Anthropic weighs $10 billion Meta compute lease for Claude demand

Anthropic is in early talks to lease AI computing capacity from Meta in a deal that could reach $10 billion over two years, Reuters reported, citing a person familiar with the matter. If signed, the agreement would make Meta a supplier to a direct foundation-model rival and underline how scarce large-scale compute has become for companies trying to keep frontier AI products online.

The proposal reportedly came from Anthropic in June. Under the structure described to Reuters, Anthropic would pay Meta monthly across the two-year term. The terms are not final, either side could exit early, and the discussions may not produce a contract.

That caveat matters. Meta does not currently operate a commercial cloud business built around selling its own compute capacity, according to the report, which makes the transaction less straightforward than buying from AWS, Microsoft Azure, Google Cloud or a specialist GPU cloud provider. Neither company has announced the deal, and no details were disclosed on the amount of capacity involved, expected margins for Meta or how the compute would be allocated across Anthropic products.

Compute demand is outpacing internal supply

Anthropic’s need for outside infrastructure has grown with usage of Claude and Claude Code. The company is already using third-party capacity, including a previously reported agreement with Elon Musk’s SpaceX valued at $45 billion over three years, or about $1.25 billion per month, for compute at the Colossus 1 and Colossus 2 data centers in Memphis, Tennessee. A Meta contract at the reported ceiling would be materially smaller, roughly one-third the size of that SpaceX arrangement.

The prospective Meta lease also shows how the AI supply chain is making competitors dependent on one another. Meta develops Llama, while Anthropic develops Claude. A commercial compute agreement would put the two companies on opposite sides of a vendor relationship even as their models compete for enterprise and developer usage.

Meta tests demand for its AI infrastructure

Meta has signaled interest in selling capacity before. At its May shareholder meeting, chief executive Mark Zuckerberg said entering cloud computing was “definitely on the table” and said companies were approaching Meta “almost every week” about access to its AI models or unused compute.

The financial backdrop is large. Meta expects capital expenditure of up to $145 billion in 2026, more than double the $72 billion it spent the prior year, with AI infrastructure accounting for much of the increase. Leasing capacity to outside customers could help offset some of that spending and create revenue outside Meta’s advertising business, though the company has not said it will build a full cloud unit.

If Meta moves ahead, it would be stepping closer to companies such as CoreWeave and Nebius, which sell AI-focused infrastructure to model developers and enterprises. It would also be taking on the operational burden of serving external customers, a different business from building data centers for internal workloads.

The reported $10 billion figure should be treated as a ceiling while talks remain early. For Anthropic, the negotiation is another sign that model performance and product adoption are constrained by access to chips, power and data center capacity. For Meta, it is a test of whether the infrastructure built for its own AI ambitions can also become a business line.

This story draws on original reporting from TechFundingNews.