SpaceX targets 10GW of AI compute capacity by 2027

SpaceX aims to expand its AI compute capacity to approximately 10 gigawatts by the end of 2027. According to company filings and earnings calls, the capacity stood at 1.4 gigawatts at the close of Q2 2026. The expansion will primarily utilize Nvidia’s Vera Rubin architecture, specifically the NVL72 and Kyber configurations.

The company plans to allocate roughly 60% of this capacity for internal use across SpaceX, xAI, the Grok chatbot, and Tesla, while the remaining 40% is intended for external customers, including potential engagements with Google and Anthropic. SpaceX is also developing Starmind, an orbital AI platform built on the same hardware, with initial satellite launches projected for 2027.

Elon Musk has suggested that this buildout could generate between $300 billion and $500 billion in annual recurring revenue. The company’s rapid scaling follows a successful deployment in Memphis, where it brought approximately 100,000 Nvidia H100 GPUs online in 122 days.