AI data centers powered by engines from old airplanes

Connecting 100 MW to the grid in Northern Virginia takes about seven years, so AI data center operators are generating electricity themselves using converted jet engines. The lack of grid capacity is starting to slow the development of AI infrastructure.

AI data center operators in the US are installing jet engines from retired passenger aircraft next to server facilities, converted to burn natural gas. The findings of the “New York Times” were reported by TechCrunch on November 3. Such units are used, among others, by Elon Musk’s xAI: eight turbines in Mississippi power the Colossus 2 data center near Memphis. They were supplied by ProEnergy, which rebuilds GE engines from Boeing aircraft.

The reason is time. Audun Martinsen from the analytics firm Rystad Energy says: “The main point is to gain access to energy as quickly as possible.” According to him, a data center with this kind of power supply can start operating within 24 months. Connecting 100 MW to the grid in Northern Virginia takes about seven years, according to the International Energy Agency.

Crusoe, which is building the first Stargate campus for OpenAI in Abilene, Texas, also uses ProEnergy’s services. The company ordered 13 units with a combined capacity of 650 MW, for delivery by summer 2027. In September, Crusoe withdrew from a $1.25 billion deal for 29 turbines from Boom Supersonic, a startup building a supersonic passenger aircraft that began selling a version of its engine as a power source for data centers. A Crusoe spokesperson then assured TechCrunch that the company was still betting on turbines, just from a different supplier. A similar offer comes from New York-based FTAI Aviation, which converts engines from retired Boeing 737s and Airbus A320s.

Converted engines are less efficient and emit more pollution than turbines designed for gas power plants. More of them are needed to produce the same amount of energy. Experts consider them a short-term solution. The Southern Environmental Law Center has challenged the environmental permits for xAI’s turbines in Mississippi.

The shortage of capacity is already affecting contracts and prices. In October, Microsoft withdrew from its role as the main customer of Nscale’s data center in Loughton in the UK. The local energy supplier will not provide sufficient capacity there before the early or mid-2030s. In the 2025/26 capacity auction, in which PJM — the grid operator covering, among other areas, Northern Virginia — buys generators’ readiness to supply electricity, the price rose by 833 percent. Independent market monitor Monitoring Analytics attributed 63 percent of that increase to data centers. That means $9.3 billion in additional costs for customers in the region.

For companies using the cloud, this means that access to energy can affect the pace of bringing new computing capacity online and the cost of services.