Analysis: since October, Anthropic has entered into agreements for at least 14.8 GW of compute capacity and may spend as much as $517B over the next decade
Anthropic in the last year has scrambled to line up cloud computing deals with SpaceX, Google and others to meet the skyrocketing demand …
Context & Ripple Effects
Anthropic’s compute build-out moved from cloud procurement to physical infrastructure commitments in 2026: reporting in May described a planned five-year Google cloud-and-chip outlay of about $200 billion, while the company also secured all capacity at SpaceX’s Colossus 1 for more than 300 MW. June reporting added that Anthropic had signed more than 12 initial direct data-center lease agreements, a first for the company.
The 14.8 GW total frames those transactions as a long-duration capacity portfolio rather than isolated cloud contracts. The reported $517 billion figure is an analysis-based estimate, not a disclosed Anthropic commitment, but it makes the financing and delivery of contracted compute central to Anthropic’s operating model.
First-order effects
- Anthropic gains contracted access to at least 14.8 GW across Google, SpaceX and other providers, shifting its near-term constraint from finding capacity toward bringing that capacity online and funding it.
- Google and SpaceX obtain a large committed buyer for cloud and data-center capacity; Anthropic’s earlier reported five-year Google cloud and chip plan already represented a major portion of Google’s disclosed revenue backlog.
Second-order effects
- Direct leases and large provider agreements make data-center owners, cloud operators and capacity brokers more dependent on a small number of AI buyers with long-term demand commitments.
- Anthropic’s suppliers must translate signed contracts into usable power and compute on schedule, concentrating execution risk in infrastructure delivery rather than model demand alone.
Third-order effects
- AI infrastructure is becoming a contractual capacity market, in which frontier-model developers secure multi-year power and compute claims before the underlying facilities are fully operational.
- If this procurement pattern persists, competitive position will depend increasingly on the ability to finance, structure and execute long-duration infrastructure contracts—not solely on access to rented cloud instances.
The trend: Frontier AI companies are treating compute as a long-dated infrastructure asset, using cloud deals, direct leases and capacity commitments to lock in supply ahead of demand.