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 capacity strategy began with a 2025 Google cloud partnership that paired access to 1 million TPUs with 1 GW of 2026 capacity. By May, it had added the full capacity of SpaceX’s Colossus 1 and was reported to be pursuing roughly $200 billion of Google cloud and chip spending over five years.
The procurement model then widened beyond conventional cloud purchasing: sources reported more than a dozen initial direct data-center lease agreements. The reported 14.8 GW total makes those separate arrangements legible as a concentrated effort to secure infrastructure before demand is served.
First-order effects
- Anthropic gains a large, multi-provider compute reservation base, while Google, SpaceX and direct data-center lessors gain a major contracted customer.
- The reported $517 billion decade-long cost estimate puts Anthropic’s ability to convert model demand into revenue at the center of its infrastructure strategy; the figure remains an estimate, not a disclosed commitment.
Second-order effects
- Direct leases pull cloud providers and data-center financiers closer to Anthropic’s buildout: the reported possibility of a Google financial guarantee shows that capacity supply can require credit support alongside hardware and cloud service.
- Google and SpaceX are no longer simply vendors in Anthropic’s stack; their own capacity planning becomes more tied to one customer’s deployment schedule and utilization needs.
Third-order effects
- If comparable agreements proliferate, frontier-model development shifts toward a contractual capacity market in which access to power, sites and long-duration commitments matters alongside model capability.
- The more AI firms procure through direct leases and provider-backed commitments, the more compute execution risk moves from flexible cloud consumption into financing and construction obligations.
The trend: AI infrastructure is becoming a long-duration, contract-financed capacity market rather than a primarily on-demand cloud purchase.