Meta and Anthropic in Talks for $10 Billion Compute Lease Deal
View original source →The New York Times and CNBC reported on July 17, 2026, that Meta and Anthropic are in advanced talks for a two-year compute lease deal worth approximately $10 billion — with Meta providing excess data center capacity to Anthropic as part of a new 'Meta Compute' infrastructure business line.
Key Points:
• The deal would have Meta leasing unused capacity from its massive AI data center buildout to Anthropic, structured as monthly payments with an early termination option for Anthropic.
• Meta has invested tens of billions in data centers to train its Llama open-source model family and now has surplus capacity it is looking to monetize through the Meta Compute initiative.
• Both companies confirmed discussions are at an early stage. No final agreement has been announced as of July 17.
• Anthropic, unlike Microsoft, Google, or Amazon, does not own its own data center infrastructure — it relies on cloud providers and compute partners for the hardware to train and run its Claude models.
• The arrangement would make Meta simultaneously a competitor to Anthropic in the AI model market (through its Llama and Meta AI products) and a critical infrastructure supplier — an unusual but not unprecedented commercial relationship in technology.
The $10 billion scale — from a company that also competes with Anthropic — is the clearest sign yet of how severe the compute scarcity problem is for AI labs without owned infrastructure. Anthropic is essentially negotiating to pay a competitor for the ability to keep competing.
The Meta Compute business line is a strategic move with long-term implications: if Meta can lock in AI labs as infrastructure customers, it gains revenue, compute utilization efficiency, and strategic insight into what capabilities its competitors are actively building.
For enterprises using Claude, the deal raises legitimate questions about Anthropic's infrastructure independence. A core supply relationship with a model market competitor creates an unusual conflict of interest that enterprise risk teams should understand and factor into continuity planning.
Why It Matters: This deal reveals the structural reality of AI infrastructure: even well-funded labs cannot self-supply the compute needed for frontier models. The ability to access infrastructure is becoming as important as the ability to build models.