Lead

Meta is in discussions to lease AI computing capacity from its data centers to Anthropic in a deal that could be worth up to $10 billion over two years, according to a report from The New York Times on July 17. If finalized, the arrangement would turn a direct competitor in frontier AI development into a customer for Meta's infrastructure.

What's New

The New York Times reports, citing people familiar with the conversations, that Anthropic proposed the deal in June 2026 and Meta is currently evaluating it. Under the proposed terms, Anthropic would pay Meta in monthly installments over a two-year period, with both companies retaining the ability to opt out early. Anthropic declined to comment, and Meta did not respond to a request for comment from Yahoo Tech.

The potential agreement follows a pattern of unusual cross-industry compute leasing deals driven by the acute shortage of AI training and inference hardware. In May 2026, Anthropic struck a similar agreement with SpaceX to rent data center capacity for approximately $1.25 billion per month through May 2029. In June, Google signed its own deal with SpaceX for $920 million per month over 32 months.

Why It Matters

The proposed deal highlights the extreme demand for AI computing infrastructure, which has outpaced supply even for the largest technology companies. Anthropic recently released Claude Fable 5, its latest frontier model, and requires massive compute resources for training and serving. Meta, despite developing its own Llama series of models, has acknowledged it is struggling to keep pace with frontier capabilities from OpenAI and Anthropic.

Leasing compute to a competitor would create a new revenue stream for Meta while giving Anthropic access to data center capacity that remains scarce despite aggressive buildouts by cloud providers. It also reflects a broader trend where AI companies with excess capacity — including those outside the traditional cloud market — are monetizing their infrastructure through bilateral agreements rather than public cloud services.

The deal is not yet finalized and either party could walk away. The reported opt-out clause suggests both companies are approaching the arrangement with flexibility as the AI infrastructure market continues to evolve rapidly.

Our Take

The most interesting signal here is not the dollar figure but the structure. Anthropic approaching Meta directly — rather than expanding its relationship with existing cloud partners — suggests the compute shortage is severe enough that even frontier labs are shopping across rival camps. SpaceX entering this market as a data center provider was already unusual; Meta doing the same turns it into a pattern.

For Meta, the deal offers a rare financial upside from its infrastructure spending while its own AI models play catch-up. For Anthropic, it is a pragmatic hedge in a market where waiting for new GPU clusters means falling behind. Neither company is likely to discuss the deal publicly until terms are settled, but the fact that talks are happening at all tells us the AI compute market has entered a new phase — one where competitors become tenants of each other's hardware.

FAQ

How much is the potential deal worth?

The New York Times reports the deal could be worth up to $10 billion over a two-year period, paid in monthly installments.

Why would Meta lease compute to a competitor?

Meta's own AI models have not matched the frontier capabilities of OpenAI and Anthropic. Leasing spare data center capacity creates a new revenue stream from infrastructure that would otherwise be underutilized while Meta continues developing its Llama model family.

Has Anthropic done this before?

Yes. In May 2026, Anthropic signed a similar agreement with SpaceX to rent data center capacity for roughly $1.25 billion per month through May 2029. Google also signed a SpaceX deal in June 2026 for $920 million per month over 32 months.

Is the deal finalized?

No. The talks are ongoing, and both companies reportedly have the ability to opt out early. The information comes from unnamed sources cited by the New York Times and has not been confirmed by either company.

Why are AI companies struggling to find compute?

Global demand for AI training and inference hardware — particularly NVIDIA GPUs and specialized AI accelerators — has consistently outpaced supply. Even major technology companies face delays in provisioning new clusters, leading to creative arrangements like cross-competitor and cross-industry leasing deals.

Sources