Meta in talks to lease Antropic’s computing power in a potential $10 billion deal

Meta is in talks to lease computing power from its artificial intelligence data centers to Anthropic in a deal that could be worth up to $10 billion over two years, three people familiar with the discussions said, a potential move toward a new artificial intelligence business for the social network company.

Anthropic proposed the deal in June and Meta is considering it, said the people, who were not authorized to discuss the confidential talks. While the specifics were being worked out, Anthropic would pay Meta in monthly increments over two years, the people said. The companies could back out of any deal early, they added.

Meta’s Anthropic proposal was about a third the size of the deal the AI ​​startup signed with Elon Musk’s SpaceX in May. Under the deal, Anthropic pays the rocket company $45 billion over three years — or $1.25 billion a month — for computing power. The agreement contained similar provisions that allowed both companies to exit the agreement early.

Meta’s talks with Anthropic are at an early stage and may not result in a deal, people with knowledge of the discussions said. Anthropic and Meta declined to comment.

The talks highlight how eager top AI companies are to get more computing power to rapidly develop the technology, as tech giants including Meta, Google and Microsoft invest hundreds of billions of dollars in building dozens of new data centers around the world. The construction boom, which has boosted spending by tech companies extraordinarily, has raised concerns on Wall Street about whether such sums can be justified.

For Meta, the deal would be particularly significant. It could open up a new line of business for the company and potentially ease pressure from investors who have questioned how much Meta is spending on data centers to develop cutting-edge AI models. Its CEO Mark Zuckerberg said his company will spend up to $145 billion this year, much of it on AI, which would be more than double the $72 billion it spent last year.

However, Meta has faced questions about whether its own AI models can compete with those developed by rivals such as Anthropic and OpenAI. And the company has acknowledged that it can build more data centers than it needs relative to the number of customers using its AI products. Selling excess computing power to companies like Anthropic could provide Meta with a new source of revenue until demand for its own AI services catches up.

In recent calls with investors, Mr. Zuckerberg hinted that selling computing power could be one way for Meta to get some return on its AI investment.

“Almost every week we have different outside companies come to us and ask us” about computing power “that they could buy from us at some price that we bought it for,” Mr. Zuckerberg said on an investor call in May. “We haven’t done that yet because we think we have a use for the calculation.

“But of course, if we get to the point where we feel like we’ve rebuilt, then that’s an option we have, and that’s partly what gives us the confidence to invest in building this,” he said.

AI companies have become more comfortable making deals with competitors out of necessity because computing power is limited. Privately valued at nearly $1 trillion and going public, Anthropic has seen a surge in demand since the release of its enterprise software product Claude Code. As a result, it has had to partner with companies that own large amounts of computing power to serve its growing customer base.

Since Meta doesn’t have a business to sell its computing power, that has complicated its talks with Anthropic, said one of the people briefed on the discussions.

Meta has also entered into its own agreements to lease computing power from other data center providers, even as it builds its own facilities. That includes a $21 billion deal with AI computing power provider CoreWeave in April and a $27 billion deal with another such company, Nebius, in March.

Since those deals were signed, the price of computing power has skyrocketed due to limited supply and increased demand, said Mandeep Singh, a technology analyst at Bloomberg Intelligence. That gave Meta the opportunity to lease its data centers while making a long-term investment in its own AI, he added.