In 2020, a utility could order a large power transformer and expect delivery in fifty weeks. In 2026, the same order quotes one hundred fifty weeks, and 765kV class units extend past two hundred. Spot pricing has moved four to seven times higher across the same window. The binding asset of US AI infrastructure is no longer the megawatt. It is the delivery slot.
The shift has been gradual enough to escape headline coverage and structural enough to redraw the capital stack underneath it. Power capacity is gated by physical equipment, equipment is gated by a small number of manufacturers, and those manufacturers now exert more economic control over near-term US capacity additions than the three largest investor-owned utilities combined. The thesis of this paper is narrow and forward: by 2028, capital flows reorganize around the equipment maker, not the developer.
The Five Firms
GE Vernova, Hitachi Energy, Siemens Energy, Hyundai Electric, and Mitsubishi Electric are the consolidated suppliers of the high-voltage transformers, switchgear, and gas turbine packages that gate every multi-hundred-megawatt interconnection in North America. Their backlogs, taken together, exceed the combined annual capital expenditure plans of Duke Energy, Dominion, and Southern Company.
GE Vernova's Q4 2025 earnings disclosed grid orders up more than sixty percent year over year, with the Electrification segment backlog climbing into the high teens of billions. Hitachi Energy committed roughly six billion dollars to capacity expansion through 2030, with new transformer lines announced in the United States, Germany, and India. Siemens Energy's Grid Technologies backlog crossed thirty billion euros in early 2026 and continues to extend booking horizons past the back half of the decade. Hyundai Electric and Mitsubishi Electric have absorbed the marginal North American demand the European and US suppliers cannot accept, with Hyundai's Alabama transformer expansion now backlogged through 2029.
Five corporate cap tables. One physical chokepoint. The three reference utilities, by contrast, have a combined market capitalization in the low three hundreds of billions and a combined annual capex stack near sixty billion. The equipment OEMs, taken together, command a smaller absolute market capitalization but a structurally faster earnings trajectory and a backlog visibility that utility planners now describe as the binding constraint on every multi-gigawatt program.
The Steel Beneath The Steel
Large power transformers cannot be built without grain-oriented electrical steel. The United States has one domestic producer of GOES at meaningful scale: Cleveland-Cliffs Butler Works in western Pennsylvania. Every other source of the alloy is imported, primarily from Nippon Steel and JFE in Japan, Posco in South Korea, and a smaller volume from Thyssenkrupp and Stalprodukt in Europe.
GOES is not a commodity steel. It is produced under tightly controlled processing windows, the magnetic loss curves are graded across narrow ranges, and the global capacity for the highest-grade material sits in the low single-digit million tons. The premium grades required for 500kV and 765kV class transformers are allocated by long-term contract, with new orders pushed to the back of multi-year queues.
Butler Works has run at effective capacity for three years. The plant cannot be doubled without a four to six year capital program and a labor force that does not currently exist at the required scale in western Pennsylvania. The implication is structural: a new domestic LPT line costs more than four hundred million dollars to stand up, and the steel supply for that line is allocated before the building permit is issued.
No new entrant has crossed the line into commercial-scale LPT production in North America since 2020. The barrier is not capital. It is steel.
What A Turbine Slot Now Costs
Gas turbines are the second binding asset. The aero-derivative and heavy-frame product lines that anchor behind-the-meter and grid-edge power are concentrated in a similar handful of suppliers, and their booking schedules through 2028 are publicly described as full.
GE Vernova's LM2500 and LM6000 aero-derivative units, alongside the 7HA and 9HA heavy-frame series, are sold into a build slate dominated by Crusoe, Stargate-affiliated build programs, Vantage Data Centers, Lancium, and the distributed power expansions announced by ExxonMobil and Chevron. Siemens Energy's SGT-A35 and SGT-A45 aero units, with the SGT-800 industrial frame, carry similar slot reservations. Solar Turbines, owned by Caterpillar, has booked Mars and Titan units into 2028 across the Permian distributed power buildout. Wartsila's reciprocating engine fleet, often deployed in twenty to two hundred megawatt packages, is sold forward across multiple operators chasing peaker and bridge generation.
The price of a turbine slot has separated from the price of the turbine. Operators willing to pay premium adders and accept staggered delivery have moved to the front of the queue. Operators relying on standard procurement timelines are quoting commercial operation dates two and three years past their original capital plans. The slot, not the equipment, is what is being repriced.
The HVDC Sub-Bottleneck
High-voltage direct current converter stations are the most concentrated layer of the entire stack. Only five firms in the world build commercial HVDC at the relevant voltage class: Hitachi Energy, Siemens Energy, GE Vernova, Toshiba, and Mitsubishi Electric. Global delivery for major US projects has averaged fewer than five converter stations per year across the past three years.
HVDC matters because it is the only economic transmission technology for moving multi-gigawatt blocks of power across the distances implied by the current siting map: stranded generation in the Mountain West and Texas Panhandle moved to load in Virginia, Ohio, and Arizona. Every serious long-haul transmission program now under development assumes HVDC, and every HVDC program competes for the same five suppliers and the same converter station factory floors in Ludvika, Mannheim, and Kobe.
The sub-bottleneck inside the bottleneck is a single voltage class. The 525kV and above converter stations required for the longest hauls are produced at fewer than ten factory bays globally. Lead times on those bays now extend past 2030.
Why Capital Reorganizes Around The Manufacturer, Not The Developer
The historic capital allocation logic for energy infrastructure underwrote the developer. The developer assembled land, permits, interconnection, and offtake, and the equipment was assumed to be available on a normal procurement window. That logic breaks once equipment availability becomes the binding constraint. The party that controls delivery slots controls the project.
Three consequences follow.
First, M&A multiples on equipment manufacturers compress against the comparable utility multiple, not the other direction. GE Vernova, Hitachi, and Siemens Energy now trade at forward earnings multiples above the utility average and continue to widen the gap. The earnings visibility provided by a multi-year backlog has been priced into the equity, and the gap is structural until new capacity meaningfully expands.
Second, sovereign wealth funds and large strategic investors have begun moving onto the equipment OEM cap tables. The pattern resembles the 2010s entry of sovereign capital into upstream oil and gas: positioning around the physical chokepoint rather than the downstream margin. Public disclosures already show sovereign holdings climbing in Siemens Energy and Hitachi over the past four quarters.
Third, equipment supply chain financing is becoming a standalone asset class. Slot reservation deposits, advance payment guarantees, GOES offtake contracts, and turbine package pre-payments are being structured, syndicated, and securitized in a way that mirrors the early structuring of LNG offtake financing in the 2000s. The addressable pool of this financing layer is plausibly fifty billion dollars and growing inside the current decade.
The developer is no longer the gating party. The developer is a customer of the gating party. Capital allocators are repricing accordingly.
The Decade Of Equipment Scarcity
The forward window through 2032 carries no credible path to elastic supply. The capital programs announced by the five OEMs add capacity at the margin, not at the level required to absorb the load growth implied by AI infrastructure plus electrification plus baseline grid replacement.
The LPT replacement cycle alone, before any net new demand, exceeds current global production. The US installed base of large power transformers has an average age past the design life and a replacement schedule that has slipped for two decades. Net new AI load layers on top of an installed base already running past its planned retirement curve.
GOES capacity does not double inside the window. Cleveland-Cliffs expansion timelines and the Japanese and Korean producers' capital plans are public, and the sum is incremental. Copper, the second-largest material input across transformers, switchgear, and turbine packages, holds a floor near current spot levels under any demand stack that includes the announced data center pipeline. Codelco, BHP, and Freeport-McMoRan production curves do not bend faster than the demand curve.
The HVDC factory bay count does not change inside the window. Adding a converter station factory is a five to seven year program.
The binding constraint compounds. Each year of unmet demand pushes orders further into the backlog. Each new operator that secures slot priority pushes the marginal operator further out. The equipment OEM order book becomes the most reliable forward indicator of US AI capacity additions, more reliable than any utility integrated resource plan or any ISO interconnection queue.
By 2028, the equipment manufacturer sits where the integrated oil major sat in the 1970s. The asset is physical, the chokepoint is geographic and industrial, and the capital stack reorganizes around the holder of the chokepoint. The megawatt is downstream. The delivery slot is upstream. The market has not finished repricing the difference.