Research Note
The Order Book
What equipment backlogs disclose about how long the power shortage lasts, and what they conceal.
The previous note argued that the AI buildout is constrained by the date power becomes available rather than by its price, and that the most durable economics sit with manufacturers whose output cannot be scaled on the timeline the buildout demands. That claim is testable, and unusually so, because the manufacturers disclose both halves of the equation: what they have sold, and what they can build.
Almost all of the commentary quotes the first half. GE Vernova’s gas equipment book reached 116 GW in the second quarter of 2026, up from 100 GW three months earlier, and management expects at least 125 GW under contract by the year end. Those are large numbers and they are reported as evidence of demand, which they are. They say very little on their own about how long anyone gets paid for it.
Two things have to be held against that figure before it means anything. What proportion of it is a firm order rather than an option, and how many years of production it represents at the output the company is actually building toward. The answers are both disclosed, and neither is flattering to the way the headline is usually read.
What is actually in the book
A slot reservation agreement is not an order. It is a paid position in the delivery queue, taken by a customer who wants the option to build and does not yet know whether it will. It converts to a firm order later, or it lapses. Ten gigawatts converted in the second quarter of 2026, which is the number that matters and the one almost nobody quotes.
The Gas Turbine Book Is Half Reservations
GE Vernova gas equipment book at Q1 2026, gigawatts, split between firm backlog and paid slot reservations
A slot reservation is a paid position in the delivery queue, converted to a firm order later or not at all.
Source: GE Vernova.
This is not a criticism of the disclosure, which is clear about the distinction. It is a criticism of how the number travels. A reader who takes 116 GW as contracted demand is treating a majority of it as something it is not, and the error runs one way: reservations can quietly fail to convert, and nothing is ever announced when they do. Cancellation makes news. Non-conversion does not.
The useful discipline is to track the two series separately and watch the ratio between them. Firm backlog rising while reservations flatten is a book hardening. Reservations rising while conversions stall is a book inflating with optionality, and it would show up in exactly the same headline figure.
How long the book takes to clear
The second question is arithmetic. GE Vernova has said it is targeting 20 GW of annualised output in the third quarter of 2026, 24 GW in 2028, and has described actions toward 30 GW by 2030. Set the present book against each of those.
Years to Clear the Book Depend on a Rate Not Yet Reached
Years required to deliver a 116 GW book at three annual output rates. Coverage is a Seven Measures calculation, the rates are company targets
Book of 116 GW divided by each stated annual output rate.
Source: GE Vernova; Seven Measures calculations.
Nearly six years of coverage is the strongest position in the whole AI power complex, and it is worth being precise about why. It is not that demand is enormous. It is that a heavy-duty gas turbine plant cannot be stood up quickly, and the incumbents are not trying to. The capacity path is a 50% increase spread over four years, executed largely inside the existing footprint, on capital commitments of a few hundred million dollars. That is a company expanding carefully into a shortage it does not fully believe.
The prior note argued that this caution is the residue of the last cycle, when gas turbine demand collapsed and these businesses were restructured or written down. The capital numbers are consistent with that reading. A manufacturer that believed the demand curve would add capacity aggressively and compete the price away; one that suspects it is looking at a five-year bubble takes the margin instead and expands only as fast as it can un-expand.
The Book Grew Through 2026
GE Vernova gas equipment book, gigawatts under contract. The end-2026 figure is the company's guidance, not a reported position
The end-2026 figure is guided as at least 125 GW and is drawn open, as every forecast in these notes is.
Source: GE Vernova.
The constraint behind the constraint
Turbines are the visible constraint. The one behind them is duller and harder to relieve. Every large power transformer needs a core wound from grain-oriented electrical steel, and the United States has one domestic producer of it, Cleveland-Cliffs, from plants at Butler in Pennsylvania and Zanesville in Ohio.
That matters more than the number of transformer factories, because a transformer plant can be built faster than a steel line that makes a specialised grade. Trade reporting through 2026 puts high-voltage transformer lead times at four to five years against roughly one year before 2020, and describes manufacturers waiting twelve to eighteen months for their own core material. Those figures come from trade sources rather than filings and should be treated as indicative, but the structure they describe is not in dispute.
Capital is arriving. Hitachi Energy has committed to a multi-billion dollar programme to expand transformer and high-voltage manufacturing, including US plants in Virginia, Missouri and Mississippi. Cleveland-Cliffs is converting its Weirton facility in West Virginia to transformer production, expected online in the first half of 2026, and has said the conversion will draw on under-utilised capacity to produce thirty to forty percent more grain-oriented steel tonnage.
Hold those two things together. The response to the shortage is measured in tens of percent of additional core steel over several years. The load it is being asked to serve is discussed in multiples. A thirty to forty percent increase from a single supplier is a meaningful commitment and it is not the same order of magnitude as the demand being described, which is the reason to expect transformer lead times to compress later than turbine lead times rather than sooner.
Which positions earn on scarcity
The equipment complex gets traded as one position and it is at least four. What separates them is not the size of the backlog but who funds the working capital, how quickly a competitor can add capacity, and whether the scarce input is the company or something the company also has to buy.
Forms of scarcity in the power equipment chain
| Position | What the scarcity rests on | Who funds the wait | What ends it |
|---|---|---|---|
| Heavy-duty gas turbines | Years of booked output and a small field of qualified suppliers. | The customer, through reservation fees and prepayments. | The suppliers themselves, on a schedule they publish. |
| Large power transformers | Lead times of four to five years and a specialised core material. | Split. Deposits are common; the manufacturer carries the rest. | Core steel capacity, which is arriving slowly. |
| Grain-oriented electrical steel | One US producer of a grade nothing else substitutes for. | The transformer maker, who buys ahead. | New lines or imports. Both are slow, neither is impossible. |
| Switchgear and busway | Assembly throughput, not a scarce input. | Mostly the manufacturer. | Capacity, which adds in quarters rather than years. |
| Engineering and construction | Licensed crews, not the firm. | The contractor, through working capital. | Labour supply, and it does not accrue to the firm meanwhile. |
SourceSeven Measures analysis, drawing on company disclosure and trade reporting through 2026.
The first three rows behave differently from the last two in a way that matters for return on capital. A turbine maker taking a reservation fee is being paid to hold inventory it has not built, which is the rarest arrangement in industrials: negative working capital on a multi-year order. A switchgear assembler with a long backlog is being paid on delivery for a product whose capacity a competitor can add inside a year, and its backlog is a measure of current demand rather than of pricing power.
Cleveland-Cliffs is the most interesting and the least clean. It holds the strongest structural position in the table and captures the least of it, because electrical steel is a commodity sold into a concentrated customer base, and the company’s equity is dominated by a carbon steel business with nothing to do with this. Owning the bottleneck and owning the earnings from the bottleneck are not the same thing, and the distinction is worth more attention than it usually gets when a supply chain gets mapped in public.
What to watch, and where it is disclosed
The advantage of this part of the chain over the rest of the AI power trade is that the evidence arrives quarterly and in a form that can be checked. These are the disclosures that would move our assessment, in the order they would move it.
Disclosures that would change the assessment
| Disclosure | Where it appears | What a change would mean | Latest |
|---|---|---|---|
| Slot reservation conversions | Quarterly results commentary. | Conversions stalling while reservations rise is the book inflating with options. | 10 GW converted in Q2 2026 |
| Firm backlog against reservations | Segment disclosure. | Firm share rising is demand hardening; falling is the opposite. | 44 of 100 GW firm at Q1 2026 |
| Stated output capacity | Guidance and capital announcements. | A step up beyond 30 GW would shorten the shortage materially. | 20 GW target for 2026 |
| Core steel capacity additions | Producer announcements. | A second US line, or a large import programme, relieves transformers. | Weirton conversion, 30 to 40% more tonnage |
| Equipment pricing in new awards | Utility rate filings and project disclosure. | Prices flattening while backlogs grow is the clearest early signal. | No usable public series |
SourceSeven Measures analysis. Readings are as at the second quarter of 2026 where a figure has been disclosed.
The last row is the honest gap. Equipment pricing on individual awards is the single most informative number here and it is not published in any form that can be tracked consistently. Utility rate filings contain fragments of it, project by project and jurisdiction by jurisdiction. Assembling that into a series is real work and we have not done it. Anyone claiming to know where equipment pricing is going without having done it is inferring from backlog, which is the thing this note argues you cannot do.
Assessment
The equipment position holds, and it holds for better reasons than the size of the headline backlog. Six years of booked output, customer-funded working capital, and a supplier field that is expanding at half the rate its order book is growing is a genuinely durable arrangement. Nothing in the disclosure suggests it breaks in the next two years.
What the disclosure also does is put a term on it. The capacity path is public and it is not slow forever: at the stated 2030 output rate the book standing today clears in under four years. This should be priced as an annuity with a known end date rather than as a structural change in the industry, and the difference between those two things is most of the valuation.
The part we would treat with most suspicion is the composition. A book that is a majority reservations can deflate without a cancellation, and the quarterly conversion rate is the only place that would show up early. If conversions slow for two consecutive quarters while the combined figure keeps rising, the headline will still look strong and the position will already have changed.