The defining shortage of the 2026 AI build-out is not chips, land, transformers or electricity — it is a manufacturing slot with a delivery year attached, and the market for that slot now clears six years forward. Two disclosures in the space of two weeks make the instrument legible. GE Vernova reported on 2026-10-28 that "Gas Power equipment backlog and slot reservation agreements grew from 116 to 122 GW" and that it expects "at least 130 GW by year-end 2026." Siemens Energy reported on 2026-11-11 that its fiscal year closed with total commitments of 104 GW, "comprising 65 GW of firm orders and 39 GW under slot reservation agreements." Neither number is a backlog in the accounting sense, and that is precisely the point: a slot reservation is a deposit that holds a future position on a production line, not an order for a machine.

This piece does four things with those disclosures. It separates the two things being sold — the machine and the date — because they have different economics and different risk. It converts the order books into years of production, which is the only unit that tells you when your slot would actually run. It checks who is buying, because the answer contradicts the story the market tells about why prices moved. And it prices the wait, which turns out to be the number that explains why buyers keep signing further out even as prices rise. Every figure below comes from a company release, an earnings call, or both; the arithmetic is ours and is labelled as such.

What is actually being sold in a gas turbine order?

Two contracts, not one. The firm order commits the manufacturer to build a specified unit and the buyer to pay for it; it lands in backlog and, for GE Vernova, inside the $197B of remaining performance obligation the company reported for the third quarter. The slot reservation agreement is earlier and softer: it holds a place in the production schedule, usually against a fee and sometimes against customer advance payments, and it converts into a firm order later. Siemens Energy's head of investor relations described the mechanics on the 2026-09-30 pre-close call: reservations "are structured commitments rather than speculative reservations and are generally intended to convert into firm orders within a defined period, typically around six months. Reservation fees and advance payments provide economic protection." The fiscal-2026 earnings release states the cash effect directly: free cash flow "continued to benefit from customer advance payments and reservation fees on the back of continued high order intake."

So the reservation is a deposit-backed option on a date. It is not a purchase order, it does not commit a frame size, and its price is set when the reservation is signed rather than when the order is placed. That last detail is what makes the reported pricing numbers so hard to read, and it is where the next section starts.

The two companies disclose the same quantity differently, which matters more than it looks. GE publishes one combined figure — "equipment backlog and slot reservation agreements" — and does not split it in the release. Siemens splits it every quarter. A reader comparing "122 GW" with "104 GW" is therefore comparing a number with a reservation component to a number with a reservation component, but only one of them tells you how much of it is soft. For a supplier this is a presentation choice. For a buyer trying to work out how much of the calendar is actually spoken for, it is the difference between a fact and an aggregate.

How many years of production are already sold?

Convert the book into time. GE Vernova shipped 5 GW in the third quarter — a 20 GW annual run rate — while signing 11 GW of orders and reservations. Against 122 GW of commitments, that is 6.1 years of output. Q3 2026 plus 6.1 years lands in late 2032. Strazik, on the same call: "This quarter we sold multiple slots for '32 delivery as the demand environment remains strong." The two statements were made independently — one is a shipment figure, one is a booking comment — and they agree to within a quarter. That agreement is the single most useful sanity check available on the whole gas-turbine narrative, because it means the order book is not a marketing aggregate: it is a calendar, and the calendar ends where management says it ends.

Siemens Energy's book converts to a shorter period because its capacity plan is larger. Total commitments of 104 GW against a target of "approximately 30 gigawatts of annual delivery capacity by 2030" is 3.5 years — but that denominator is a 2030 number, not today's, so the real duration is longer. The company is expanding from 35 to 50 large gas turbine units a year, contributing from fiscal 2027, and from about 50 to about 80 medium-sized units in the year just closed. It also said on the pre-close call that slots for 2029 and 2030 were "filling up very quickly," and on the results call that reservations convert within "a defined period" against milestones it manages.

GE Vernova (Q3 2026)Siemens Energy (FY2026)
Commitments under contract122 GW (backlog + slot reservations)104 GW
Disclosed splitnot split in the release65 GW firm + 39 GW reserved
Signed vs shipped, latest period11 GW signed, 5 GW shipped42 GW of new unit orders in the year
Delivery capacity~20 GW/yr at the Q3 run rate~30 GW/yr targeted for 2030
Years of book at that capacity6.1 → late 20323.5 at the 2030 figure, longer today
Price signal2026 order price/kW ~50% above 2025"very strong"; conversions expected above FY26 margins

One more consistency check, and it is the one that should worry anyone modelling this as a temporary squeeze. Siemens Energy now puts the annual market at an estimated 140–150 GW in fiscal 2026, up from 108 GW in fiscal 2025 and 64 GW in fiscal 2024, and says it holds "a global market share of around 30%." Thirty per cent of 140–150 GW is 42–45 GW a year of demand addressed to a company that is targeting 30 GW a year of delivery capacity by 2030. The gap is not a forecast error; it is the pricing mechanism. When share times market exceeds capacity, the clearing variable is price and the queue, which is exactly what the order books show.

Who is actually buying the calendar?

Not primarily the buyers the coverage blames. Strazik said GE Vernova serves "approximately 100 customers across 25 countries, with about 80% traditional customers and 20% tied to data centers." Siemens Energy's numbers point the same way from a different dataset: data-center-related customers account for roughly €15B of €68.1B in fiscal-2026 orders — under 10% — and "demand excluding data centers increased by around 50% between fiscal year 2024 and fiscal year 2026, driven by electrification, fuel shifts, grid stability, and security of supply."

Read those two disclosures together and the AI story inverts. Data centres are perhaps 15% of the gas turbine buyers in this cycle, and something like 85% of the sold-out calendar belongs to utilities replacing retiring plants, to coal-to-gas switching, and to grid-stability procurement. What data centres changed is not the volume; it is the marginal bid. A utility buying a replacement unit has a planning horizon and a regulated return. A hyperscaler with an energised-shell deadline has a date it cannot miss, and it will pay whatever the last available slot costs. In a market with no spare capacity, the price is set by the buyer with the least elasticity — so 15% of the volume sets the price for the other 85%, who then discover that their regulated cost recovery has to absorb a price move they did not cause.

That is also why the geopolitical and policy commentary around this cycle keeps missing. Siemens Energy's CEO was asked on the results call about US midterm elections and any slowdown in data-center demand; his answer was that he had seen no slowdown, and that demand was "well-diversified." The order book supports him. The demand is broad; the urgency is narrow; the price follows the urgency.

What does one year of waiting cost?

This is the arithmetic that explains why buyers keep signing further out while complaining about price, and it needs only one input the companies have already given. GE Vernova's CEO: "Our orders in '26 relative to '25 pricing is up directionally 50%," referring to full-year gas equipment order pricing per kilowatt, with the caveat that quarterly comparisons move with product mix. Third-quarter orders alone were priced more than 20% above fourth-quarter 2025 orders.

Take a buyer choosing between booking now and booking in twelve months. Booking later costs 0.5 × P per kW in escalation, where P is the order price per kilowatt. Waiting also delays output by a year, and a year of output on one kilowatt of capacity at an 80% capacity factor — 7,008 hours — earns roughly $210 per kW at a $30/MWh spark spread (ours; 7,008 × $30 ÷ 1,000). Setting the two equal gives a break-even order price of $420/kW. Above that price, a year of waiting costs more than the year of revenue it postpones. Every heavy-duty gas turbine order in this market is above $420/kW by a wide margin, so the conclusion is not close: at 2026 escalation rates, the rational move is to take the earliest slot available at almost any price, which is precisely the behaviour that fills 2032.

Put numbers on a mid-size order. A 500 MW order priced at $600/kW of equipment is $300M. A 50% escalation is $150M for one year of delay, against $105M of gross margin from that year's output at 80% capacity factor and a $30/MWh spread (ours; the $600/kW is an assumption, stated so you can move it). Waiting costs $45M more than it earns. Push the capacity factor to 90% and the margin rises to $118M — still short. Push the spread to $50/MWh and the year is worth $175M, which finally beats the escalation — which is to say that the wait only pays for itself if you believe spark spreads rise by two-thirds. Very few buyers in 2026 are underwriting that.

The same arithmetic runs in reverse for the seller, and this is where the reported numbers get interesting. GE said fourth-quarter pricing per kilowatt could come in below the third quarter "because more 50 Hertz units will move from reservations" — that is, units reserved earlier, at earlier prices, are converting now and dragging the blended order price down. Siemens said the opposite of its own book: fiscal-2027 conversions are expected "at a more attractive margin profile than the fiscal year 2026 orders." Both can be true, because a reservation vintages its price on the day it is signed. Which leads to the least obvious conclusion in this piece: the reported order price is a lagging indicator of when the reservation was signed, not of what a turbine costs today. The cheapest way to buy a 2029 turbine in 2026 is to have reserved it in 2025.

What happens when a project slips?

The seller takes the slot back. Siemens Energy's own description, from the results call: "If a project does not progress within the agreed timeframe, we can reallocate the slots to another customer where our contractual rights allow. This protects utilization and ensures scarce capacity is deployed to firm, value-creating projects." That sentence is the whole risk allocation of the reservation model in one line. The buyer's option is callable on a schedule: miss a milestone and the date you paid to hold is re-sold to the next buyer in the queue. The reservation fee is the premium; the order price is the strike; the milestone calendar is the expiry.

It is the mirror image of the capacity-market instrument we wrote about on 2026-11-11. There, a data centre pays $333.44 per MW-day for an option on megawatts standing by, and the risk that the load never arrives sits on the buyer's balance sheet through an 85% take-or-pay floor and $1.5M per MW of collateral. Here, the buyer pays a reservation fee for an option on a date, and the risk that the project never arrives sits on the buyer through the loss of the slot. Same instrument logic, opposite direction: in capacity markets the load buys insurance from the generator; in the turbine market the developer buys an option from the factory and forfeits it if the project dies.

The honest caveat is that nobody has seen the forfeiture phase yet. Siemens Energy states plainly that it has "not seen any delays or cancellations of previously signed slot reservation agreements or orders." That is a statement about a book built during a boom, with a typical six-month conversion window, and it will remain true until it is not. The first real test of the reservation model is not a price decline — it is the first quarter in which a hyperscaler's project slips a milestone and the slot is reallocated. Watch the disclosure language at that moment, not the GW number.

Why does the service contract decide the price?

Because the machine is not the product. Siemens Energy, on the results call: "At today's pricing, a 20-year service contract represent lifecycle revenue of more than EUR 400 million per gigawatt, resulting in a service revenue opportunity of more than EUR 100 billion from the newly added capacity," based on roughly 250 GW of fleet additions expected from fiscal 2024 through fiscal 2030. It also said new long-term service agreements signed in fiscal 2026 "carry margins around 5 percentage points above those in the existing backlog," and that the installed fleet approaches 1,000 GW in the 2030s.

Do the division. €400M per GW over 20 years is €20M per GW-year of service revenue (ours). At 42 GW of new unit orders in fiscal 2026, the service revenue written in a single year is about €16.8B over the following two decades — against €68.1B of total orders booked in that year. And because service margins run above new-unit margins, the profit mix tilts further toward the annuity with every GW sold. This is why the manufacturer's stated objective is not volume: "Our objective is not to maximize gigawatts, but to maximize the returns generated by the capacity we deploy." A turbine sold at a discount to a buyer who will sign a 20-year service agreement is a customer-acquisition cost; a turbine sold at a premium to a buyer who will service it elsewhere is a worse deal at a better headline price. Which is also why the reservation system is disciplined about who gets a slot: reservations convert against milestones, and the company says its project-selection process screens for projects that "can support 15 to 20 years of long-term service revenues."

For a buyer this reframes the negotiation. The number on the term sheet is the equipment price; the number that decides the deal is the service agreement attached to it, because that is where the manufacturer's margin actually lives. A 5-percentage-point margin gap on a 20-year annuity is worth more than a few per cent on the machine, and it is the reason a reservation signed early is valuable to both sides: the buyer locks a date and a price band, the seller locks two decades of service revenue.

What would have to be true for this to unwind?

Three things, and only one of them is about AI. First, capacity has to arrive faster than the book grows. GE is guiding to 24 GW of output in 2028 and toward 30 GW by 2030 against 20 GW today; Siemens targets 30 GW by 2030 from a base that is expanding 35→50 large frames and 50→80 medium frames. If both hit those numbers, the combined Western delivery capacity roughly doubles by 2030 — which is what turns a six-year book into a three-year book without a single order being cancelled. That is the benign path, and it is the one both companies are underwriting with capital.

Second, the marginal buyer has to stop being inelastic. The 15% sets the price only because it cannot wait. If hyperscaler capex digests — if the 2027–2028 build-out slips a year — the marginal bid leaves the market and the remaining 85% of utility demand clears at utility prices, which is a different number entirely. Note that this is a demand-side risk with a specific signature: it shows up first as reallocated slots, not as lower prices, because the price on a reservation is set when it is signed.

Third, non-Western supply has to become bankable. Siemens Energy's own market estimate of 140–150 GW a year is a global number that includes frames the Western three do not build, and its ~30% share implies it is not counting the Chinese OEMs inside its own denominator advantage. Those suppliers have the capacity; what they do not yet have, at scale, is the 20-year service network and the utility reference list that a regulated buyer needs. Every year that gap persists is a year the price holds.

What are the caveats?

The headline numbers are aggregates of two different instruments, and only one company splits them. GE's 122 GW combines equipment backlog with slot reservation agreements and is not split in the release; Siemens' 104 GW is explicitly 65 GW firm and 39 GW reserved. Comparing them as if they were the same quantity overstates GE's firm position and understates the softness in Siemens'. The 122 GW is also not the same as the $197B backlog GE reports — that figure is remaining performance obligation, an accounting measure, and the GW number is an operating one.

Per-kilowatt pricing is a mix metric, and the companies say so. Strazik's own caveat is that "quarter-to-quarter comparisons can move with product mix," and he flagged fourth-quarter per-kW easing as 50 Hz units convert from reservations. Siemens' investor-relations note is blunter: "order intake per gigawatt should not be used as a pricing indicator because frame size, geography, Hertz rating, scope and the new-unit and service mix can materially distort that metric." We used the ~50% figure as a full-year, like-for-like direction, not as a per-model price.

The market-size numbers are the sellers'. The 140–150 GW estimate for fiscal 2026, the 110–120 GW planning assumption, and the ~30% share are Siemens Energy's own; no independent tally of global gas turbine orders exists at that granularity, and the figures include frame classes outside the Western duopoly. We used them for internal consistency (share × market vs capacity), not as a market forecast.

The break-even arithmetic is ours and depends entirely on two assumptions. The $420/kW threshold assumes a $30/MWh spark spread and an 80% capacity factor, and it scales linearly: double the spread and the threshold doubles. The $600/kW order price in the 500 MW example is illustrative, not observed — neither company discloses a list price per kilowatt, and every published per-kW figure in this market is either a mix-weighted average or an analyst estimate. What is not an assumption is the escalation rate: the ~50% year-over-year move is management's own statement, repeated in several independent accounts of the call.

And the reservation model has not been stress-tested. Siemens Energy says it has seen no delays or cancellations; that statement has a six-month conversion window behind it and a boom in front of it. The disclosures that would tell you whether the model is robust — a conversion rate, a forfeiture rate, the size of reservation fees relative to order value — are not published by either company. Until they are, treat the order book as a calendar with a real but unmeasured cancellation option attached, and price the date accordingly.