SpaceX announced August 4 it will build all future AI infrastructure exclusively on Nvidia's Vera Rubin architecture, with CEO Elon Musk calling it "the best AI computer." The commitment anchors SpaceX's aggressive compute expansion: 2 gigawatts of deployed capacity by end of 2026, scaling to 10 GW (with 15 GW as realistic outcome) by end of 2027. Q2 capex hit $18.37 billion, with $15.83 billion dedicated to AI infrastructure—more than double quarterly revenue and setting the pace for ~$100B annualized revenue by year-end 2026. Analysts estimate this 10 GW deployment could generate ~$100B incremental Nvidia revenue.
SpaceX is simultaneously deploying ground-based AI data centers and launching the Starmind AI satellite constellation, with each satellite carrying Nvidia Vera Rubin GPUs and Vera CPUs for direct orbital compute. Starmind AI1 is expected to launch in 2027. SpaceX plans to apply for FCC approval to deploy up to 1 million satellites. The company has already secured $6.7 billion in forward cloud services contracts (through October 2026) with customers including Google and Anthropic. For terrestrial deployment, the cable-less rack design of Vera Rubin NVL72 improves serviceability and reduces installation errors.
For infrastructure investors and AI ops teams, SpaceX's exclusive commitment signals Nvidia's competitive lock on hyperscaler custom-silicon customers. AMD (Helios) and Intel remain shut out. The scale and timeline carry execution risk: SpaceX's capex run rate implies $28B H1 2026 spend (up from $7B in H1 2025), raising questions about cash burn and power/cooling availability. For Nvidia, this single customer commitment represents one of the largest discrete GPU allocation guarantees in the company's history. Satellite thermal and reliability engineering in orbit remain unproven at this scale.