SpaceX Overhauls AI Data Centers Ahead of Google's $920M Deadline

SpaceX has redesigned its AI data centers around power and cooling redundancy after reliability failures at sites in Tennessee and Mississippi, The Information reported, with the rework running against a Sept. 30 deadline tied to Google's compute contract.

Jake Palmer, who led physical infrastructure for SpaceXAI, left in late July alongside several other data-center executives; rocket-production leader Wesley Salandro and Starbase operations director Logan McConnell moved into larger roles. The Tennessee Macrohand facility "had to rely on over 100 mobile chillers and recorded uptime well below the company's 99.9% internal target," The Motley Fool wrote, citing the reporting. In Mississippi, the company kept temporary gas turbines running longer than planned after supply-chain problems delayed 41 permanent units.

The commercial stake is specific. Google agreed to pay $920 million a month at full capacity for access to around 110,000 Nvidia GPUs, with delivery due by Sept. 30. After a one-month grace period, Google may terminate the agreement, accept fewer GPUs, or reduce its payments.

Some facilities operated for months without backup cooling and power systems, and outages interrupted AI model training. SpaceXAI has scaled capacity from 0.4 gigawatts to 1.4 GW over the past year and still aims for 2 GW by year-end, though the redesign could slow that push. Nvidia signed eight Australian operators this week for a comparable 2GW buildout by 2027. AI revenue during the second quarter topped $2.5 billion. Power and cooling as the binding constraint on AI capacity has been a recurring theme across our AI infrastructure coverage this week.