Meta in talks to lease $10B in AI compute to Anthropic over two years

Meta is reportedly in talks with Anthropic to lease computing power in a deal worth as much as $10 billion（約1.6兆円） over two years, according to the New York Times. Anthropic proposed the arrangement in June, and Meta is currently considering it, with three people with knowledge of the talks confirming the details to the newspaper. Under the proposal, Anthropic would pay Meta in monthly installments over the two-year period, with an early-exit clause available to either party. Computing power, or compute, refers to the data center capacity used to train and run artificial intelligence models; the deal would let Anthropic rent Meta's excess infrastructure rather than build its own facilities. By industry standards, the scale appears modest. The proposed deal runs about a third of the agreement Anthropic signed with Elon Musk's SpaceX in May, under which the AI firm pays roughly $1.25 billion（約2000億円） monthly, or $45 billion（約7.2兆円） over three years, for computing power. Similar early-exit provisions apply to that larger contract as well. Both Anthropic and Meta declined to comment on the reported negotiations. The talks remain in early stages and may still collapse before closing. The urgency behind the deal reflects the broader AI industry's racing demand for compute. Leading AI companies are competing intensely to secure computing capacity, while Meta, Google, and Microsoft pour hundreds of billions into new data centers worldwide. That construction boom has unsettled Wall Street, with investors increasingly questioning whether such extraordinary levels of spending can ever be justified by real returns. Theo Jaffee of MTS noted that "Anthropic needs a lot of compute, and Meta has a lot of compute. Anthropic has really good models. Meta, until very recently, didn't have very good models, and now they have, you know, I would say an A-minus to B-tier frontier model." For Meta, a potential deal would carry unusual weight. It could create fresh revenue and ease pressure from shareholders skeptical of the company's aggressive infrastructure budget. Mark Zuckerberg has said Meta will spend as much as $145 billion（約23兆円） this year, most of it on AI—a figure more than double the $72 billion（約12兆円） spent the previous year. Meta has also admitted it might build more data centers than its AI products currently require. Selling that surplus offers an obvious solution to this overbuilding risk. On a May investor call, Zuckerberg hinted that outside firms regularly ask to buy compute at a premium. He said Meta had resisted so far because it still expected to use the capacity internally. Overbuilding, however, would make leasing the surplus a far more logical option. Anthropic, valued near $1.2 trillion（約190兆円） and preparing to go public, has seen demand surge since launching Claude Code. The growing scarcity of compute has pushed direct rivals toward cooperation. Meta itself already rents capacity elsewhere, including a $21 billion（約3.4兆円） CoreWeave deal and a $27 billion（約4.3兆円） agreement with Nebius. Rising compute prices now let the company consider renting its own centers out to others.

The reported talks reflect two urgent pressures in the AI industry: Anthropic's desperate hunt for computing capacity and Meta's need to justify an extraordinary infrastructure spend that has unsettled investors. Mark Zuckerberg announced Meta will spend as much as $145 billion（約23兆円） on AI this year—more than double the $72 billion（約12兆円） spent the previous year—yet the company has admitted it may have built more data centers than its AI products currently require. This overbuilding creates a potential liability; leasing surplus capacity to Anthropic would convert that excess into revenue and signal to shareholders that the massive infrastructure bet can generate returns beyond internal use. Anthropic faces the opposite problem: the startup needs vast computing power to train and run its models but lacks the capital to build data centers at the scale required. The proposed arrangement lets Anthropic rent from Meta's infrastructure rather than construct its own, easing both companies' constraints. The talks also reflect broader industry dynamics—leading AI companies including Meta, Google, and Microsoft are pouring hundreds of billions into data center construction worldwide, while investors grow skeptical of whether such spending can be justified by real returns. Anthropic itself has seen demand surge since launching Claude Code and is valued near $1.2 trillion（約190兆円） while preparing to go public, making its compute needs all the more pressing.