Why Some AI Data Centers Are Running on Old Jet Engines

Some AI data centers are now making their own electricity with gas turbines that started life as jet engines. The developers are doing this because connecting to the power grid can take years.
Data centers are the large buildings full of computers that train and run AI systems. Gas turbines are machines that burn fuel to spin a generator, which produces electricity.
The engines come from retired aircraft and are rebuilt to burn natural gas. ProEnergy supplies converted turbines made by rebuilding a GE engine often found on Boeing jets. FTAI Aviation turns CFM56 engines into 25 MW turbines for AI and data center power. FTAI President David Moreno has said the conversion takes the company 30 to 45 days.
The customers include xAI, which installed eight converted jet-engine turbines in Mississippi to help power its Colossus 2 data center in Memphis. Crusoe, a Denver-based AI data center startup that had recently raised $3.9 billion as of September 2026, ordered 13 of ProEnergy's converted generating units in April for its AI factories, including its campus in Abilene, Texas.
Speed is the main draw. Rystad Energy analyst Audun Martinsen said speed is the main attraction of converted jet engines, and that jet-powered units can have a data center running within 24 months. The grid route is much slower. According to the International Energy Agency, connecting 100 megawatts to the grid in Northern Virginia takes about seven years.
Grid limits are changing other plans, too. Microsoft pulled out as anchor customer for Nscale's data center in Loughton, England, after the local utility said it could not deliver enough power until the early-to-mid 2030s. In September 2026, Crusoe walked away from a $1.25 billion deal to buy 29 natural gas turbines from Boom Supersonic.
The trade-off is in the machines themselves. Converted jet engines are less efficient than turbines built for conventional gas plants, and they produce more pollution for each unit of power. Making the same amount of electricity takes more of them. Aeroderivative turbines, which are based on aircraft engine designs, start up quickly and pack a lot of power into a small space. That makes them well suited to filling the gap while utilities catch up. GE Vernova offers onsite generation for data centers, ranging from smaller aeroderivative units to large heavy-duty systems.
The wider market is moving the same way. Siemens Energy reported record third-quarter sales, margins and orders in August 2026, driven by demand for gas turbines from expanding AI data centers. Reuters has reported that the rush for small gas turbines is set to raise long-term power costs and emissions for data centers. The International Energy Agency projects that global data center electricity use will roughly double by 2030. Batteries, meanwhile, are now cheaper than the natural gas turbines used at many data centers, according to TechCrunch's October 2026 coverage.
The broader context here is timing. A bridge generator that runs within two years can carry a campus until a grid connection arrives, much as a temporary bridge carries traffic while a permanent one is built. For an operator facing a seven-year wait, that difference can decide whether a project makes financial sense. The pollution and efficiency penalties, however, last as long as the machines run. Whether these units are shut down once grid power arrives, or kept running for steady, long-term power, will decide how much of the speed advantage is temporary.
In my view, the more useful question is not whether converted engines work, since they clearly do in service, but what gets built around them. Onsite generation changes where a data center can be sited, how its fuel is delivered, and how permits are handled. It also moves the main bottleneck from the utility's substation to the operator's own gas supply and maintenance schedule. If those are managed well, the approach can bring computing capacity online years sooner than the grid would allow, which is a real gain for training and running AI models.


