SpaceX and Nvidia are preparing to take AI infrastructure beyond Earth, with a space-optimized version of Nvidia's Vera Rubin NVL72 system planned for the Starmind AI1 satellite. Elon Musk said the companies aim to launch the system into orbit in the fourth quarter of 2027, with what he has described as "gigawatt-class" scaling to follow in 2028.

Announced by Musk alongside Nvidia Chief Executive Jensen Huang at Nvidia's GTC conference in Washington, D.C., the initiative represents a major step in SpaceX's effort to establish computing infrastructure in orbit. Nvidia has confirmed that Starmind AI1 will run on its Vera Rubin architecture, while SpaceX says the satellite will pair onboard computing with high-bandwidth laser links to the Starlink network, a constellation that already relies on optical cross-links between spacecraft. Musk has cited connectivity for the orbital system on the order of 200 gigabits per second.

Orbital AI Moves Toward Deployment

The Vera Rubin NVL72 is normally a rack-scale AI system built around 72 Rubin GPUs and 36 Vera CPUs, and it is the planned successor to the GB200 NVL72 racks built on Nvidia's current Blackwell generation. Nvidia has said the first rack-scale Rubin systems are scheduled to arrive in the second half of 2026, which would put the orbital variant roughly a year behind the platform's terrestrial debut. SpaceX and Nvidia are now adapting that architecture to withstand the demanding conditions of space.

According to SpaceX, Starmind satellites will rely on solar power and radiative cooling while processing AI workloads in orbit. The company also intends to build the satellites at its Gigasat Factory in Texas, where production could eventually reach thousands of spacecraft.

Rather than continuously transmitting large volumes of raw data to Earth, the system is designed to send processed results back through laser links. That approach could reduce bandwidth requirements and enable faster AI services. Orbital computing efforts to date have run at a far smaller scale, from HPE's Spaceborne Computer-2 experiment aboard the International Space Station to a demonstration satellite from startup Starcloud carrying a single Nvidia H100 GPU; nothing close to a 72-GPU rack has flown in orbit before.

Scaling AI Beyond Earth

The commercial rationale rests on several advantages. Space-based systems can draw on continuous solar energy in sun-synchronous orbit and sidestep many terrestrial constraints tied to land, electrical grids, and cooling infrastructure, constraints that have become a growing bottleneck for AI data center construction on the ground.

Orbital deployment nonetheless poses major challenges. Launch costs, radiation exposure, hardware reliability, thermal management, and satellite manufacturing must all be able to support large-scale operations.

Nvidia has positioned Vera Rubin as a major platform for high-performance AI inference and training. SpaceX's adoption could open a new market for Nvidia technology while offering SpaceX another route to expand computing capacity for AI applications.

The companies are targeting 2027 for the first launches, with larger-scale deployment expected in 2028. That schedule is tied to the ramp of Nvidia's Rubin platform on the ground, which the company has slated for the second half of 2026, and to SpaceX's parallel effort to stand up large-scale satellite production in Texas.