On Tuesday, The New York Times reported that eight converted jet engines are generating electricity for one of the largest AI campuses in the country — the clearest evidence yet that bring-your-own-power has crossed from improvisation into strategy. The engines, rebuilt by Missouri-based ProEnergy from retired General Electric aircraft turbines, sit at a power plant in Southaven, Mississippi, supplying xAI’s Colossus 2 data center across the state line in Memphis, Tennessee. They are a workaround: a way to run a gigawatt-scale campus while its grid connection is still years away.
First, the geography, because some secondary coverage has already gotten it wrong. The data center is in Memphis, Tennessee; the turbines are in Southaven, Mississippi, just over the state line in DeSoto County. There is no facility in “Memphis, Mississippi,” and the distinction is not pedantry: the turbines were permitted by Mississippi regulators, and the legal challenges are proceeding in Mississippi courts. Where a generator sits determines who permits it and whose air-quality rules apply.
What the machines are
The eight units are built around the GE CF6-80C2, an aircraft-engine core from Boeing 747s. ProEnergy strips the bypass fan and reworks the core into its PE6000 aeroderivative generator set, which burns natural gas instead of jet fuel and is rated at 50 megawatts per unit. Eight of them add up to roughly 400 megawatts — a power plant assembled from retired airframes.
The Mississippi air permit covering the units was issued on March 11, 2026. The eight ProEnergy units are permitted to emit carbon monoxide at four parts per million — double the limit applied to the plant’s 33 other turbines — and their permitted limits for volatile organic compounds run 25 percent higher. Converted aircraft engines are less efficient and more polluting per megawatt than turbines designed for stationary plants; the permit reflects that, in writing. The NAACP and the Southern Environmental Law Center have appealed the permit; a separate NAACP suit over temporary turbines at the same site was kept alive by a federal judge on October 30.
The eight converted engines are not the end of the buildout at Southaven: a July 30 state order describes 41 permitted turbines as the plant that “will be the permanent power source” for xAI’s computing operations, and the company is preparing a permanent gas plant of about 1.2 gigawatts. The jet engines are the bridge to a gas plant that is itself a bridge to a grid connection with no published arrival date.
The arithmetic that makes a jet engine a power plant
The reason operators are buying retired aircraft engines is arithmetic, and the numbers are stark. According to figures cited by the International Energy Agency, connecting a 100-megawatt data center to the grid in Northern Virginia takes about seven years; our interconnection explainer put the general range at three to seven years when major upgrades are needed. A data center using jet-powered units can be running in 24 months, Rystad Energy analyst Audun Martinsen told the Times. That gap is the market: five years of waiting is worth more than the efficiency penalty of a machine designed to fly.
The Times reported that OpenAI used gas turbines at its first Stargate campus in Abilene, Texas, in 2025. Crusoe, the Denver-based company building that campus, has gone further: on April 16, ProEnergy announced a contract to deliver 13 PE6000 generator sets — 650 megawatts in total, at 50 megawatts each — to power Crusoe’s upcoming hyperscale AI factories, with delivery by summer 2027. The contract covers Crusoe’s pipeline of projects, not a single named site — a developer buying 13 turbine sets at once is building a fleet.
The second supplier in this market arrived through a different route. FTAI Aviation, a New York-based aircraft engine leasing and repair company, launched a power business late last year to convert its CFM56 engine inventory — the workhorses of the Boeing 737 and Airbus A320 fleets — into aeroderivative turbines of about 25 megawatts each. On July 22, its J&F Power Systems joint venture with the Jereh Group announced an initial $1.465 billion order for Mod-1 mobile gas turbine generator sets from an unnamed international cloud service provider, under a five-year master supply agreement, with batch deliveries through November 2027. The figure is often rounded to $1.5 billion; the announced number is $1.465 billion, the customer unnamed. FTAI says it expects to produce more than 100 power turbines a year once production ramps.
The demand side is not mysterious: new purpose-built turbines carry lead times measured in years, and the queue for them is crowded with developers, utilities, and power producers all ordering at once. Venture Global’s $17.2 billion in frame agreements for gas turbines, disclosed in mid-October, shows how much of that future supply is already spoken for. A retired aircraft engine is, by contrast, available now — sitting in a boneyard or a leasing company’s warehouse, its core still serviceable. For a developer whose binding constraint is the calendar, the choice between a purpose-built turbine in 2030 and a converted jet engine in 2027 is barely a choice.
That is what makes this a strategy rather than a stopgap. A stopgap is a machine you rent while you wait; a strategy is a supply chain you build, a fleet you order, a master supply agreement you sign for five years. Both suppliers are selling a product line — manufacturing, test-fitting, service contracts — not one-off conversions. When the two largest suppliers in a niche market are both scaling production lines and signing multi-year agreements, the niche is a segment.
The federal record, corrected
The federal context has also been misreported, and the correction matters for what operators should expect from Washington. There was no August federal push requiring data centers to build their own generation. The August 26 executive order concerned foreign-produced bulk-power equipment — transformers, inverters, batteries, and turbines tied to covered foreign entities — and directed the Energy Department to write implementing rules by December 24. It is a supply-chain security measure, not a self-generation mandate — and it may complicate the equipment picture with new sourcing constraints.
The self-generation “mandate” came later, and in a different form. On October 21, President Trump wrote on Truth Social that new and large data centers were “mandated by me to make their own electricity” and should give or sell any excess back to the local community or grid. As this desk documented the following day, there was no accompanying executive order or federal rule, and none has been published since. The post escalates a voluntary Ratepayer Protection Pledge launched on March 4, under which seven technology companies — Amazon, Google, Meta, Microsoft, Oracle, OpenAI, and xAI — agreed to build, bring, or buy the new power their data centers need and cover the cost of grid upgrades. The pledge has no enforcement mechanism; the post dresses it in the language of a mandate without supplying one.
Meanwhile, the federal regulator has moved in the opposite direction from a self-generation requirement. On October 9, the Federal Energy Regulatory Commission rejected most of PJM’s plan to make new data centers bring their own capacity or face curtailment first in emergencies (Docket ER26-3515). The commission kept PJM’s definition of a large load — 50 megawatts or more at a single site — and a registry to track those loads, effective October 12, but threw out the Interim Resource Adequacy Service and the bring-your-own-new-capacity provisions as reaching into retail electricity policy the Federal Power Act leaves to the states. The market is moving toward self-generation faster than the regulator is willing to require it.
What self-generation does to the queue
The interconnection queue is the institution this trend is quietly dismantling, and the consequences cut in both directions. Every campus that powers itself removes a gigawatt-scale load from the queue, shortening the wait for the loads behind it. Cleanview, an energy research firm, counted more than 100 gigawatts of data-center-owned power plants announced as of October 21, up from about 90 gigawatts in June. If even a fraction is built, the queue stops being the industry’s binding constraint — and utilities’ load forecasts, which assume those campuses will eventually connect, start to look overstated.
But bypassing the queue has costs that do not disappear when the load does: the transmission and generation a utility builds for a large load is paid for, over decades, by the customers on the system. A campus that self-generates but keeps a grid connection for backup still relies on the system while contributing less revenue — the cost-shift problem our bring-your-own-power explainer flagged, now operating at hundreds of megawatts per site. And a campus that never connects leaves the utility planning for load that never arrives, and infrastructure sized for a future self-generation has quietly canceled. The grid will have to plan for customers who might build their own supply — and make the grid alternative attractive enough that most of them choose it. Right now, on speed alone, it is losing.
There is also the bridge problem, which Southaven illustrates: the eight converted engines were the bridge, the 1.2-gigawatt permanent gas plant is the destination, and the grid connection it was meant to bridge to has no published arrival date. Bridges have a way of becoming permanent, and regulators are beginning to ask whether “temporary” gas generation at data centers is truly temporary. A converted jet engine is, by design, a machine with a second life; the question is whether that life ends when the grid arrives.
Reliability, fuel, and the business of running a plant
The operational case for aeroderivative turbines is genuinely strong: they are dispatchable, starting fast, ramping quickly, and following a load that changes by the hour. That is the property that matters most for a data center, whose load is flat and always-on but whose tolerance for interruption is effectively zero. It has neither solar’s intermittency nor batteries’ duration problem: it burns gas and makes power, on demand, for as long as the fuel flows.
The fuel is the constraint. A gigawatt of gas-fired generation burns on the order of 7 million cubic feet of gas a day, and a site at that scale needs a pipeline lateral with firm capacity — itself a potential bottleneck. The maintenance burden is real, too: aeroderivative turbines need regular overhauls, and a fleet of dozens of converted engines is a power-plant operations business. The operators buying these machines are becoming power plant operators, whether they planned to or not. Vertiv’s $1.45 billion acquisition of a microgrid controls company last month, covered on this desk, is the other half of the same story.
The emissions trade-off, stated plainly
None of this is free, and the price is paid in emissions. A converted jet engine produces more pollution per megawatt than a purpose-built combined-cycle plant, and the Southaven permit says so in its own numbers. The campuses running these machines sit in communities that did not ask for a power plant, and the legal challenges in Mississippi are the predictable result. The industry’s climate commitments — clean-power pledges, renewable PPAs, net-zero targets — sit uneasily alongside a fleet of rebuilt 747 engines burning gas around the clock. Operators will frame the turbines as transitional, a bridge to cleaner on-site options or to grid power. But a bridge that runs for five or ten years at hundreds of megawatts is a material emissions source, not a footnote.
The honest version of the trade-off is this: the industry has chosen speed over efficiency, and deployment over cleanliness, because the cost of waiting is measured in the race to deploy AI. That is defensible in the abstract and contested at any particular site — and the permitting fights in Mississippi are where it is being fought.
What to watch
Four things will determine whether jet-engine generation stays a bridge or becomes the architecture. The first is the delivery schedule: whether ProEnergy’s 13 Crusoe units arrive by summer 2027 as contracted, and whether J&F Power’s batches reach its unnamed cloud customer through November 2027.
The second is the Mississippi litigation. The appeals against the Southaven permit, and the federal case over the temporary turbines, will establish how much latitude regulators give converted-engine generation — and whether the emissions premium becomes a permitting liability.
The third is the federal picture. The Energy Department’s implementing rules for the August executive order are due by December 24, and will show whether the supply-chain security regime reaches the equipment developers are buying. And the question left open by the October 21 post — whether anything with legal force follows it — is unanswered.
The fourth is the queue itself. If interconnection timelines shorten, the case for buying a retired jet engine weakens with every month saved. If they do not, the converted-engine fleet will keep growing, and the grid will keep losing its largest new customers to a power source that was, until recently, flying passengers across the Pacific. The jet engine was never meant to be a power plant. That it now is one says less about the engine than about the grid.