THE CRUNCH
Elon Musk claimed two weeks ago that he was "highly confident" SpaceX could build 10 GW of data centre capacity in 2027 alone. That sounded like classic Musk hyperbole, but the analysts at Semianalysis have published a detailed assessment arguing the target is realistic, if everything goes right. Adding 10 GW on top of an existing 3 GW would mean SpaceX built as much capacity in about a year and a half as Google accumulated over its entire existence, which would make it the fastest-building hyperscaler on the planet.
The evidence for the pace is visible from the air. Colossus 2's power generation capacity went from 0.4 GW in January to a projected 2.4 GW by the end of the year, and Semianalysis tallies roughly 9 GW of potential capacity across the sites it tracks, including Colossus and Starbase facilities. As of 25 September, Musk said Colossus 2 already had 550,000 GPUs working.
Power is the real bottleneck, and SpaceX's workaround is unorthodox. The Colossus 2 campus has 89 turbines delivering about 2.5 GW plus 1,377 Tesla Megapack batteries, but emissions rules will eventually cap on-site turbine use at roughly 1.24 GW. A Mississippi agreement lets the company run the extra turbines temporarily, with a promised phase-out by July 2027, and the analysts expect batteries and grid supply to fill the remaining gap. The turbines sit on truck flatbeds as mobile generation, and Semianalysis suspects Tesla is building electrical equipment in factory modules ready to install, which would let SpaceX skip the supplier lead times other hyperscalers face.
Built is not the same as working, though. Zoning, permits, grid connections, water supply and public opposition have stalled builds everywhere, and Microsoft's Satya Nadella has said plainly: "You may actually have a bunch of chips sitting in inventory that I can't plug in. In fact, that is my problem today." Even if Colossus 2 finishes on schedule, replicating that speed elsewhere is not guaranteed. Still, Semianalysis notes two data centres under rapid construction at Starbase, suggesting the approach can scale, and even a miss on the 10 GW figure would likely leave SpaceX outpacing everyone else.