SpaceX to More Than Quintuple Its AI Compute Capacity with Nvidia by End of 2027

During SpaceX's Q2 earnings call, Elon Musk said the company aims to increase its AI compute capacity to over two gigawatts by the end of 2027, with the figure closer to ten gigawatts. Current expansion stands at 1.4 gigawatts. Musk stated that SpaceX will use Nvidia's upcoming Vera Rubin platform, which he called the best architecture. Existing Colossus clusters use Nvidia's H100, GB200, and GB300 systems. IPO documents revealed existing and planned Colossus expansions totaling approximately 100,000 H100, 110,000 GB200, and 110,000 GB300 processors, plus at least 220,000 planned GB300s, totaling about 540,000 GPUs. The planned capacity increase would mathematically require over one million Rubin GPUs, potentially over two million if fully implemented with Vera Rubin; SpaceX did not specify the actual Rubin share. Musk's AI company xAI merged into SpaceX in February 2026 in a mostly stock-funded deal, valuing SpaceX at $1 trillion and xAI at $250 billion, combined $1.25 trillion. Musk aims to operate data centers in orbit long-term. SpaceX's AI segment, including xAI and X, reported Q2 2026 revenue of $2.56 billion with an operating loss of $1.26 billion, compared to Q1's loss of $2.47 billion on $818 million revenue, totaling a $3.73 billion loss in H1. Revenue growth stems mainly from new cloud contracts renting Colossus capacity, not Grok, with one unnamed AI customer contributing about $1.52 billion, likely Anthropic. Competitors distribute capacity more broadly: Anthropic uses AWS Trainium, Nvidia GPUs, and Google TPUs for training and running Claude, with up to 5 GW additional AWS infrastructure (Trainium and Graviton) and about 3.5 GW additional TPU capacity announced. OpenAI has announced 6 GW AMD, at least 10 GW Nvidia, and about 2 GW AWS Trainium capacity.