Meta is exploring the possibility of leasing computing power from its AI data centers to Anthropic. According to sources cited by The New York Times, the two companies are discussing a deal that could be worth up to $10 billion over a two-year period. The talks illustrate just how scarce high-quality AI infrastructure has become.
The negotiations are still in the early stages, and a final agreement is by no means certain. According to insiders, Anthropic made the proposal in June. If the deal goes through, the AI developer will pay Meta a monthly fee for two years for access to GPU capacity. Both parties would have the option to terminate the agreement early.
New revenue stream for Meta
For Meta, an agreement would mean more than just a major client. The company is expected to invest up to $145 billion this year, largely in AI infrastructure and data centers. Investors have long wondered whether those investments will pay for themselves quickly enough.
By temporarily leasing excess capacity to third parties, Meta can generate a return on part of those investments sooner. CEO Mark Zuckerberg hinted during a recent earnings call that companies regularly inquire about available computing power. Until now, Meta has turned down such requests because it expects to need the capacity itself. However, if it turns out that the company has built more infrastructure than it is currently using, commercial leasing is a possibility, according to Zuckerberg.
The potential partnership is notable because Meta and Anthropic are simultaneously competing in the market for large language models. Nevertheless, the limited availability of AI infrastructure is increasingly forcing competitors to collaborate.
Anthropic has seen demand for its services rise sharply in recent months, partly due to the popularity of Claude Code. To meet that growing demand, the company is seeking additional capacity beyond traditional cloud providers. According to SiliconANGLE, additional computing power could also enable the company to further expand the capacity of Claude Code and its API. Following its earlier infrastructure agreement with SpaceX, Anthropic also increased the usage limits for both services.
Anthropic had already signed a similar agreement with SpaceX earlier this year. Under that deal, the AI developer is leasing computing power worth $45 billion over three years. The potential Meta agreement, at a maximum of $10 billion, would be significantly smaller but would still rank among the largest infrastructure deals in the AI sector. Similar to the proposed Meta deal, the SpaceX agreement includes a provision allowing both parties to terminate the contract early. According to SiliconANGLE, the agreement is formally structured as a lease with a term of 180 days and a 90-day notice period.
Meta also leases capacity itself
Interestingly, Meta is not only a provider but also a consumer of external AI capacity. While the company is building new data centers worldwide, it also entered into multi-year agreements with CoreWeave and Nebius this spring to secure additional GPU capacity.
This underscores just how tight the market for AI infrastructure currently is. Due to the explosive demand for generative AI, prices for computing power are rising rapidly, and developers are seeking out every available source of capacity.
If the agreement with Anthropic goes through, Meta will take a new step in its AI strategy. The company would then no longer position itself exclusively as a developer of AI models, but also as a provider of large-scale AI infrastructure to other parties. In the long term, this could develop into an additional source of revenue alongside advertising, cloud, and AI services.