SpaceX S-1 Filing Reveals $15 Billion/Year Anthropic Compute Deal
SpaceX's newly public S-1 filing has revealed the staggering scale of its compute partnership with Anthropic. The AI safety company agreed to pay $1.25 billion per month—$15 billion annually—through May 2029 for access to SpaceX's Colossus I and Colossus II AI training data centers in Memphis, Tennessee. The deal includes a 90-day exit clause for either party, likely reflecting the volatile nature of the AI industry and the fact that Anthropic's Claude competes with Elon Musk's Grok. The filing also disclosed that SpaceX spent $12.7 billion in capex on AI in 2025 (61% of total spend) and $7.7 billion in Q1 2026, compared to just $1 billion on its space division. SpaceX's AI division lost $6.3 billion on $3.2 billion in revenue in 2025. Meanwhile, Anthropic is reportedly approaching its first quarterly operating profit, with sales expected to reach at least $10.9 billion. Anthropic is also reportedly in talks to use Microsoft's Maia 200 AI chips via Azure, suggesting even the Colossus deal isn't enough compute capacity for Claude's growing demands.
The numbers are eye-watering but reveal the real economics of AI in 2026: compute is the bottleneck, and companies are willing to pay unprecedented sums for GPU access. SpaceX is essentially pivoting into an AI infrastructure company.
Why is Anthropic paying SpaceX instead of using cloud providers?
Compute capacity for training and running large AI models is extremely scarce. SpaceX's Colossus data centers represent some of the largest GPU clusters in the world, and traditional cloud providers can't scale fast enough to meet demand from frontier AI companies.