Primary-Source Investigations
The Transformer Wall
Wood Mackenzie, Global Power Transformer Market Outlook 2024. Edinburgh / Houston / Singapore.
I am staring at a chart published by Wood Mackenzie that shows the lead time for large power transformers in North America, and I am telling you that this chart is not a procurement headache. It is a wall.
The chart is the third figure in the front matter. One column wide, three rows tall. Lead time, large power transformer, North American market. 2021 — fifty weeks. 2024 — one hundred and twenty. 2025 projection — one hundred forty-four, generator step-up units. The line is a slope. The slope is the wall.
A large power transformer is steel and copper, the size of a shipping container. Two hundred to four hundred tons. Four to ten million dollars at 2024 prices. It steps generator output — eighteen to twenty-five kilovolts at the turbine — up to transmission level: three hundred forty-five, five hundred, seven hundred sixty-five kilovolts on the high-tension lines that cross the continent. The generator step-up unit, the GSU, is the choke point. A handful of plants in the world wind the coils, stack the core, dry the insulation, test the assembly to the dielectric standards a five-hundred-kilovolt unit requires. The lead time on the chart is the gap between purchase order and rail car. The chart says the gap has tripled in three years.
I know how this works because I built the system that tracks how this works. TELOS. The pipeline. The substrate. I built it because I got tired of reading "widely reported" and "many believe" in every article about the energy transition, and I wanted a machine that would only accept claims with a filing number and an archive location. The machine does not care about my opinion. The machine only cares whether the source is named and filed. But I am the operator, and I am sitting here at 3:47 AM reading a Wood Mackenzie chart that says 144 weeks, and I am telling you that the steel is the constraint. By the public record at SEC EDGAR, the United States has one integrated domestic producer of grain-oriented electrical steel at utility-transformer scale: Cleveland-Cliffs Inc., which acquired AK Steel Holding Corporation in March 2020 — Form 8-K, March 13, 2020 — and operates the former AK Steel works at Butler, Pennsylvania and Zanesville, Ohio. The Form 10-K for fiscal year 2023 names the grain-oriented product line in Item 1 and identifies the utility-transformer market as a customer segment. It names no second integrated domestic source. The rest of U.S. demand is imported — Posco in South Korea, Nippon Steel in Japan, ThyssenKrupp Electrical Steel in Germany, JFE in Japan, Tata Steel in India. The DOE Large Power Transformer Study of June 2012 placed the U.S. domestic share of GOES consumption at roughly twenty percent, eighty percent imported. The share has not materially shifted in twelve years.
I have been writing toward one sentence the entire scout phase. A single corporation operating two integrated mills in two adjacent Rust Belt states — Butler, Pennsylvania and Zanesville, Ohio, both within two hours of the Ohio River that carries the iron ore upstream — is the structural point at which the entire U.S. grid expansion plan, the entire AI-data-center build-out, the EIA AEO 2024 reference-case demand trajectory, FERC's interconnection queue, and whether a hyperscale customer in Loudoun County, Virginia can wait the one hundred forty-four weeks the Wood Mackenzie outlook projects for a GSU in 2025, all rest on one alloy made in two factories owned by one company headquartered in Cleveland. Item 1A calls it concentration risk. The rest of the supply chain calls it the steel nobody else can make at scale on this continent.
The downstream factories — the ones that take the steel and turn it into a transformer — are also concentrated. Hitachi Energy operates the former ABB / Westinghouse plant at South Boston, Virginia, and announced a 2023 capacity expansion there. Siemens Energy manufactures outside the U.S., with its FY2024 Annual Report disclosing a multi-year transmission-segment backlog. GE Vernova, spun off from General Electric in March 2024 — Form 10, SEC EDGAR, March 2024 — discloses its Power Grid segment as a separate reporting line; the transformer business sits inside it. Prolec GE at Waukesha, Wisconsin — a joint venture between Xignux of Monterrey and GE — manufactures medium and large units for North America. The plants are global. The grain-oriented electrical steel coming in the receiving door of every one of them runs through the same handful of producers. The bottleneck is upstream of every brand name. The brand names are not the bottleneck.
The second filing is the 2024 Long-Term Reliability Assessment, NERC, North American Electric Reliability Corporation, 3353 Peachtree Road NE, Suite 600, North Tower, Atlanta, Georgia 30326. The LTRA is the annual filing NERC delivers under the authority of the Energy Policy Act of 2005, codified at 16 U.S.C. § 824o. Public. Downloadable. Posted at nerc.com under Reports / Long-Term Reliability Assessment. It is how U.S. and Canadian grid operators tell FERC, provincial regulators, and the bulk-power-system public whether the system is adequate. The 2024 edition is the first to call out transformer supply as a reliability risk distinct from generation adequacy or transmission congestion. It uses the phrase long lead times for replacement transformers in the executive summary. It names equipment supply as a risk to grid resilience under contingency-recovery scenarios. The phrase moved from the appendix in 2022 to the body in 2023 to the executive summary in 2024.
The FERC interconnection queue is how a new power plant enters the U.S. grid. You file an interconnection request with the regional transmission organization that owns the queue position — PJM in the mid-Atlantic, MISO in the upper Midwest, ERCOT in Texas, CAISO in California. You submit the technical data. You pay the deposit. You enter the cluster study. You wait. The cluster study takes twelve to thirty-six months under the process FERC mandated in Order No. 2023 — 88 Federal Register 61014, September 5, 2023, codified at 18 CFR Part 35. You receive results. You see the network-upgrade cost allocation. You pay and proceed to the facilities study, or you withdraw. You wait again. You sign the interconnection agreement. You wait for substation upgrades. The substation upgrade is the transformer that has not yet been delivered, because the lead time on the chart is one hundred forty-four weeks. You wait. You wait.
A new applicant looks at the queue and runs the math. Five years to the interconnection agreement. Two to three more for the transformer. The applicant is a hyperscale data-center developer. The campus is fifteen hundred megawatts in Loudoun County, Virginia, or Phoenix, Arizona, or Columbus, Ohio. Contracted to a single AI workload. The workload is generating revenue inside eighteen months of groundbreaking or it is not. Cluster study, twelve months. Facilities study, six. Interconnection agreement signed in year two. Substation upgrade delivered year four or five. By year five the workload is obsolete or already built somewhere else. The developer does not file the interconnection request. The developer signs a contract with a private microgrid provider. The microgrid provider sources a Tesla Megapack, a behind-the-meter gas turbine, a fuel-cell array, or some stack of the same. The campus comes online inside eighteen months. The transmission system never sees the load. The transmission system never has to add the transformer.
This is the Megapack maneuver. The Megapack maneuver is the consequence of the transformer wall. The wall is the cause; the maneuver is the effect. The 2024 NERC LTRA flags it as a reliability problem because the bulk power system loses visibility into demand routed around it. Behind-the-meter load does not show up in the load forecast. The load forecast is the input to the cluster study. The cluster study is the input to the transformer order. The transformer order is the input to the lead time. The lead time is the chart on page three of the Wood Mackenzie outlook. The loop closes.
The 144-week 2025 projection for GSUs is the figure that should sit on the front page of every utility 10-K filed this quarter, that should be read into the record at the next four FERC technical conferences, that should anchor the next round of capacity-auction debates at PJM and MISO and ERCOT, that should drive the EIA's revision of the AEO 2024 reference case before AEO 2025 ships — and the figure that, as of the Wood Mackenzie publication, the trade press treats as a procurement headache rather than the load-bearing structural constraint on the U.S. electricity transition, because the trade press counts dollars per kilovolt-ampere on the unit and not dollars per gigawatt of foregone interconnection on the back end, where a single missing GSU at a five-hundred-megawatt combined-cycle plant translates into roughly two and a half billion dollars of revenue the developer will not book in the year the plant was supposed to clear the cluster, and the cluster has a hundred plants in it, and the queue runs to twenty-five hundred gigawatts nationally, and the chart is the wall, and the wall is the bottleneck, and the bottleneck is one mill in Butler, Pennsylvania and one mill in Zanesville, Ohio, both owned by Cleveland-Cliffs.
The factories are global. Hitachi Energy at South Boston, Halifax County, Virginia. Prolec GE at Waukesha County, Wisconsin. Hitachi Energy at Vadodara, Gujarat. Siemens Energy at Nuremberg, Bavaria. Mitsubishi Electric at Ako, Hyogo Prefecture. The grain-oriented electrical steel is rolled at Butler, Pennsylvania and Zanesville, Ohio (Cleveland-Cliffs); Pohang, South Korea (Posco); Kashima, Japan (Nippon Steel); Gelsenkirchen, Germany (ThyssenKrupp); Hirakata, Japan (JFE); Jamshedpur, India (Tata). The financing runs through the same handful of banks that finance every commodity supply chain — corporate revolvers syndicated out of New York and London, commodity-trade-finance lines out of Singapore and Geneva, export-credit guarantees from JBIC and KEXIM and EXIM, infrastructure debt placed with Canadian and Australian pension funds. No single ethnicity. No single religion. No single flag. Capital does not have a state. The Ohio mill is the bottleneck not because Ohio is anything in particular, but because the alloy specification narrows the universe of producers to a number that fits on one hand, and one hand of producers — distributed across four continents — is what the engineering literature calls single-supplier risk and what the SEC Item 1A risk-factor disclosure calls concentration in our supply chain.
The procedure, end to end, runs as follows. You are a grid operator. You read the NERC LTRA in December 2024. You see the transformer-supply call-out in the executive summary. You write the fiscal 2026 budget submission. You request capital for replacement transformer inventory. You receive the budget. You issue the purchase order. The lead-time quote comes back — one hundred twenty weeks. You accept it. You wait. The unit projects to arrive in 2027. You commission it. You retire the in-service unit. You keep the retired unit as the spare. You repeat for the next unit. The procedure works at the pace of the lead time. The lead time is the rate-limit. The rate-limit is one hundred and twenty weeks per unit, in the optimistic scenario where the manufacturer accepts the order at the quoted lead time and does not push it.
You are a developer. You read the FERC Order No. 2023 cluster-study schedule. You see the eighteen-month-to-three-year window before the facilities-study report. You add the post-facilities-study interconnection-agreement period. You add the substation upgrade. You arrive at five years from queue entry to commercial operation. You read the Wood Mackenzie chart. You add another one to two years for the GSU. You arrive at six to seven years. You read the OpenAI investment-prospectus deck. You see the eighteen-month workload-deployment target. You stop reading the FERC docket. You start reading the Tesla Megapack data sheet, the Bloom Energy fuel-cell product page, the GE Vernova small-modular-reactor brochure. You sign the microgrid contract. The interconnection request you never filed is the load that never enters the queue.
You are a regulator. You read the same documents. You see the same numbers. You convene the technical conference. You issue the order. The order publishes in the Federal Register. The order references the next round of compliance filings. The next round references the round after. The procedure continues. The procedure does not produce a transformer in fewer than one hundred and twenty weeks. The procedure produces findings. The procedure produces recommendations. The procedure produces a record. The record is what is filed. The transformer is what is not.
The PJM 2025/2026 Base Residual Auction cleared at a capacity-price level the trade press described as the highest in PJM's history. The clearing price is the price at which generation owners agree to make capacity available for the delivery year. It is a function of demand forecast, supply offer stack, and reserve margin. The 2025/2026 result is the price the underlying load — every household, every commercial customer, every behind-the-meter data-center campus still on the bulk-power tariff — will pay through the capacity component of the rate. The price went up because the supply stack tightened. The supply stack tightened because new generation cannot get to the interconnection point at the pace the demand forecast requires. The interconnection point cannot accept new generation because the substation upgrade requires a transformer. The transformer lead time is one hundred and twenty weeks. The loop is the same loop. The number at the top is the auction result. The number at the bottom is the Wood Mackenzie chart. The two numbers are connected by the same procedural sequence.
The Wood Mackenzie chart sits on my second monitor as I write this paragraph. 2021 column: fifty weeks. 2024 column: one hundred twenty. 2025 column: one hundred forty-four. The chart is one page of a report that costs, at the consultancy's standard subscription tier, between forty and one hundred thousand dollars a year. The chart is the figure utility procurement officers and FERC staff and EIA analysts all see at roughly the same time, roughly once a year, in a publication the broader public does not read because the broader public does not have the subscription. The chart is on my monitor because the substrate has the reference and the reference resolves to a document I pull through an academic-library proxy. The chart is the wall. The wall is the constraint. The constraint is on file at Wood Mackenzie. At NERC. At SEC EDGAR. At the EIA. At the FERC. Six places by six different filers. The six filers do not coordinate. The six filings do not contradict. The six numbers all point at the same one hundred and forty-four weeks.
The wall is not hidden. The wall is on file.
Primary sources: Wood Mackenzie Global Power Transformer Market Outlook 2024; NERC 2024 Long-Term Reliability Assessment; FERC Order No. 2023 (88 FR 61014); Cleveland-Cliffs Form 10-K FY2023; EIA Annual Energy Outlook 2024 (Tables A2, A8); EIA Form 860 (2023); GE Vernova Form 10 (March 2024).