Who Is Buying Used Power in America, and Why They Are Paying New Prices for It

A transaction that explains the market

In June, a Philippine LNG developer sold two Siemens F-class gas turbines and a steam turbine that had sat in crates at the Port of Pagbilao for a decade. They had never been fired. The buyer was an Indiana coal company, Hallador Energy, and the price was $350 million, which Energy World Corp told the Australian stock exchange was nearly four times what it had paid for the equipment ten years earlier. Hallador did not buy a power plant. It bought crates. By the company's own estimate, shipping the equipment to the United States and putting it through a Siemens recommissioning adds about $100 million, and after that comes a 22-month engineering and construction program that has not yet been awarded. First power is late 2028 at the earliest. A new order placed the same week would ship in 2031.

That trade is the whole market in one transaction. A seller quadrupled its money on equipment it never used, and a buyer paid it gladly, because the alternative was a factory slot three years further out. In the last turbine boom, in 2001, an online auction of twenty used LM2500-class units drew no bids at all. Today the discount to new is gone, and it has gone for every class of surplus equipment: the unused machine at a port, the retired frame at an industrial site, the aeroderivative coming off a bridge contract. What a buyer pays now turns on one question, which is how soon the machine can make power, and the answer to that question is set less by whether the machine has run than by whether it has been through a shop and has a package, a warranty and a date. Idle generation equipment has never been worth more. The rest of this piece is about who is buying it in the United States. The buyers abroad, from failing grids to reconstruction, are a separate piece.

Quick takeaways

• The surplus-to-new discount is gone. In June an unused pair of F-class turbines sold for nearly four times their original cost and will cost the buyer about $450 million to own delivered and recommissioned, with first power no earlier than late 2028. A new frame ordered today ships in 2031.
• The OEM queue is the market. GE Vernova has 122 GW under contract and expects 2026 pricing about 50 percent above 2025; Mitsubishi says a large frame ordered now arrives around 2031; Siemens Energy quotes three years or more.
• Data centers bought aeroderivatives first because they install in weeks, follow a GPU load, and can leave. As campuses pass a gigawatt, the buying has moved to 100 to 250 MW frames, which is exactly what the factories cannot deliver this decade.
• Fermi's 19-turbine fleet is the template: three contracted new, six acquired through a stranded LNG project, seven leased, three bought used from an industrial site.
• The winners are the developers who look where nobody else is: flare gas, retired jet engines, cancelled projects, closed industrial sites.
• Utilities are buyers now too. Coal retirements fell to 2.6 GW in 2025 against 8.5 GW planned, and Entergy Louisiana is seeking approval to pay $1.5 billion, plus $309 million of upgrades, for an existing 1.26 GW plant to get megawatts today.
• The window is probably five years wide. It closes as the OEMs add capacity; it only ever applied to equipment with documented provenance and a credible shop behind it.

The queue that made a market

Every secondary market is a shadow of a primary one, and the primary market for gas turbines has stopped functioning as a market in any normal sense. It is a queue. GE Vernova reported on October 28, 2026 that its gas equipment backlog plus slot reservations reached 122 GW in the third quarter, up from 116 GW three months earlier, and that it expects at least 130 GW by year end. On the call, the company said it sold multiple aeroderivative slots for 2032 delivery in the quarter, and that it expects gas orders booked this year to be priced about 50 percent higher per kilowatt than orders booked in 2025. Siemens Energy's gas turbine backlog stands at 69 GW and its CFO puts lead times at three years or more. Mitsubishi Power's Asia Pacific chief said in late October 2026 that a large frame ordered now delivers around 2031.

The scale of what happened is hard to overstate. Between 2018 and 2023 the world ordered somewhere between zero and eleven gigawatts of gas turbines a year. In 2025 it ordered 51 GW, the most since 2000, and the first quarter of 2026 alone added 18 GW. JPMorgan counted a record 38 GW of global orders in the second quarter, half of it American. The factories that survived the lean decade can build perhaps 60 to 70 GW a year between them, and Wood Mackenzie counted 110 GW of orders outstanding at the end of 2025. The OEMs are expanding, and they are expanding carefully, because every one of them remembers 2002, when the last boom collapsed and left factories building machines for projects that had been cancelled.

Southern Company's chief executive has said publicly that the company pays non-refundable fees to hold its place in GE Vernova's line; one Kentucky utility paid $25 million to reserve a single turbine. Siemens Energy's head of gas services told Bloomberg that holding a slot for six months typically costs 10 to 15 percent of the purchase price. The all-in cost of a new combined cycle plant has gone from under $800 per kilowatt for projects built in 2022 to the $2,400 to $2,800 per kilowatt that NextEra's chief executive now cites, with construction stretching from four and a half years to six or more. NRG priced a 1.2 GW bring-your-own-power plant at $2,700 per kilowatt with a late 2029 start.

When the thing you need costs three times what it did and arrives in five years, the question for every buyer becomes the same: what exists now? The answer is the installed base. Gas turbines supply roughly a fifth of the world's electricity, and a surprising fraction of that fleet is idle, mothballed, stranded by a cancelled project, or running a few hundred hours a year at a plant whose owner would rather have the cash. That is the inventory. The queue is what gave it a price.

Why the data center bought the jet engine

The first wave of behind-the-meter data center power in the United States was built on aeroderivatives, and the reason is not romance about aviation. It is the calendar. A GE LM2500 or LM6000 is a modified aircraft engine bolted to a generator. It arrives on a trailer, sits on a pad instead of a foundation, and starts in about five minutes. GE Vernova's own data center literature says a TM2500 can be installed in roughly two weeks and that a developer can begin at 20 or 30 MW and add units as racks fill. A heavy frame machine of the same vintage needs a concrete foundation, a crane the size of a building, and an enclosure that takes months to erect. For a developer whose tenant is paying by the megawatt-month, that difference is the whole business case.

The second reason is that aeroderivatives behave like batteries with a fuel line. A GPU campus does not draw a flat load. Training runs ramp hard and drop hard, and a large frame turbine that is forced to follow that profile runs inefficiently and wears fast. The aero was designed to spool up and down thousands of times a year on a wing. When GE Vernova announced Crusoe's order of 29 LM2500XPRESS units for the Stargate campus at Abilene, the company's own pitch was the start-stop behaviour, dual fuel, and the fact that each 35 MW unit displaces about a dozen diesel gensets. That is nearly a gigawatt of aeroderivative capacity on one site, and it is the template everyone else copied.

The third reason is speed through the permitting process, and xAI showed how far that can be pushed. A trailer-mounted turbine is still a stationary source under the Clean Air Act, and EPA's January 2026 rule says so in terms, but it arrives, connects and runs on a timescale that a frame unit's foundation and enclosure cannot match, and in Memphis the machines were generating long before the permits caught up with them. Within months of breaking ground in Memphis it had dozens of Solar SMT-130 and GE TM2500 units feeding Colossus, and by September 2025 roughly 460 MW of trailer-mounted turbines were installed or under construction across Memphis and a former Duke Energy plant site across the state line in Southaven. Mississippi has since permitted 41 turbines and up to 1.2 GW at that site. The machines were making power long before any frame unit could have cleared a foundation design review.

There is a fourth reason that matters most to anyone holding used iron. An aeroderivative can leave. The data center operator who installs 300 MW of LM6000s as bridge power while a utility interconnection crawls through a five-year queue is not buying a stranded asset. In three years the units go on a truck to the next campus, or back to market. ProEnergy has built a business on exactly this logic, selling 21 of its PE6000 packages, each roughly 48 MW on a refurbished CF6 core pulled from a retired 747, to two data center projects totalling more than a gigawatt, explicitly as five to seven year bridge capacity, with delivery in 2027 against three to five year waits for a new LM6000. Twenty-year-old aircraft engines are now a power generation supply chain. That tells you how far the market has moved.

Bring your own power, and bring something bigger

Once a campus is designed at a gigawatt or more of steady baseload, 35 MW units stop making sense: the fuel penalty of simple-cycle aeros against a combined-cycle frame is real, the maintenance footprint of thirty machines is real, and the footprint of thirty enclosures on a site that wants to use its acreage for halls is very real. Enverus expects 22.5 GW of behind-the-meter data center demand to be built between 2026 and 2030, almost all of it on gas. Cleanview's census of behind-the-meter projects counts about 90 GW of planned self-generation across 59 campuses, more than a quarter of all planned US data center capacity, of which roughly 2 GW is actually operating. NextEra's chief executive told investors in December that the market had moved to bring your own generation, and that the company had locked in 4 GW of GE Vernova slots to serve it. The Texas legislature set the terms for it in law with SB 6, which governs how large loads connect, curtail and co-locate, and FERC ordered PJM in December to build a tariff for co-located load.

The frame unit is back, and the frame unit is where the secondary market gets interesting, because a 100 to 250 MW heavy-duty machine is exactly what the OEMs cannot deliver before the end of the decade.

No company illustrates this better than Fermi America, whose campus outside Amarillo is planned at 17 GW, 11 GW of it gas. Fermi's filings are the most detailed public record of how a developer actually assembles power when the queue is closed. As of its second quarter investor deck, Fermi had 19 turbines on hand or contracted, none yet operating, for about 1,266 MW of simple-cycle capacity. Only the three F-class frames came through a conventional OEM contract.

The SGT-800s and their heat recovery boilers were acquired through a Texas LNG developer and shipped to Houston on seven vessels from four countries, where Fermi staged them in a foreign trade zone to beat tariff deadlines. The three 6Bs, which the company calls the Jersey Boys, ran for years at a heavy manufacturing site in New Jersey before Fermi bought them, had them refurbished and repainted in Houston, and trucked them to the Panhandle, where they arrived in mid-October 2026. The TM2500s are a twenty-year lease from a rental fleet. Lenders financed the Siemens turbines alone with a $500 million facility, which tells you how bankable the equipment has become, and Fermi's own announcements now describe OEM lead times as running into the early to mid 2030s.

The procurement playbook Fermi wrote is now being copied across the industry. Hallador pulled 460 MW of unused F-class out of a stranded Asian LNG project for Indiana. xAI bought a retired 640 MW Duke Energy plant site in Mississippi, at a price that was not disclosed, for its transmission infrastructure, then filled it with trailer-mounted units while it pursued a permit for 1.2 GW. Elon Musk confirmed a report that the company intended to buy an entire power plant overseas and ship it to Memphis. Pacifico Energy ordered eight Ansaldo AE64.3A frames for a Texas campus, a model that had not been sold in the United States in a generation, because Ansaldo had slots and the big three did not. Applied Digital is building 1.2 GW of steam plants with new Babcock & Wilcox boilers, an older technology rather than a secondary-market buy, because boilers can be built.

What the serious buyers have in common is that they stopped asking what is on the price list and started asking what is on the ground. A 1990s F-class at an industrial site, a cancelled project's crated equipment at a port, an oil company's retiring cogeneration plant: all of it is now in play, and the data center developers were the first to understand that.

The ones who go one step further

Crusoe is worth dwelling on, because its history is a parable about where power actually comes from when nobody will sell it to you. The company was founded in 2018 in Denver by two people who had noticed that oil producers in North Dakota and Wyoming were burning off associated gas at the wellhead because there was no pipeline to carry it. Crusoe put generators and shipping containers of computers next to the flare stacks and turned waste gas into Bitcoin. By October 2021 it had 44 of these sites running and was cutting about 10 million cubic feet a day of flaring. By 2025 that business had grown to more than 250 MW across seven states and Argentina, and Crusoe sold it to NYDIG so it could put the same instinct to work on AI. The result is Abilene: a 1.2 GW campus whose first two buildings went live in September 2025, with a 350 MW on-site gas plant and substation built in under six months, financed by $11.6 billion of debt against a fifteen-year Oracle lease.

The lesson is not that flare gas is the future. It is that Crusoe's founders had spent seven years looking at energy that everyone else had written off, and when the AI buildout arrived they already knew how to think.

The same reflex is visible everywhere the market is actually moving. ProEnergy saw a thousand CF6 jet engines coming off retiring 747s over the next decade and built a power plant around them. Crusoe, in turn, bought thirteen of those packages for 650 MW. FTAI did the same with smaller engines, and BTIG estimates each 25 MW unit sells for about $25 million, roughly $1,000 per kilowatt for a converted aircraft engine. Hallador looked at a Philippine port. xAI looked at a decommissioned Duke site. Energy World Corp's chairman, explaining why he had just quadrupled his money on crated turbines, credited American data centers. Developers have been reported hunting refurbished 1990s frames in Alberta and along the Gulf Coast, and some are buying turbines on speculation with no site attached.

This is the real shape of the secondary market in 2026. It is not a classifieds page of surplus peakers. It is a race among developers to find generation that is not where anyone is looking: industrial sites that never thought of themselves as power producers, projects cancelled with equipment already delivered, plants whose grids no longer need them. The examples in this piece are the ones that became public. Most do not. The developers who will be energized first are not the ones with the best lawyers in the interconnection queue. They are the ones who found the turbine nobody else knew was for sale.

The utilities are shopping too

The secondary market used to be a place utilities sold into. Now they are buyers, and the reason is that the plants they planned to retire cannot be replaced on schedule.

The retirement calendar has been rewritten. The EIA reports that US utilities retired 2.6 GW of coal in 2025, the smallest amount in fifteen years, against 8.5 GW planned. TVA's board voted in February to postpone the closures of Cumberland and Kingston, nearly 3.9 GW between them, with no new dates. Georgia regulators let Georgia Power run Bowen and Scherer, about 4 GW, into the late 2030s against roughly 8,500 MW of expected load growth. Duke pushed Belews Creek, Cliffside and Marshall out two to four years while asking for 9.7 GW of new gas by 2033. PacifiCorp's Hunter plant now runs to 2042. Talen's Brandon Shores, a 1.3 GW coal plant that was supposed to close in mid-2025, is being paid about $145 million a year in fixed costs by PJM to stay open until 2029, with another $35 million for the neighbouring Wagner units. The Department of Energy has issued more than forty emergency orders since May 2025 keeping some 4.4 GW of retiring plants online, including the 1,560 MW Campbell plant in Michigan, which Consumers Energy says has cost its customers $295 million to keep running. In September the DC Circuit vacated the first of those orders, which means the legal floor under this strategy is cracking even as the engineering case for it grows.

Why are utilities paying hundreds of millions to run fifty-year-old boilers? Because the alternative is a turbine slot in 2031. Grid Strategies now forecasts 166 GW of peak load growth by 2030, more than half of it data centers, against a 2022 forecast of 24 GW. NERC's latest long-term assessment adds 224 GW to summer peak over ten years while 105 GW of retirements are still on the books. A utility cannot close that gap with new frames before 2030, so it has three options: run the old plant, buy someone else's, or find equipment that already exists. Entergy Louisiana chose the second, asking regulators to approve a $1.5 billion purchase of the 1.26 GW Cottonwood plant in Texas plus $309 million of upgrades, with closing targeted for January 2027, and its CFO explained the premium with two words, megawatts today. Vistra paid $4 billion for 5.5 GW of Cogentrix plants and LS Power paid $5 billion for 4.4 GW from Constellation, and Enverus notes that operating gas plants now trade at roughly half the cost of building new, and that gas M&A valuations have roughly doubled since 2024. Hallador, a coal miner that became a utility supplier, chose the third option and went to the Philippines.

The quiet consequence is on the service side. A fleet of 1990s frames that was meant to be scrapped by 2030 is now expected to run to 2040, and every one of those machines needs hot gas path parts, rotors, generators and controls that the OEMs are too busy building new units to supply. The secondary market for whole turbines is only the visible part. Beneath it is a secondary market for everything inside them, and the utilities that extended their plants are the biggest buyers of that too.

What the iron is worth

There is no index for used gas turbines, and a published asking price is only an asking price, so treat any single number with suspicion. What the public comps do show is that the equipment price is the smallest part of the answer. The question a developer with a tenant actually asks is what a megawatt costs on the day it first makes power, and that depends on the state of the machine and the date more than on the sticker.

Read down the equipment column and the Hallador comp looks like the bargain: unused F-class for less than a new one. Read across the row and it is not. Hallador's machines sat in crates in a tropical port for a decade, and the $760 per kilowatt that bought them is the start of the bill, not the end. The company's own estimate adds about $100 million for shipping and a Siemens recommissioning, which brings the equipment to roughly $980 delivered, and the first revenue is still two and a half years away, after a 22-month engineering and construction program that has not been awarded. Carry that for two years and the equipment alone is north of $1,000 per kilowatt before anyone pours concrete. Unused is not the same as ready.

That is why the market has stopped pricing on condition and started pricing on the date. A refurbished frame that has been through a shop with a documented record, has been repackaged, carries a warranty and has a 2028 operating date is a different product from a crate at a port, and in 2026 it is clearing at prices in the same range as what the OEMs quote for a new machine with a 2031 date. We expect that to hold for as long as the two dates are three years apart. It will not hold forever. GE Vernova is adding capacity toward 30 GW a year by 2030, Siemens Energy and Mitsubishi are both roughly doubling output, and Wood Mackenzie expects orders to peak this year. The window in which a 1990s turbine is worth what it cost new is real, and it is probably five years wide. Owners who are waiting for a better price are waiting for a market that has already arrived.

Two cautions for buyers, because the prices only hold for equipment that is what it claims to be. First, provenance: Ecuador contracted $706 million of emergency generation in 2024 and not one of those engines ever produced a megawatt. Second, the shop: Hallador's machines go through a refurbishment program with a documented record before they reach Indiana, and Fermi's 6Bs went through a certified facility before they went anywhere near a pad. The spread between a turbine with a documented overhaul and a turbine with a photograph is wider now than it has ever been, and it should be.

What this means

If you own idle generation, you are sitting on something the largest companies in the world cannot buy from the factory. That includes the obvious owners, utilities with mothballed peakers and IPPs with stranded plants, and it includes the ones who have never thought of themselves as being in the power business, like the industrial site in New Jersey that had three Frame 6Bs out back. The equipment has a buyer in Texas, a buyer in Indiana and a buyer in Louisiana, and the buyers abroad, which the companion piece covers, are bidding against them. For the first time in the history of the industry those buyers are competing with each other and with the OEM's own price list.

If you are a developer, the lesson of the last two years is that the interconnection queue and the OEM queue are the same queue wearing different clothes, and nobody wins by standing in it. Crusoe did not. xAI did not. Fermi did not. The campuses that are energized in 2027 will be the ones whose developers treated power as a sourcing problem rather than a procurement problem, went to find equipment that already existed, and were willing to do the engineering to make it fit. The ones that are energized in 2031 will have ordered new.

The market for used power generation equipment has existed for as long as there have been turbines. It has never mattered this much, and its assets have never been worth more. That is not a forecast. It is a description of a transaction that closed in June, and of the dozens behind it that have not been announced.