Anthropic has committed to compute capacity contracts potentially valued at $517 billion over the past 11 months, nearly three times the $180 billion the company previously projected to investors for spending through 2029.
Since October, Anthropic added at least 14.8 gigawatts of potential compute capacity through these agreements, building on an earlier 1–2 gigawatts. The deals span cloud services, direct chip purchases and long-term data center leases—a multi-pronged approach reflecting the scale required for frontier AI development.
The surge follows unexpected demand for Claude Code and Cowork. For Anthropic, securing this infrastructure is essential to narrowing the compute gap with larger rivals like OpenAI and Google, where access to massive processing power has become the primary competitive moat. The company is positioning itself for public market scrutiny, where investors will scrutinize not just revenue but the capital efficiency of its infrastructure deployment.
The deals underscore a brutal economics reality in AI: the winner-take-most dynamics of model training mean companies must commit to years of capacity ahead of validated product-market fit. Anthropic is betting that demand for Claude justifies locking in this capital before its IPO pricing.
