SpaceX Slows AI Data Center Buildout After Outages, Reshuffles Leadership

SpaceX is overhauling its rapid AI data-center expansion after an outage left multiple Colossus facilities in Tennessee and Mississippi without backup power and cooling, disrupting Grok model training, per The Information. The company is now installing backup power and cooling systems and running more thorough testing before facilities go live, reversing its previous build-fast-fix-later approach; a mismatch between older Hopper-generation and newer Blackwell GPUs at Colossus 1, plus latency between facilities more than 10 miles apart, had also slowed cluster performance. Elon Musk reassigned Starlink executive Michael Nicolls to lead AI infrastructure after several data-center executives departed, backfilling with staff pulled from the rocket and Starlink programs. The slowdown comes as SpaceX races a September 30 deadline under its deal to supply Google with roughly 110,000 Nvidia GPUs for $920 million/month — after a grace period, Google can terminate, cut the GPU allocation, or reduce payments if SpaceX misses it. SpaceX is also building an in-house turbine foundry in Bastrop, Texas to speed power deployment. SPCX traded up slightly (~$149.87, +1.14%) on the news; Goldman Sachs projects SpaceX's AI revenue could reach $322 billion by 2030, up from $3.2 billion in 2025.