SpaceX and Nvidia will put a Vera Rubin rack in orbit — the Starmind AI1 satellite

Announced during SpaceX's first earnings call as a public company, the two firms are jointly developing the compute payload for Starmind AI1 — a satellite designed to run data-center-class AI workloads in low Earth orbit. The payload pairs Nvidia's Rubin GPUs with Vera CPUs in the Vera Rubin NVL72 rack-scale configuration, the same silicon Nvidia sells for terrestrial AI factories. SpaceX said it will use Nvidia exclusively to power its next generation of AI infrastructure. Production is planned at SpaceX's Gigasat factory in Bastrop, Texas, with initial deployments targeted as early as 2027. The companies describe an eventual constellation of up to a million satellites acting as nodes in a distributed orbital supercomputer — an ambition, not a plan with a date.

Why it matters

Orbital compute has been a pitch deck idea for years; what changes here is who is making it and with what. SpaceX is the only operator that can launch and service rack-scale hardware at constellation cadence, and an exclusive commitment from it is a durable customer lock for Nvidia at exactly the moment hyperscalers are diversifying toward their own silicon. The stated rationale — power and grid limits on the ground — is also the honest one: the terrestrial constraint is now electricity and interconnects, not chips, and orbit trades those for launch mass, radiation and the unsolved problem of dumping tens of kilowatts of waste heat into vacuum. Treat 2027 as a first-article date, not a capacity date, and the million-satellite figure as direction rather than forecast.

What to watch

Whether Starmind AI1 gets a launch date, what power and thermal envelope the first payload actually flies with, and whether any customer workload — SpaceX's own or third-party — is named for it.