Construction Bottlenecks
Transformers at four-year lead times, gas turbine slots sold into the 2030s, high-voltage cable, and skilled labour. The constraint on building moved from money and permits to factory capacity, and a queue is not a market.
Words the industry uses are marked like this at the sentence that defines them, so you can tell a term you will hear on a desk from a phrase this book happens to be using.
A developer with financing, land, a signed offtake and a completed interconnection study can still take five years to produce a megawatt-hour, and the reason will have nothing to do with any of the four. Instead the delay is often due to a transformer, or a turbine, or a crew.
For most of the period this book describes, the constraints on building were money, permits and the queue. Demand was flat, factories had spare capacity, and equipment was something a procurement department bought rather than something a chief executive worried about. That changed between roughly 2021 and 2026, and the change is large enough that the delivery schedule for physical hardware now sets the pace of overall economic growth, and the data-centre buildout in particular. This chapter is about what is scarce, why, and what a market does when the thing it needs cannot be bought at any price for four years.
Transformers
A transformer changes voltage. Nothing on the grid works without one, and there is one at every step of Chapter 3: a generator step-up unit lifting a plant’s output to transmission voltage, substation transformers stepping it back down, and a small pole-mounted or pad-mounted unit at the end of almost every street.
Lead times for the large units have gone from something on the order of a year before 2020 to as much as four years, with US power transformers running around 128 weeks and generator step-up units around 144 weeks as of recent quarterly surveys. Prices have moved with them: power transformers up something like 70 to 80% against 2019, distribution transformers up in a similar band or more.
Demand for generator step-up units rose by a factor of nearly four between 2019 and 2025 as renewables and nat gas connected at scale, and substation transformer demand roughly doubled. Large units are close to bespoke, built to a utility’s specification rather than pulled from a shelf, which makes them hard to stockpile and hard to substitute between buyers. And they need grain-oriented electrical steel, rolled and annealed so that its crystal grains line up in one direction, which is what lets a core carry magnetism the efficient way round and makes it the material Chapter 2 said iron was for. There is one remaining domestic producer in the United States, which puts a hard ceiling on how fast domestic output can rise regardless of how many winding shops get built.
Nothing about that steel is uniquely American, and the transformers themselves are built all over the world. Siemens Energy in Germany, Hitachi Energy in Switzerland, Toshiba and Mitsubishi Electric in Japan, Hyundai Electric and Hyosung in South Korea, TBEA in China and CG Power in India all build large units, and roughly a quarter to a third of the finished distribution transformers installed in the United States are already imported, mostly from Mexico, Canada and Taiwan. Between half and three quarters of the cores inside the units assembled domestically are imported as well. So the obvious answer to a domestic bottleneck is to buy abroad. Three things get in the way of that.
The first is that the shortage is not an American one. Large transformer lead times run to three or four years across the world, and the queue is longest in the places with the most demand: some North American and European orders are quoted at five years, against roughly a year in Asia-Pacific, where most of the factory capacity was built. Grain-oriented electrical steel is tight everywhere, at prices half again to nearly double their pre-pandemic level. A utility that goes abroad is joining a shorter queue rather than skipping one, and only if it goes as far as Asia.
The second is trade policy, which in the United States has been pointed at this industry for eight years. Section 232 put a 25% tariff on imported electrical steel in 2018, and the rate went to 50% on 3 June 2025. Copper, the other thing a transformer is mostly made of, was tariffed at 50% from 1 August 2025. What happened in between is the part that matters. A tariff on steel is not a tariff on a part made from steel, so mills shipped grain-oriented steel to Mexico or Canada, cut and stacked it into laminations or a finished core there, and sent the core into the United States duty free. Core imports went from $126m in 2018 to $524m in 2025. On 15 August 2025 the 50% rate was extended to those derivative products, which closed the route.
The third is that these are made to order. A large transformer is built to a utility’s specification, tested against it, and has to match the dimensions, the protection scheme and the fault duty of one particular substation. Qualifying a new supplier is a programme of its own, and it takes the kind of time the buyer went abroad to save.
On price, the pole-mounted and pad-mounted units are a few thousand dollars each and are bought by the thousand. A high-voltage power transformer between 69 and 230 kV runs from a few hundred thousand dollars to about two million, and the extra-high-voltage machines above that, up to 765 kV, run from two million to ten million and beyond. There is a wide gap between the official statistics and what buyers report: the producer price indices put transformers up around 40% in real terms between 2020 and 2024, while utilities and distributors describe increases of 70 to 95% over roughly the same period. An index tracks a specified product, and a good deal of what a utility now pays for is expedited delivery, a better queue position, and a specification bent to whatever the factory can actually build.
The consequence for a market is that a transformer is now an asset with a secondary value, and utilities have started to behave accordingly: ordering speculatively before a project is approved, holding strategic spares, and in some cases standardising specifications across utilities so that a unit ordered for one substation can be diverted to another. Standardisation is the interesting response, because it is an attempt to make a bespoke item fungible after the fact, and fungibility is the precondition for anything being traded at all. Cleveland-Cliffs has gone further and is building a plant at Weirton, West Virginia, to turn its own steel into finished distribution transformers, integrating forward into the shortage rather than only supplying it, alongside a $170m expansion of the Butler mill due around 2028.
Gas turbines
The turbine story is even more constrained, because there are three suppliers on earth for the large machines and all three are full.
GE Vernova’s gas turbine backlog reached about 116 GW by the middle of 2026, from roughly 100 GW a quarter earlier, with the company guiding to at least 125 GW under contract by the end of the year and targeting 30 GW of annual output by 2030. Siemens Energy is carrying the largest order book in its history at roughly 136 billion euros. Mitsubishi Power has been described as sold out into 2028. All three are booking reservations four to five years forward, and slots for 2031 delivery are being contracted now.
What emerged in response is a market that did not exist five years ago. A buyer no longer orders a turbine; a buyer reserves a manufacturing slot, paying a fee years before the machine is specified, and converts the reservation into a firm order later. The slot is the scarce thing. It is dated, it is transferable in practice, it trades at a premium, and holding one is a competitive advantage independent of whether the project behind it goes ahead.
That is close to being a forward market in manufacturing capacity, with one difference: there is no exchange, no published price, and no clearing house, so the reservation is a bilateral contract with a manufacturer who is also the counterparty on every other reservation. A reader of Commodities 101 will recognise the shape from the markets there that never developed a futures contract, where an obviously scarce good stays bilateral because production is concentrated in very few hands.
The second-order effects are already visible. A project developer who cannot get a turbine buys a company that has one, which is part of why generation assets have re-rated higher in value so sharply. A hyperscaler that cannot wait builds on site with whatever is available, including reciprocating engines (diesel engines) and aero (jet engine) derivative units that would never have been chosen on efficiency. And the same machines are wanted for liquefied natural gas trains, industrial cogeneration and mechanical drive, so power competes for turbines against buyers who are not in the power business at all, which is why a damaged LNG facility can take years rather than months to restore.
Everything else in the queue
Table 26-1: What is scarce, and what kind of scarcity it is
| Item | Nature of the constraint | Can it be relieved with money |
|---|---|---|
| Large power transformers | Bespoke design, and a narrow supply of grain-oriented electrical steel | Slowly, and the steel sets the ceiling |
| Gas turbines | Three suppliers, decades of tooling and skilled assembly behind each | Only on a five-year view |
| High-voltage cable | A handful of makers, and a smaller number of vessels able to lay it | Partly, and the ships are the harder half |
| Switchgear and breakers | Broader supplier base, but the same demand surge | Yes, over a year or two |
| Skilled line crews | A four-year apprenticeship that cannot be compressed | No, not inside four years |
| Battery cells | Large and still growing manufacturing base | Yes, which is why storage is the fast option |
Labour is the constraint least responsive to price, for the reason given in Chapter 3: a journeyman lineworker is four years of apprenticeship and several thousand supervised hours, and no amount of money shortens that for a person in their first year. Battery cells are the constraint most responsive to price, because that industry has been scaling for a decade and can add a line in months.
A queue is not a market
The economically interesting property of all of this is that a four-year lead time is a rationing mechanism, and it rations by something other than willingness to pay.
In an ordinary shortage, the price rises until demand matches supply, and the buyers who value the good most get it. That is happening here to a degree, and the 70 to 80% price increases are the evidence. But a manufacturing slot is also allocated by relationship, by order history, by volume commitment and by whoever asked first, which means a well-connected incumbent with a modest project can hold a slot that a newcomer with an urgent one cannot buy.
This is the same problem the interconnection queue of Chapter 25 has, and the responses look similar. FERC Order 2023 moved the connection queue to first-ready rather than first-come. Manufacturers have reached for the equivalent, weighting allocation toward buyers who have specified the machine, committed the site and posted money, which is a commercialised version of the first-ready rule.
What it means for the market
Two consequences follow for anybody pricing power more than a couple of years out.
Supply cannot answer a price signal quickly, so scarcity persists for longer . Chapter 12 sets out the case that high prices in tight hours call forth new plant. That mechanism still operates, but the lag between the signal and the steel is now most of a decade, which means a tight market stays tight through several years of prices that in theory should have fixed it.
The value of an existing asset rises with the difficulty of building its replacement. A nat gas plant that can be replicated in eighteen months is worth roughly what it costs to build. The same plant, when the turbine for its replacement cannot be delivered before 2031, is worth considerably more, and the reopened reactors of Chapter 27 are the most extreme version of the same arithmetic. When new supply is slow, the premium moves to whatever already exists.