SpaceX is partnering with NVIDIA to design compute payloads for its StarMind AI1 satellites, which will launch thousands of units equipped with Rubin GPUs and Vera CPUs starting in late 2027, according to an announcement by SpaceX on August 4, 2026.
Each satellite will carry a 150 kilowatt peak compute payload paired with solar arrays and transmit processed results through Starlink's laser network. The partnership pairs space-grade processor design with SpaceX's manufacturing capacity: the company plans to build and deploy the constellation from its Gigasat Factory in Bastrop, Texas, with production expected to begin in late 2027.
SpaceX says its orbital approach eliminates land, electrical grid connections, and active cooling systems by running inference and processing on-orbit, then beaming results to ground stations via Starlink's inter-satellite laser links. The 150 kW payload per satellite would allow significant workload capacity in a single unit.
NVIDIA's involvement centers on the processor design: Rubin GPUs handle AI inference tasks while Vera CPUs manage scheduling and networking overhead. Both are purpose-built for space environments, where radiation, thermal cycling, and power constraints differ sharply from terrestrial data centers. NVIDIA has previously built custom silicon for specialized environments, most notably the H100 variants for national laboratories and hyperscalers.
The Bastrop factory announcement reflects SpaceX's scale-up intent beyond Starlink. The company has operated limited satellite production capacity at its Starbase facility in Texas. A dedicated Gigasat Factory would support manufacturing thousands of units annually, a volume comparable to how terrestrial semiconductor fabs operate for mass production.
The late 2027 timeline puts first deployment roughly 18 months away from the August 2026 announcement. SpaceX would need to complete satellite integration, orbital qualification, and launch scheduling within that window. Starlink's existing constellation exceeds 6,000 active satellites, so adding AI-specific units with inter-satellite laser links fits within the company's demonstrated launch cadence.
SpaceX is pursuing a market where multiple actors are active. Amazon has disclosed plans to use AWS infrastructure for satellite data processing, and other aerospace companies have announced orbital edge-compute projects. SpaceX's advantage is vertical integration: it manufactures the satellites, operates the launch system, and runs the terrestrial gateway network through Starlink, eliminating dependencies on third-party carriers or ground operators. The cost per satellite and per-gigaflop of sustained compute in orbit will determine whether this approach proves economical at scale. SpaceX has not disclosed unit economics or pricing for customer compute access.
