Some AI data centers have stopped waiting for the grid and started making their own electricity with engines pulled off retired aircraft. TechCrunch AI reports, citing the New York Times, that these rebuilt engines now burn natural gas, and that xAI has already installed eight of them. Nobody considers them a permanent fix. But they show how far power supply has fallen behind the AI buildout.
Speed beats everything else right now
“It’s all about getting power as fast as possible,” Rystad Energy analyst Audun Martinsen told the Times. With jet-powered units, he said, “they can have a data center running in 24 months.”
Compare that with the grid. Hooking up 100 megawatts in Northern Virginia takes about seven years, according to the International Energy Agency. Start today and that’s roughly late 2028 versus 2033.
Those two numbers don’t measure quite the same thing, so treat the gap as a rough guide. What stands out is that two years now counts as the fast option.
An executive at Mitsubishi Power also told Bloomberg last week that most power capacity is expected to come on by 2031. Mitsubishi makes this equipment, so it has a close view of the market, but it’s also a seller and not a neutral forecaster.
Slow power is already costing deals
This has moved past a planning headache. Last month, Microsoft pulled out as the anchor customer for Nscale’s data center in Loughton, England, after the local utility said it couldn’t deliver enough power until the early to mid-2030s.
That’s the new risk for developers. You can have the site and the customer, and still lose the customer because of a utility’s timeline.
Who’s making the bet
- xAI: It installed eight converted turbines in Mississippi to help power Colossus 2 in Memphis, the Times reported, citing state permits and satellite imagery. The units come from ProEnergy, which rebuilds a GE engine commonly found on Boeing jets.
- OpenAI: It followed Musk’s lead in 2025 with gas turbines at its first Stargate site in Abilene, Texas.
- Crusoe: The startup building the Abilene campus ordered 13 of ProEnergy’s converted units in April. In September, it walked away from a $1.25 billion deal for 29 turbines from Boom Supersonic, though a spokesperson told TechCrunch it still plans to use turbines.
Washington is pushing the same way. The Trump administration is encouraging data centers to generate their own electricity, which makes on-site gas look less like a workaround and more like expected behavior.
The trade-offs come due later
Converted jet engines are less efficient and pollute more than turbines designed for conventional gas plants. It also takes more of them to make the same amount of power, according to the Times.
Some experts quoted in the coverage think efficiency and environmental concerns could grow in importance within a year or two, which would make these engines a short-term play. TechCrunch doesn’t name those experts, so take the timing as informed opinion.
Where this goes in the next 1-3 years
This part is my read, not the reporting:
- Through 2027: Power availability drives site selection more than land or tax breaks. Expect more on-site gas orders, and more competition for the rebuilt units.
- Around 2028: The first jet-powered wave is running at scale. Fuel bills and emissions become harder to ignore, especially for companies with public climate goals.
- 2029 to 2031: If Mitsubishi’s timeline holds, newer turbines and grid connections start to replace these units as primary power. Some engines probably stick around as backup.
What to do now
- If you buy compute: Ask providers where the power for your capacity comes from and when it’s contractually available. An announced campus isn’t the same as energized capacity, and Loughton shows why.
- If you depend on one region or vendor: Line up a second source. A utility delay can move your timeline by years.
- If you’re building or financing sites: Model fuel costs, permits and emissions rules across the full life of the project, and plan how you’d swap the turbines out later.
- If you report on sustainability: Find out now whether your AI vendors run on-site gas, because it could show up in your own emissions questions.
These engines have already been retired once. How soon they get retired again depends on how fast the grid catches up, and right now the people selling the equipment are pointing to 2031. Full details are in the original TechCrunch AI report.