Building your own power: GE Vernova's queue, not the grid's, sets the date
An AI campus that cannot wait years for the grid has an obvious way out. It can build its own power station beside the building and burn gas.
Many have taken it, and now they queue beside the utilities for the same machines. At the end of September GE Vernova, which makes many of them, had 122 gigawatts of gas turbines under contract. It ships about 20 gigawatts a year.
So building your own power does not skip the queue. It changes which queue you stand in, and the new one is about six years long and still growing.
Explain it like I don't work in finance
Your street's water main is too small, and the council's new one is years away. So you decide to sink your own well. The big drilling firm is booked for years, because every house had the same idea.
A smaller firm with a lighter rig can come sooner, but its diary is filling too. And either well still needs a pump, made by someone else.
Making power on your own side of the utility’s meter is called going behind the meter. It skips the grid’s waiting list. It does not skip the need for a machine.
The biggest machine, the heavy-duty gas turbine, is the cheapest at scale and the slowest to get. Its makers sell paid places in their factory line years ahead. After it ships, it needs about another year and a half on site before it makes power.
The quicker machines have queues of their own. A turbine built around a jet engine ships about two years sooner and runs within months, but its maker now takes orders far faster than it ships them. Fuel cells arrive by lorry, and their maker is doubling its factory to keep up.
Supply cannot simply catch up. GE Vernova is raising output inside the factories it already has, paid for by its customers’ deposits, and its next step waits on castings due in 2027.
The gas has a clock as well. A new pipe is paid for by its buyers under long contracts signed before it is built, and a recent one is due about four years after its first bids. The newest pipe from the Permian, contracted to fuel new power plants for one tech giant, opens in 2030. The heavy turbines are sold out to the same year.
So a paid place in a turbine queue is not power. It is a date, and GE Vernova is now selling places for 2032.
For investors
Leaving the grid’s queue means joining four others
A campus that leaves the grid’s queue passes four other clocks: the machine, the castings inside it, the air permit and the fuel. The machine is the longest, so it sets the date.
It comes three ways. A heavy-duty turbine of hundreds of megawatts needs about 18 months on site after it ships. An aeroderivative, about 35 MW, ships about 24 months sooner and needs about six. Bloom Energy’s fuel cells come in boxes added in steps, and Bloom expects to power a full data centre for Oracle within 90 days, an expectation rather than a measured delivery.
The way to measure each queue is years of output: the orders waiting, divided by what the factory can ship in a year. A queue that grows faster than the factory means a later date for the next buyer, whoever it is.
The turbine queue is six years long and still growing
GE Vernova counts firm orders and slot reservations, paid holds on future production slots, together as gigawatts under contract. In nine months the count rose from 83 GW to 122 GW. At its present 20 GW a year, that is a rise from 4.2 to 6.1 years of output.
The turbine queue grew by almost half in nine months
Gas turbine gigawatts under contract at GE Vernova, firm orders and paid reservations, at each quarter end
The queue is hardening as it grows. In the third quarter GE Vernova turned 13 GW of reservations into orders, so firm orders now exceed reservations. It sold “multiple slots for 2032 delivery”, and it expects at least 130 GW under contract by the end of 2026. A turbine in one of those slots would make power no earlier than 2033.
The maker is paid to keep the queue, not to clear it
GE Vernova is not building its way out of the queue, and its plan says so. Output rises from 20 GW a year now to 24 GW in 2028 and 30 GW in 2030, through “lean and incremental machinery in our existing factory footprint”, “all funded by customer down payments”. No new heavy-duty factory has been announced. Even at 30 GW a year, today’s 122 GW would still be four years of work.
The buyers are paying to stand in line. In the Power segment, which also holds nuclear and hydro, contract liabilities and current deferred income doubled from $16.5 billion at the end of 2025 to $33.2 billion at 30 September 2026. The filings put the cash rise down mainly to down payments on orders and reservations. GE Vernova expects 2026 orders to be priced about 50% above 2025’s per kilowatt, and about 20% of its roughly 100 gas customers are data centres.
The next step up waits on parts the maker does not control. A turbine’s hottest blades are cast from superalloy, a nickel metal made for heat, by a few specialist foundries. GE Vernova’s chief executive said he could “see clearly the castings and forgings arriving in ‘27” for the 2028 step. Howmet Aerospace, “the largest manufacturer of gas turbine blades in the world” by its own account, grew gas turbine revenue 45% in the third quarter. It is not the only gate: several of its largest customers have their own furnaces, and Siemens Energy is adding blade capacity in Florida.
The fast lane has a waiting list of its own
The smaller machines were meant to be the way round the queue, and they are queued too. In the first nine months of 2026 GE Vernova took orders for 127 aeroderivatives, turbines built around a jet engine, and shipped 39.
Even the fast lane takes orders faster than it can ship them
GE Vernova aeroderivative gas turbines, units ordered and units shipped
On 28 October its chief executive said fourth-quarter aeroderivative orders would be lighter, “not because the demand is softer” but because GE Vernova is “now going out so far in the future before we have availability”. Crusoe, which builds AI campuses, ordered 29 of them in December 2024 and mid-2025, nearly 1 GW. I found no statement that they are running, nor whether as primary power.
Fuel cells are the other fast route, and buyers pay for them in volume, not price. Bloom Energy’s third quarter product revenue rose 179%, to $1,073.6 million. Oracle intends to procure up to 2.8 GW, 1.2 GW of it contracted. Yet Bloom’s product margin sat near 35% for two years before reaching 37.9% in the third quarter, one quarter and not yet a trend.
Either route still needs transformers and switchgear, and those have their own queue.
The permit and the gas run on clocks of their own
An air permit can decide whether a big plant gets built at all. On the site of a Pennsylvania coal station, Homer City Generation is building up to 4,678 MW of gas plant beside which Amazon Web Services has filed plans for 39 data centres. A new source of NOx, a smog-forming gas, that size would normally face full federal and state review.
Homer City avoided it by netting: counting the coal boilers’ recent cut against the new plant’s rise. The plant adds 1,142.8 tons of NOx a year; retiring the boilers removed 2,031.8. A cut counts only within five years, so a coal site shut long ago has none to spend. Even here, the plant must buy 369 tons a year of credits for another pollutant before it runs, and first power is targeted for early 2028.
The gas itself is plentiful; the pipe is the clock. Most Permian gas flows out of oil wells, so oil drilling sets how much is made. From early 2025 to early 2026 rigs fell by a fifth while marketed gas rose by a twelfth, to 28.4 billion cubic feet a day. At the coast, power plants share it with export: Golden Pass LNG shipped its first cargo in April 2026.
A pipe is booked by the gas buyers before it is built. Shippers sign long-term firm contracts first; Kinder Morgan’s newest are take-or-pay, paid whether or not gas flows. Transco’s Southeast Supply Enhancement shows the length. Its open season, when buyers bid for space, ran in June and July 2023, and its target in service date is 1 November 2027. That is about four years for 55 miles of new pipe beside an existing line, and 93% of it was signed by three power companies.
The buyers are increasingly power plants. More than 75% of Kinder Morgan’s $14.4 billion project backlog supports power generation and local gas utilities. Its Permian Link, about 2.8 billion cubic feet a day, will serve new power generation for an “A-rated hyperscaler” from the third quarter of 2030.
Every route lands years out, and a turbine ordered today lands last
Laid side by side, from first decision to first power, the routes I can date from primary sources tell one story. Own power still takes years, the fuel arrives alongside the plant, and the turbine slot sold today comes last.
Own power still takes years, and a turbine slot sold today comes last
Dated routes to on-site gas power and its fuel, from first decision to first power or gas
The fastest route I can date from primary sources was built by a pipeline company, and it is small. Ohio’s siting board approved Socrates North on 18 September 2025: 200 MW of gas generation “not physically connected to the electric power grid”, run by Williams’ Will-Power for Sidecat, a Meta affiliate. Williams lists two 200 MW Socrates plants due in the second half of 2026, the first phase done “within budget and on time”. Williams builds and runs the plants, and is itself a pipeline company.
The big routes take far longer. For Meta’s Hyperion campus in Louisiana, the state regulator approved three Entergy Louisiana combined cycle plants in August 2025, 2,262 MW due in 2028 and 2029. In March 2026 Entergy signed a second Meta agreement for seven more, 5,278 MW in service in 2030 and 2031, with the regulator’s vote due in December. Those plants join the same turbine queue as everyone else.
So for gas, the fuel clock runs alongside the plant, not behind it. Pipe bid for in 2023 lands in 2027. Pipe sanctioned in October 2026 lands in 2030, level with heavy turbines sold out to 2030. The one open test is whether Permian Link’s hyperscaler has turbines slotted to make power before the third quarter of 2030. Nuclear is the opposite case, where the fuel runs years behind the new reactors.
What could open the queue
The case against is real. Signings slowed to 11 GW in the third quarter, from about 20 GW in each of the two before, while shipments rose to 5 GW. Reservations can lapse. If output reaches 30 GW in 2030 and signings keep slowing, years of output would fall even while the pile holds. And a small off-grid plant like Socrates can go from approval to its due date in about a year.
None of this yet frees a large campus. Signings still run at twice the rate of shipments, so the pile grows. Socrates is 400 MW, under a fifth of Hyperion’s first three plants. And the claim I made in September, that own generation is not faster than a comparable grid connection, has no primary head-to-head of energisation dates to test it. I concede that limb is open.
The big drilling firm is still booked for years, the lighter rig’s diary is filling, and the pipe was booked by the neighbours before the well was dug. Building your own power buys a place in a queue. What it buys is a date, not power.
The turbine is booked for longer than the tenant
Put two dates side by side. A heavy turbine slot sold now ships in 2032 and makes power no earlier than 2033. Meta’s lease on the Hyperion campus runs from 2029 for four years, renewable up to 20. Set side by side, the machine’s queue runs longer than the tenant’s firm promise.
A turbine ordered today makes power about when a four-year lease would end
A heavy turbine slot sold now, against the plants and the lease at Meta's Hyperion campus
Hyperion’s first plants are not caught: approved in 2025, they land before the lease begins. A campus that orders a big turbine now is. It commits to a machine further ahead than Meta has firmly committed to Hyperion. Meta’s agreement with Entergy carries minimum charges and an exit fee, which is how a utility guards a plant that may outlast its customer.
Every clock is paid for before it starts
The second surprise is who pays for the wait: the buyers, before anything is built. GE Vernova’s Power deposits doubled in nine months, to $33.2 billion. Three power companies signed 93% of Transco’s new pipe before it was laid. Kinder Morgan’s newest pipes are take-or-pay, and one was signed by a hyperscaler.
Every clock is paid for by its buyers before it starts
What buyers commit before the machine, the pipe or the plant exists
That turns the usual picture of a shortage around. The makers and pipe owners are paid up front, so they gain from the queue without betting on it. The transformer makers in the equipment queue behave the same way, adding floor space only as fast as orders justify. The buyers carry the risk. If AI demand cools, a reservation can lapse, but a take-or-pay pipe and a minimum charge are still owed.
The length of the queue is reported every quarter; the deposits and the lease term sit deeper in the filings. Three figures will show whether it holds. The first is GE Vernova’s count at the end of 2026, expected at 130 GW or more. Then come its Power deposits each quarter, and the count at the end of 2027. With heavy turbines sold out to 2030 and slots sold for 2032, it should hold well past my test. The campus that left the grid’s queue did not leave the waiting behind; it now pays for the wait in advance.
What would prove me wrong
The first test is GE Vernova’s count at the end of 2027 against the end of 2026. Shipping about 20 GW a year, it holds level only if it signs as much as it ships, net of lapses. A fall would mean buyers no longer queue faster than the machine ships, though faded demand, lapsed reservations and faster output would all fit.
The second is the fast lane. Since the start of 2025 GE Vernova has ordered 124 more aeroderivatives than it shipped, about 2.3 years of recent shipments. If that backlog is a year of shipments or less by the end of 2027, the fast lane has room.
Research notes
Two initiation notes rest on these claims: GE Vernova and Bloom Energy, where the margin correction was first made. Whether any of this is already in share prices is a question for these notes, not for the journal.
The claims on record, word for word
The test this piece stands behind, quoted word for word from the claims page:
Building your own power (30 October 2026)
Claim: “Building your own gas power swaps the grid’s queue for the makers’ queues, and the fast lane is queued too: GE Vernova’s gas turbine gigawatts under contract at 31 December 2027 will be no lower than at 31 December 2026, and at 31 December 2027 its aeroderivative backlog, units ordered less units shipped since 1 January 2025, will still exceed its aeroderivative units shipped in the previous twelve months.”
Breaks if: “GE Vernova reports fewer gigawatts of gas turbines under contract, firm orders and slot reservations together, at 31 December 2027 than at 31 December 2026, or at 31 December 2027 its aeroderivative units ordered less units shipped since 1 January 2025 are equal to or fewer than its aeroderivative units shipped in calendar 2027.”
Three earlier claims on this queue:
The best argument against my own thesis, and why I still think it is wrong (22 September 2026)
Claim: “Building your own power on site does not escape the queue. Turbines, transformers and switchgear have queues of their own.”
Breaks if: “Turbine lead times fall while the backlog is still being worked through, or large sites energise on their own generation faster than comparable sites get a grid connection, or transformers and switchgear stop being the constraint.”
Where later evidence bears on it: its figures came from a trade magazine; GE Vernova’s own filings now confirm them, and Siemens Energy’s is left out, unconfirmed at source. The second limb, own generation against comparable grid connections, is untested: I have no primary head-to-head of energisation dates. Williams’ Socrates now gives the fast end with primary dates, 200 MW off the grid approved 18 September 2025 and due by end 2026, but no grid date beside it.
Bloom Energy is not selling electricity. It is selling time. (24 September 2026)
Claim: “Bloom Energy’s customers are paying for an earlier switch-on date, not for cheaper electricity.”
Breaks if: “Bloom keeps winning sites where a grid connection is available on schedule, or its deployment times stretch towards turbine and grid timelines, or its gross margin holds while grid waits shorten.”
Where later evidence bears on it: I was wrong. I read Bloom’s rising gross margin as “the time premium, expressed in money”. It rose mainly from product mix; product margin sat near 35% for two years, so buyers paid in volume, not price. Oracle intends, rather than has committed, to procure up to 2.8 GW, 1.2 GW contracted. The claim stands; its margin test is weak.
GE Vernova’s customers are paying years in advance for a place in its queue (9 October 2026)
Claim: “GE Vernova is expanding gas turbine output inside its existing factories, paid for by customers’ deposits, so its queue will stay at four or more years of its planned output through 2030 rather than clear.”
Breaks if: “Before 2030, gigawatts under contract fall below four years of the annual output GE Vernova has planned for that year (20 GW now, 24 from 2028, 30 from 2030), or GE Vernova announces a new heavy-duty turbine factory, or Power deposits and price per kilowatt fall while orders stay strong.”
Where later evidence bears on it: holding. At 30 September 2026, 122 GW was 6.1 years of 20 GW output, above four years of the 2030 step, and Power deposits reached $33.2 billion. One correction: the $27.7 billion the piece said was held “for gas turbines” is the whole Power segment, which also covers nuclear and hydro, though the filings tie the rise to gas orders and reservations.
The figures behind the charts
The three machines, with GE Vernova’s chief executive’s timings of 22 July 2026:
| Route | Each unit | Ships | On site before power | Queue today |
|---|---|---|---|---|
| Heavy-duty gas turbine | Hundreds of MW | Largely sold out to 2030; places sold for 2032 | About 18 months | 122 GW under contract at GE Vernova |
| Aeroderivative | About 35 MW | About 24 months before a heavy-duty unit | About six months | 127 ordered, 39 shipped, nine months of 2026 |
| Solid oxide fuel cell | Boxes added in steps | Bloom’s aim: a full data centre’s power within 90 days | Included | Bloom’s factory doubling to 2 GW a year by end 2026 |
The 90 days is Bloom’s expectation for Oracle (24 July 2025), not a measured delivery.
GE Vernova’s gas turbine queue:
| Quarter end | Firm orders | Reservations | Under contract | Signed | Shipped | Years of 20 GW |
|---|---|---|---|---|---|---|
| Dec 2025 | 40 GW | 43 GW | 83 GW | 4.2 | ||
| Mar 2026 | 44 GW | 56 GW | 100 GW | 21 GW | 4 GW | 5.0 |
| Jun 2026 | 53 GW | 63 GW | 116 GW | 20 GW | 3 GW | 5.8 |
| Sep 2026 | 64 GW | 58 GW | 122 GW | 11 GW | 5 GW | 6.1 |
Sources: GE Vernova results releases, 22 April, 22 July and 28 October 2026. Years of output is under contract divided by 20 GW, my arithmetic. Its grid equipment backlog was $42.2 billion at 30 September 2026, up 60% in a year, partly from buying Prolec GE.
GE Vernova’s aeroderivatives:
| Aeroderivative units | Ordered | Shipped |
|---|---|---|
| 2025 | 63 | 27 |
| First nine months of 2026 | 127 | 39 |
Sources: GE Vernova 2025 Form 10-K and Form 10-Q to 30 September 2026; release and call, 28 October 2026. Ordered less shipped since 1 January 2025 is 124 units, about 2.3 years of shipments over the previous twelve months, my arithmetic. Each Crusoe unit leaves the factory about 95% assembled (22 July 2025); Crusoe said on 18 March 2025 that its Abilene turbines back up the grid.
Bloom Energy: product gross margin was 36.8% in 2024, 35.2% in 2025 and 35.6% in the first half of 2026; 37.9% in the third quarter against 35.0% a year earlier, my computation from its release of 22 October 2026. American Electric Power signed for up to 1 GW (November 2024). Howmet’s gas turbine revenue rose 38% in the second quarter of 2026 and 45% in the third (29 October 2026).
Homer City: seven GE Vernova 7HA.02 combined cycle units and ten Mitsubishi FT8 aeroderivatives. NOx figures are tons a year on a 2021 to 2022 baseline. Without netting, a new NOx source in Pennsylvania’s ozone region faces the lowest achievable emission rate and credits for 1.15 tons of cuts elsewhere for each ton added. The 369 tons are volatile organic compound credits. Pennsylvania DEP, plan approval 32-00457A, 18 November 2025.
The Permian Basin:
| Permian Basin | Q1 2025 | Q1 2026 |
|---|---|---|
| Active rigs | 302 | 242 |
| New wells per rig | 4.8 | 5.5 |
| Marketed gas, billion cubic feet a day | 26.3 | 28.4 |
Source: EIA Short-Term Energy Outlook, October 2026, Tables 5a and 10a, with Baker Hughes rig data. Wells already flowing lose roughly a third of the basin’s gas in a year, my arithmetic from EIA’s monthly estimate. In February to April 2026 Texas LNG exports exceeded the gas burned by Texas power plants (EIA).
New pipe:
| New pipe | Capacity | Who signed, for what | In service |
|---|---|---|---|
| Transco Southeast Supply Enhancement | 1,596,900 Dth a day | Duke, Southern and Santee Cooper, 93%, for power; nine local distributors | Target 1 November 2027 |
| Kinder Morgan Permian Link | About 2.8 Bcf/d | An A-rated hyperscaler, for new power generation | Q3 2030 |
| Kinder Morgan Southeast Connector | Up to 2.6 Bcf/d | Take-or-pay shippers | In phases from Q4 2030 |
Sources: Transco’s FERC application (CP25-10), 28 October 2024, whose open season ran 19 June to 14 July 2023; Kinder Morgan, 21 October 2026. A dekatherm (Dth) is roughly a thousand cubic feet of gas. The 93% is 1,480,000 of 1,596,900 Dth a day, my arithmetic.
Hyperion: Meta’s power agreement with Entergy Louisiana carries minimum charges and an exit fee; its term is redacted. The campus, majority owned by Blue Owl funds, is leased to Meta from 2029 for four years, renewable to 20 (Meta 2025 Form 10-K). The second agreement’s vote is set for December 2026 (Entergy Form 10-Q, 31 July 2026).
Chart dates: Socrates starts at the siting board’s approval of Socrates North and ends at Williams’ second half of 2026; Hyperion at the regulator’s approval in August 2025; the second Entergy deal at its signing in March 2026; Permian Link at Kinder Morgan’s release of 21 October 2026; the turbine slot at GE Vernova’s call of 28 October 2026. Years given without a month are placed mid-year.
Who is exposed if building your own power means joining the makers' queues
This maps who operates in each layer. It is not a recommendation, and naming a company is not a view on its shares. Figures in the piece are disclosures used as evidence for the mechanism.
- Turbines
- GE Vernova, Siemens Energy, Mitsubishi Power and Ansaldo Energia make gas turbines.
- Blades and fuel cells
- Howmet Aerospace casts turbine blades, and Bloom Energy makes fuel cells.
- Plants, campuses and fuel
- Homer City Generation and Entergy build gas plants, Crusoe builds AI campuses, and Kinder Morgan and Williams run gas pipelines; Williams also builds off-grid gas plants.
- On the other side
- AI campuses counting on their own gas power as a fast route, and anyone reading a reserved turbine slot as power on a date.
Sources: GE Vernova results releases and calls of 22 April, 22 July and 28 October 2026, its 2025 Form 10-K and Forms 10-Q to 30 September 2026; GE Vernova and Crusoe (22 July 2025) and Crusoe (18 March 2025); Howmet Aerospace's 2025 Form 10-K, Technology and Markets Day (10 March 2026) and releases of 6 August and 29 October 2026; Siemens Energy, 11 February 2026; Bloom Energy with Oracle (24 July 2025, 13 April 2026) and American Electric Power (November 2024), and its releases to 22 October 2026; Pennsylvania DEP plan approval 32-00457A (18 November 2025) and Homer City releases of 26 March and 20 October 2026; Meta's and Entergy's 2025 Forms 10-K, Entergy's Form 10-Q of 31 July 2026 and LPSC docket U-37425; EIA Short-Term Energy Outlooks of September and October 2026, Today in Energy (23 April 2026) and monthly gas data (30 September 2026); Kinder Morgan, 22 April and 21 October 2026; Transco's FERC application, docket CP25-10 (28 October 2024); Williams, 3 August 2026, and its Socrates project page; Ohio Power Siting Board case 25-188-EL-BGN (18 September 2025). Leads came from Leo Cui's videos on GE Vernova and the energy sector. Personal research, not investment advice.