SpaceX plans to scale its AI compute capacity from approximately 1.4GW at the end of Q2 2026 to 10GW by the end of 2027, representing a more than sevenfold increase in roughly 18 months. The company is betting exclusively on Nvidia’s Vera Rubin architecture for this expansion. Roughly 60% of the capacity will be reserved for internal use covering xAI, the Grok chatbot, SpaceX, and Tesla, while 40% will serve external customers such as Google and Anthropic. Elon Musk projects annual recurring revenues between $300 billion and $500 billion, with capital expenditures estimated at $15.8 billion to $18.4 billion in Q2 2026 alone. SpaceX is also developing Starmind, an orbital AI platform built on the same Vera Rubin architecture as its terrestrial servers.
Source: Read the original article
