Skip to content
NewsFunding & M&A

Anthropic in Preliminary Talks to Lease $10 Billion in Compute Capacity from Meta

· by Pondero Newsdesk

The short version

Two of the biggest frontier AI rivals may become each other's infrastructure vendors: Anthropic has approached Meta about a $10 billion compute arrangement, the New York Times reported on July 17 and CNBC confirmed.

Anthropic in Preliminary Talks to Lease $10 Billion in Compute Capacity from Meta

Two of the largest frontier AI companies may be about to become each other's infrastructure vendors. Anthropic has approached Meta about leasing roughly $10 billion in compute capacity, the New York Times first reported on July 17 and CNBC confirmed the same day, citing sources familiar with the discussions.

What the talks involve

The conversations are in very early stages and neither company issued a statement, per The Next Web. Per CNBC, the arrangement under discussion calls for monthly payments over roughly two years, with an early-exit option. Anthropic initiated the approach in June 2026, after Meta publicly signaled it would sell excess capacity to third parties.

The deal would follow Anthropic's existing SpaceX arrangement: $1.25 billion per month for Nvidia GPU access inside the Colossus 1 data center. Google has a parallel SpaceX deal running at $920 million per month. Despite approaching a $1 trillion valuation, Anthropic still places usage limits on its most advanced Claude models because chip supply is a hard constraint per TNW.

Meta's case for doing it

Meta's capital expenditure for 2026 could reach $145 billion per TNW, and Chief Executive Mark Zuckerberg has been signaling a cloud push since at least October 2025, when he noted that companies "regularly ask if we have compute they could buy from us at some premium." In May 2026 he said Meta was actively considering entering cloud.

To staff that effort, Meta brought in Dave Brown, a former senior AWS executive. The company cut 8,000 jobs in May 2026 while redirecting capital into AI infrastructure. Selling compute to Anthropic would help Meta generate revenue from that spend before a full cloud marketplace is ready.

The rival-as-vendor dynamic

Meta builds its own Llama series, which competes directly with Anthropic's Claude. A large compute lease would make Meta both a rival and a critical infrastructure provider. That structure is becoming routine in AI: SpaceX sells GPU access to both Anthropic and Google; Google sells Gemini capacity to Meta while rationing its own supply per TNW. Infrastructure dependency no longer tracks competitive lines in the current chip environment.

Why it matters for teams building on Claude

Anthropic's rate limits and access restrictions on frontier models trace directly to GPU supply. A second large compute source alongside SpaceX could ease that pressure over time. For teams with production dependencies on Claude's API, the relevant implication is that access ceilings on the most capable models may loosen if Anthropic closes this deal and secures more diversified supply.

The counterparty picture is also worth watching. Anthropic's capacity roadmap is already tied to SpaceX, a company with its own commercial AI interests. Adding Meta, which ships competing models, deepens that entanglement. Teams making long-term architectural bets on Claude should factor in that Anthropic's supply chain now runs through parties with distinct competitive interests.

What to watch next

Neither company has confirmed the discussions or given a timeline. A formal announcement would be the first known large compute arrangement between two direct frontier-model competitors. Whether Anthropic's reported IPO preparation accelerates or delays a signing is the clearest near-term signal to track.

Sources