The new capacity is already spinning

Every conversation about powering the AI buildout eventually reaches the same machine: the gas turbine. And every one of those conversations now runs into the same fact. The big ones are spoken for.

The thesis: the fleet you already own is the new capacity. Large-frame turbines ordered now are next-decade assets. Until they ship, get built, and commission, the demand surge lands on the gas fleet already in the ground. That fleet will run harder and stay online longer, and it is already showing wear. Value moves toward reliability, outage execution, upgrades, parts, and the monitoring that sees failures coming. That shift touches owners, independent service providers, their suppliers, and OEM service arms alike.

Here is what the public record shows.

The line for new iron runs into the 2030s

The three largest heavy-duty OEMs all reported record or near-record gas books this summer.

GE Vernova ended the second quarter with 116 GW of gas equipment under contract, up from 100 GW a quarter earlier: 53 GW of firm backlog plus 63 GW of slot reservation agreements. It expects at least 125 GW by year-end. On the earnings call, management said it now has agreements signed into 2031.

Siemens Energy's Gas Services business reported a 69 GW backlog plus 26 GW of slot reservations at the end of its fiscal third quarter, with a book-to-bill of 2.65 in the segment's earnings release.

Mitsubishi Heavy Industries said its first-quarter large-frame orders are generally scheduled for delivery between 2028 and 2030. Utility Dive puts its large-frame backlog at 35 GW, up from 23 GW a year earlier.

None of the three counts "backlog" the same way. Slot reservations are not firm orders, and one company's gigawatts are not directly comparable to another's. But the direction is unambiguous. A frame ordered today is an asset for the next decade, not the next summer.

Not much new gas arrives this year

The near-term additions data says the same thing from the other side.

EIA reports developers plan a record 86 GW of new utility-scale capacity in 2026. Only 6.3 GW of that is natural gas: 3.3 GW of combined cycle and 2.8 GW of combustion turbines. The rest is mostly solar, storage, and wind.

The appetite for gas is plainly there. Berkeley Lab's Queued Up 2026 Edition counts 253 GW of natural gas capacity actively seeking interconnection at the end of 2025, up 86% in a year, even as solar, storage, and wind volumes in the queue fell. Most queued capacity historically never gets built, but the shift in what developers are asking for is hard to miss.

Demand is not waiting for the queue to clear, either. Enverus Intelligence Research forecasts 29.6 GW of behind-the-meter gas generation serving new industrial load from 2026 to 2030, with data centers about 88% of that demand. Reciprocating engines, small and medium turbines, and fuel cells take 61% of it, because they reach operation in roughly 18 to 24 months versus up to 80 months for a large-frame combined cycle.

So new load gets served three ways: smaller machines that ship faster, big frames that arrive later, and the fleet that already exists.

So the existing fleet stays on

Retirements are slowing. EIA says 4.6 GW of gas-fired capacity is scheduled to retire in 2026, almost 1% of the operating gas fleet, mostly older steam units, and that retirement delays may continue. Some units in that list were already held past earlier retirement dates.

The peaking fleet is also being asked for more. EIA's Electric Power Monthly shows net generation from gas-fired combustion turbines rising from 160,082 thousand MWh in 2024 to 163,873 thousand MWh in 2025, even as total gas generation dipped. That is a modest change on paper. It means more starts and more hours on machines that were built to cycle, and every start and fired hour moves a unit closer to its next inspection.

And the fleet is showing wear

NERC's 2026 State of Reliability found the weighted equivalent forced outage rate for conventional generation rose to 9.2% in 2025, against historical norms of 7–8%. NERC notes this was not driven by any single storm. It describes a broader baseline decline in the availability of much of the coal and combined-cycle fleet, spread across the year.

For combined cycle specifically, the technical assessment shows forced outage rate rising from 4.2% to 5.7%, with roughly 19,154 GWh more unavailable energy than the year before.

Read alongside the backlog numbers, this changes what reliability means. When new capacity is years away, an hour of forced outage on an existing unit is not just a maintenance metric. It is capacity the system cannot replace on short notice.

The aftermarket is repricing

The service side of the market is already responding.

GE Vernova told analysts that transactional service orders per unit continue to rise by double digits annually as customers invest in upgrades, larger outage scopes, and higher prices. The same call offered a look at the next wave: about 130 of its HA-class units are running today, against 325 commissioned or under contract. Each faces a major outage roughly every four years once in base-load service. Management framed that outage profile as a driver for the middle of the next decade.

Baker Hughes reported Gas Technology Services orders up 33% year over year to $1.31 billion in the second quarter. The same release lists an LNG award that includes fleet-wide gas turbine upgrades. Siemens Energy said both new-unit and service business grew substantially in Gas Services.

One counterpoint is worth keeping. MHI noted that strong new-equipment orders have slightly reduced the share of after-sales service in its revenue mix. The equipment boom and the service boom do not move in lockstep, and for now OEM factories are busiest building new machines.

What this means, by seat

Owners and operators. Availability is now a capacity asset. Outage planning, parts strategy, and the decision to uprate, extend, or retire a unit belong in capital planning, not just the maintenance budget. Expect more scrutiny from grid operators and capacity markets on forced outage performance as reserve margins tighten. Lock in outage windows and long-lead parts early, because many of them come from the same constrained supply chain as new units.

Independent service providers. As a category, ISPs face a larger, harder-run installed base and owners who want options alongside OEM long-term agreements. The constraint is less demand than capacity: skilled labor, repair throughput, and access to parts and engineering data. Providers that can show turnaround and quality with real performance data have the strongest case.

Suppliers behind the service market. Coatings, inspection, tooling, engineering, and replacement parts scale with outage volume, not new-unit volume. A fleet that runs more and retires later is steady demand. The same capacity limits apply: qualified people and qualified processes do not scale on a quarterly schedule.

OEMs. Service is the long tail of today's equipment boom. The units now in backlog become the service fleet of the 2030s. The near-term tension is allocation: factory capacity, engineering, and field teams are shared between new-unit commitments and the existing fleet's outages.

The common thread is that every seat at the table is competing for the same scarce inputs: people, parts, and outage windows.

Where AI fits: monitoring and operations

The practical AI story in this market is not autonomous plants. It is seeing problems earlier and planning around them.

A harder-run fleet produces more operating data, and forced outages are expensive precisely because they are unplanned. Condition monitoring, anomaly detection, and asset-performance software exist to turn some unplanned outages into planned ones. OEMs, independent providers, and software firms are all selling into that gap. Baker Hughes, for example, reported asset-performance, condition-monitoring and analytics awards in the same quarter its service orders jumped.

The back office matters as much as the turbine deck. Outage scheduling, parts forecasting, crew planning, and work-scope estimates are where schedule slips start. The same AI wave that is creating the load can compress that paperwork, provided the underlying data is clean and owners keep control of it.

A useful test for any AI claim in this space is simple. Does it move a forced outage into a planned one, or shorten a planned one? If not, it is a demo.

What Driftwood Club is watching

• OEM results. Third-quarter reports from GE Vernova and Baker Hughes, Siemens Energy's fiscal-year results in November, and MHI's next quarter. We'll track gigawatts under contract, slot conversions, and service order commentary.
• NERC's winter assessment and any updates to forced outage data.
• EIA's monthly generator inventory for changes in planned gas additions and retirement dates.
• Berkeley Lab and ISO queue data for how much queued gas actually advances.

Join the room

If you own or operate turbines, service them, supply the people who do, or build them, and especially if you are bringing AI into that work, this is the conversation Driftwood Club is for. Join Driftwood Club. Tell us which seat you sit in. We introduce members working on the same problems, and we send the sourced reading each week.

The next gigawatt is on order. The ones you need this decade are already spinning.