/); [Quote Check](https://open-factory.com/tools/quote-check/) reads the bands directly. The first normalized rows are already public beside [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/): the [Riverside Public Utilities 15 kV switchgear bids](https://riversideca.legistar1.com/riversideca/attachments/28dbba7d-a43b-456d-99cc-3148790fb517.pdf) 2018 to 2025 (four bidders on SUB-806, low bid $1,343,878 against a $1,450,000 engineer's estimate) and the [City of Dover PME-9 pad-mounted switchgear bid tabulation](https://www.cityofdover.gov/media/Purchasing/Tabulations/FY%2020%20Info/20-0007CS%20Bid%20Tab.pdf) (five unit bids from $20,703 to $28,235, November 2019). ## What it is compiled from - **Municipal utilities.** Board agendas and bid-award memoranda through Legistar (Riverside, Santa Clara, Dover and others), which publish bidders, amounts, engineer's estimates and the specification. - **Cities and states.** The [City of Austin open-data portal](https://data.austintexas.gov/) purchase-order lines; BidNet, Bonfire and PlanetBids postings; state procurement portals. - **Federal.** [USAspending.gov](https://www.usaspending.gov/) awards by NAICS 3353 (electrical equipment), 335312 and 333611 where the description names a transformer, switchgear lineup, genset or chiller. - **FOIA.** Requests to public utilities for bid tabulations not posted online, logged with the date filed and the date answered. Every row keeps its link. Every band prints n, the region, the award-date range and the unit convention. ## What is free and what is paid - **Free:** bands by class with n and date, this page, and the [Quote Check](https://open-factory.com/tools/quote-check/) and [Transformer Sizing and Cost](https://open-factory.com/tools/transformer-sizing-cost/) calculators with default inputs once bands exist. - **Pro:** bands by region and award-year, and the full award table with specs. - **Institutional:** row-level data with documents, the API, quarterly updates for a year, an onboarding call and ad hoc calls. ## Articles built on it [Reading a Switchgear Quote: The Eleven Line Items That Hide the Schedule](https://open-factory.com/articles/switchgear-quote-anatomy/), [How to Write an RFQ for Capital Equipment in 2026](https://open-factory.com/articles/rfq-capital-equipment/) and [What a Haas VF-2 Really Costs in 2026](https://open-factory.com/articles/haas-vf2-true-cost/) use the first public award and list-price rows. Index-level price context comes from the PPI series in the [Lead-Time Monitor](https://open-factory.com/models/lead-time-monitor/). ## Version history - **0.1 (September 2026, collecting).** Raw pulls from Austin, USAspending and Legistar; first normalized switchgear rows published with articles. Launch threshold set at 300 awards across six classes. ## What version 1 adds Contributed quotes from the [Equipment Price Benchmark](https://open-factory.com/models/equipment-price-benchmark/) merge into the same bands under the cell-size rules (five contributors, eight quotes, no contributor above 40%, all quotes three months old or older), and machine-tool awards from university and community-college procurement open a fifth and sixth class. ## Known gaps Federal award descriptions often bundle equipment with installation. Those rows are excluded from unit bands until the split is documented. Legistar coverage depends on which utilities post attachments. Sales tax treatment differs by state and is stated per row. --- # Model: Equipment Price Benchmark URL: https://open-factory.com/models/equipment-price-benchmark/ Status: Coming Headline: Contribution form open no distribution is printed until a cell meets the size rules; contributions received so far: none Contributed quotes and invoices for machine tools, robots and electrical gear, normalized to spec. Launches as a contribution form and a promise; distributions publish only when a cell has five or more contributors and eight or more quotes. ## Where it stands The Equipment Price Benchmark is a **contribution form and a promise**, nothing more. Zero contributions have been received. No count, no distribution and no band will print until a cell holds five or more contributors and eight or more quotes, with no contributor above 40% of the cell and every quote at least three months old. Those rules are on the [methodology page](https://open-factory.com/methodology/) and will not be relaxed to get a number out sooner. The one public reference point today is Haas, the only US machine-tool builder with a published list: VF-2 $70,995, VF-2YT $78,995, VF-2SS $83,995, VF-3 $87,995, verified 31 August 2026 via [MachineToolIndex](https://machinetoolindex.com/machines/haas-vf-2) against the [Haas price list](https://www.haascnc.com/shop/category/pricelist.html). What the benchmark will show is how far a delivered, optioned, rigged machine lands from that list, and what Mazak, Okuma and DMG Mori machines with no public list actually sell for. ## How a quote becomes a cell 1. **Contribution.** A quote or invoice, uploaded or transcribed, with the configuration fields for its class (spindle speed, travels, tool changer, probing, automation for a machining center; payload, reach, controller for a robot; kVA, voltage, impedance for a transformer; ampacity, sections, breaker frame for a switchboard; kW, fuel, tier for a genset). The contributor confirms there is no confidentiality clause. 2. **Normalization.** The price is reduced to the class's unit convention ($/machine at base configuration with options listed, $/kVA, $/section, $/kW) and tagged with region and quote month. 3. **Cell test.** A cell is a class, a configuration band, a region and a rolling four-quarter window. It publishes only when it meets the size rules. 4. **Publication.** Median, interquartile range, n and contributor count. Nothing else. ## What is free and what is paid - **Contribute to see:** contributors see the distributions for their cells. - **Pro:** all published cells. - **Institutional:** cell statistics as CSV and API, with quarterly updates for a year, an onboarding call and ad hoc calls. Never rows. ## Articles that set it up [What a Haas VF-2 Really Costs in 2026](https://open-factory.com/articles/haas-vf2-true-cost/) is the first cell's public anchor. [Reading a Switchgear Quote](https://open-factory.com/articles/switchgear-quote-anatomy/) and [How to Write an RFQ for Capital Equipment in 2026](https://open-factory.com/articles/rfq-capital-equipment/) define the configuration fields for the electrical cells. Public awards from the [Public Bid-Tab Price Book](https://open-factory.com/models/bid-tab-price-book/) anchor each cell before any contribution arrives. ## Version history - **0.0 (September 2026).** Contribution form open. No contributions. Cell rules published. ## What version 1 adds Version 1 exists when the first cell passes the size test. It will print the cell, its n and its contributor count, and merge into the Bid-Tab Price Book bands where the classes overlap. A lawyer's memo on information exchange and defamation precedes the first publication. ## Known gaps Everything. This page exists so the rules are public before the data is. ---