Data centers turn to commercial aircraft jet engines as AI power crunch bites (www.tomshardware.com)

Faced with multi-year delays getting new grid capacity, U.S. data centers are deploying aeroderivative gas turbines — essentially retired commercial jet-engine cores mounted in trailers — as fast, modular “bridging” power. Vendors like ProEnergy and Mitsubishi Power are repurposing GE CF6-80C2 and LM6000 cores and Pratt & Whitney‑derived FT8 designs into self-contained units that can spin up in minutes and deliver tens of megawatts each (vendors advertise ~34–48 MW per unit). OpenAI’s Stargate project, for example, is installing nearly 30 LM2500XPRESS units (~34 MW apiece) to cold‑start servers in under ten minutes. These aero‑derivatives have pedigree in military and offshore use but are appearing at data centers at scale for the first time, reflecting tight U.S. power supply and the outsized, immediate energy needs of large AI builds (single projects can demand 100+ MW). Technically, these units run in simple‑cycle mode — fast to start but far less thermally efficient than combined‑cycle plants because they don’t recover exhaust heat. They typically burn natural gas with distillate (No. 2 diesel) backup, often truck‑delivered, and require emissions controls like selective catalytic reduction to meet NOx limits. That makes them costly, noisy, and more carbon‑intensive per MWh, but viable short‑term stopgaps while substations or long‑lead resources (including modular nuclear or grid upgrades with multi‑year lead times) come online. Unless adoption scales enormously, they’re unlikely to move aviation‑fuel markets significantly because most run on pipeline gas or diesel rather than jet fuel; however, widespread use would raise fuel logistics, emissions, permitting, and sustainability tradeoffs for the AI community.