SpaceX and NVIDIA announced on August 4, 2026 that they are jointly developing the compute payload for the Starmind AI1 — a satellite designed to run artificial intelligence workloads from low Earth orbit using NVIDIA's most advanced silicon. The announcement came during SpaceX's inaugural quarterly earnings call, the company's first public financial report since listing.
Key Highlights
- Each Starmind AI1 satellite delivers up to 25 times the AI compute power of an NVIDIA H100 GPU
- A planned megaconstellation of up to 1 million satellites would form the world's largest distributed AI supercomputer
- SpaceX selected NVIDIA as exclusive chip supplier; Musk said they chose NVIDIA because "they are the best"
- NVIDIA shares gained roughly 3% and SpaceX stock climbed 9% on announcement day
The Starmind AI1 Satellite
At the heart of the Starmind AI1 payload is NVIDIA's Vera Rubin NVL72 system — combining Rubin GPUs and Vera CPUs. Each satellite sustains roughly 120 kW of average compute power, with peak capacity reaching 150 kW.
Power comes from a 210 kW solar array operating at 250 watts per square metre, with a solar wingspan stretching approximately 70 metres — about two-thirds the length of a Boeing 747. The satellites orbit between 311 and 1,243 miles (500 to 2,000 km) above Earth in sun-synchronous trajectories that keep the solar panels illuminated.
Unlike terrestrial data centers, heat is radiated directly into the vacuum of space, eliminating the power-hungry water chillers that make traditional facilities energy-intensive and expensive to operate. Communication between satellites runs via petabit-scale optical links integrated into the Starlink network.
A Megaconstellation of AI Nodes
SpaceX and NVIDIA are targeting a megaconstellation of up to 1 million satellites. Each satellite functions as an independent AI compute node — together forming what would be the world's first distributed AI supercomputer operating from space. For comparison, the current orbital population sits at roughly 15,000 active satellites.
The satellites will be manufactured at the Gigasat Factory in Bastrop, Texas, and launched aboard SpaceX's Starship rocket.
Timeline
Prototype testing for the Starmind AI1 is scheduled for early 2027, with mass production expected later that year if development stays on schedule. The FCC has granted SpaceX's initial application to operate the constellation, with a final regulatory decision still pending.
What This Means for AI Infrastructure
Orbital compute addresses three persistent bottlenecks in terrestrial AI infrastructure: land availability, power grid capacity, and cooling costs. A space-based supercomputer running on solar power with vacuum cooling could reduce the marginal cost of AI inference at scale — particularly for latency-tolerant workloads where a brief orbital hop is acceptable.
For NVIDIA, SpaceX's exclusive commitment represents one of the largest sustained hardware contracts in the space industry and a powerful signal that next-generation AI compute is heading beyond the atmosphere. The race for AI compute has officially left the ground.
Source: Interesting Engineering