SpaceX Taps Nvidia to Build Starmind AI1 Satellite Compute Payload
SpaceX is teaming up with Nvidia to design the Starmind AI1 satellite compute payload, bringing data center-class computing power into orbit. Each Starmind satellite will run on Nvidia’s newest chips, the Rubin GPUs and Vera CPUs, which the company says can deliver up to 25 times the AI processing power of an older H100 GPU.
To handle that much hardware, SpaceX has boosted the satellites’ peak power capacity by 67%, pushing it to around 250 kilowatts. That’s enough energy to support a full Rubin NVL72 rack, a setup packing 72 Rubin GPUs into a single satellite.
The Idea Behind Starmind
The idea behind Starmind is simple but ambitious: instead of building AI data centers on crowded, power-hungry land, SpaceX wants to put them in space, where sunlight offers a steady, abundant power source and the vacuum helps with cooling. It’s a different mission from Starlink, which moves internet data. Starmind is meant to actually process it, running AI workloads directly in orbit before passing results back down to Earth.
The timing is notable. SpaceX announced the Starmind AI1 satellite compute payload deal just hours before reporting its first quarterly earnings as a public company. Both Nvidia and SpaceX shares climbed on the news. Still, big questions remain. SpaceX’s AI division reportedly lost $6.4 billion against $3.2 billion in revenue last year, and the Federal Communications Commission has not yet approved the license needed to launch the satellites. For now, Starmind is a bold chip order still waiting on regulatory green light.
