Anthropic has signed roughly $517 billion in compute capacity agreements over the past 11 months, a figure that underscores how quickly frontier AI labs are converting capital into physical infrastructure. The total, compiled by The Information from public announcements and its own reporting, represents 14.8GW of compute capacity that Anthropic will access over the next few years. The scale matters because compute access is now a primary constraint on AI development, and multi-hundred-billion-dollar commitments signal that leading labs expect demand for training and inference capacity to keep rising well beyond current levels.
The agreements span both data center leases and cloud provider contracts. On the data center side, Anthropic reportedly signed a $45 billion compute capacity agreement with Nscale toward the end of August, a 191MW agreement with Riot Platforms for its Texas data center, and more than a dozen letters of intent for data center leases with multiple US developers. On the cloud side, the company has a $200 billion agreement with Google for TPUs that it could deploy itself, a large cluster of Amazon Web Services' custom Trainium hardware, and a $50 billion partnership with AI cloud firm Fluidstack. Other counterparties named in the reporting include Akamai, CoreWeave, SpaceX/xAI, Google Cloud, Lambda, Volta Infra, AMD, and Microsoft.
The evidence comes from a single secondary source, Data Center Dynamics, which summarized The Information's analysis. The dossier notes that Anthropic previously told investors it expected to spend around $180 billion on renting servers through 2029, but the newer tally of $517 billion suggests that estimate has risen significantly. The reporting also indicates that Anthropic may have signed additional deals that were not made public, meaning the true total could be higher. An unnamed source said the company had secured between 1GW and 2GW of capacity before the 11-month window covered by the analysis, further expanding the cumulative footprint.
For the data center and cloud sectors, the concentration of this spending is notable. Google and AWS will combined contribute 11GW of the 14.8GW total, reinforcing the position of hyperscale cloud providers as the dominant suppliers of AI compute. The remaining capacity is spread across specialized data center developers and AI cloud firms, suggesting a bifurcated market in which a few large platforms capture most of the value while smaller operators compete for incremental deals. Anthropic's confidential IPO filing with the US SEC in June adds another dimension: investors will likely scrutinize these compute commitments as both a growth enabler and a long-term cost obligation.
The main limitation is that the dossier relies on one secondary account of The Information's analysis, and Anthropic had not commented at the time of publication. The $517 billion figure is an aggregation of reported agreements rather than a confirmed balance sheet liability, and the timing of capacity delivery remains unclear. What to watch is whether Anthropic discloses more detail in its eventual IPO filings, and whether the pace of compute contracting across the industry continues at this magnitude or begins to moderate as capacity comes online.