Meta, Anthropic in talks to share AI computing capacity

On July 17, the New York Times reported that Meta and Anthropic are in early talks about Meta selling computing capacity to Anthropic at a premium. The reporting confirmed remarks Mark Zuckerberg had made the previous October, when he noted that companies had been approaching Meta with requests to purchase GPU access from the company. At that time, the comments seemed exploratory; the July reporting shows that at least one major AI company has moved the concept from informal inquiry into active negotiation. No agreement has been finalized, and both companies declined to comment on the details of the talks. The arrangement would give Anthropic direct access to Meta's computing infrastructure, addressing the intense competition for GPU resources that has constrained model development across the industry. For Meta, the arrangement would convert idle or excess capacity into revenue—a model well-established in cloud computing but relatively new in the AI infrastructure space. The talks underscore how rapidly the economics of AI development are shifting, with companies that have invested heavily in compute beginning to treat it as a tradeable resource.

The computing capacity market has become a critical chokepoint in AI development. Zuckerberg's October comment—that companies were asking Meta to sell GPU access at a premium—suggested Meta had recognized an opportunity to monetize its infrastructure investments. That observation has now materialized into formal discussions with Anthropic, one of the most significant AI research companies. The model mirrors how major cloud providers (Amazon Web Services, Google Cloud, Microsoft Azure) already operate: those with excess capacity can lease it to customers facing demand constraints. For Meta, this would create a new revenue stream; for Anthropic, it could ease supply constraints that otherwise limit model training and deployment speed. The talks remain preliminary, but they illustrate how the industry's largest players are beginning to function as infrastructure providers for one another.