SpaceX has reshuffled leadership of its AI data-center operations after engineering concerns and reliability problems emerged at facilities in Tennessee and Mississippi. Jake Palmer, who led physical infrastructure for SpaceXAI, left in late July with several other data-center executives, and SpaceX veterans from its rocket and Starlink businesses have taken larger roles, including previously reported moves involving Wesley Salandro, Logan McConnell and Starlink executive Michael Nicolls. Some sites operated for months without backup cooling and power; the Macrohard facility relied on more than 100 mobile chillers and logged uptime well below an internal 99.9% target, with temporary systems contributing to outages that interrupted AI model training. Mississippi regulators allowed temporary gas turbines to run longer than planned after supply-chain delays held up 41 permanent units. Weeks before the latest shake-up, SpaceX added an AI infrastructure risk factor to its August 4 quarterly filing, citing construction delays, workforce turnover, power constraints and equipment shortages. The company ended June with 1.4 gigawatts of compute capacity, up from 0.4 gigawatts a year earlier, spent about $15.8 billion on AI infrastructure in the second quarter, and is racing toward a September 30 deadline to deliver committed GPUs under Google’s $920 million-a-month agreement, after which a one-month grace period applies before Google may terminate or cut payments proportionately. There is no indication the problems will affect that deal, though they test Elon Musk’s claim that SpaceX can bring compute online faster and better than peers.