SpaceX Acquires 440,000 Nvidia GPUs to Expand AI Infrastructure
SpaceX is expanding its AI compute capacity with 440,000 Nvidia GPUs to rent to other businesses and fund future orbital data center satellites.
Space Exploration Technologies Corp. is expanding its AI compute capacity by acquiring approximately 440,000 Nvidia GB300 Blackwell GPUs. The company expects 220,000 of these units to be operational within days, with the remainder active by October. This addition supplements an existing fleet of 780,000 Blackwell and Hopper processors.
To manage this growth, SpaceX appointed former Starship executive Kevin Holland as VP of Supercompute. Holland is tasked with activating the next six phases of compute by the end of 2026. This leadership shift follows the February merger of xAI into SpaceX, which created an entity referred to as SpaceXAI in July. CEO Elon Musk is currently overseeing the construction of a third data center, Minihard, in Memphis to bring the new chips online by late December.
SpaceX currently operates two Colossus supercomputing clusters in Tennessee and Mississippi. The company generates significant revenue by renting this capacity to other businesses, including Anthropic, which pays $1.25 billion monthly, and Google, which pays $920 million monthly. SpaceX intends to use these rental profits to fund the launch of orbital data center satellites starting as early as 2028.
