SpaceX to use NVIDIA GPUs for its Starmind project — Here’s what they’ll face in space
SpaceX has revealed that its ambitious Starmind orbital data centre project will initially rely on NVIDIA's most advanced chips, as it races to meet an aggressive 2027 launch target for its first satellite, AI1. The project, championed by Elon Musk, aims to move AI computing into space to sidestep the land, power and cooling constraints facing terrestrial data centres, and could eventually see up to a million satellites deployed in low Earth orbit. While Musk is confident, experts remain sceptical that deploying cutting-edge chip architecture in the harsh conditions of space is technically or economically feasible on this timeline.
Each AI1 satellite is expected to carry around 72 NVIDIA chips, using the firm's Vera Rubin GPU and CPU platform, which NVIDIA claims offers up to 25 times the compute power of its older H100 model. The satellites will generate roughly 175kW of computing power on average, drawing on solar arrays producing 210kW while orbiting in near-constant sunlight, and will shed heat via large 1,700-square-foot liquid radiators venting into space. Longer term, SpaceX intends to shift production to its own Terafab chip plant, a $119 billion venture with Tesla, though for now it is leaning on NVIDIA to hit its schedule.
- SpaceX plans Starmind orbital AI data centre using NVIDIA's Vera Rubin chips
- First AI1 satellite launch targeted for 2027, despite expert scepticism
- Aims to bypass Earth's power, land and cooling limits for AI computing