Nvidia’s rack-scale AI hardware is heading off-planet. SpaceXAI, the venture formed from SpaceX and xAI, will fly a Vera Rubin NVL72 system aboard the first spacecraft in its Starmind constellation, with a launch penciled in for next year.
The August 24 agreement also puts Vera CPUs at the center of the company’s Grok workloads, where they take on orchestration, code execution and data processing between model calls.
Vera Rubin NVL72 is the platform Nvidia is pushing as the standard for AI factories, and SpaceXAI becomes the first operator to run it outside a terrestrial data center.
Nvidia’s Ian Buck, vice president of hyperscale computing, said agentic workloads need machines that do more than produce answers, which is why the orbital system leans on Vera for the busywork around inference.
Both sides want a shared Nvidia software stack spanning Grok, earthbound data centers and orbit, so agents can move between them without rework.
Gwynne Shotwell, SpaceXAI’s president, said compute payloads will first ride some Starlink satellites, ahead of dedicated AI1 orbital data centers arriving in late 2027.
