Meta in talks to lease $10 billion in computing power to Anthropic

Meta Platforms is in early-stage talks to lease computing power to Anthropic in a deal that could be worth as much as $10 billion over two years, the New York Times reported, marking the social media giant's pivot toward monetizing its vast AI infrastructure.

"If we do have some excess capacity, we could do the kind of deals SpaceX cut with Anthropic and Google," John Blackledge, an analyst at TD Cowen, said. "We'll monetize half a gigawatt or one gigawatt, whatever the excess is."

The discussions remain preliminary, according to three people with knowledge of the matter. Meta shares narrowed their decline to 3% on the news. The company plans to spend as much as $145 billion on capital expenditures this year, nearly double the $72.2 billion it spent in 2025, with the bulk going toward AI data centers including a 5-gigawatt expansion of its Hyperion facility in Louisiana and a new 1-gigawatt site in Canada.

For Meta, the deal would mark a strategic shift from pure AI consumer to infrastructure provider, putting it in direct competition with Amazon Web Services, Microsoft Azure, and Google Cloud. For Anthropic, it secures critical computing capacity for model training in a market where grid power shortages and hardware bottlenecks constrain supply.

Zuckerberg has telegraphed this move for months. During Meta's first-quarter earnings call in May, he said external companies were reaching out "almost weekly" about purchasing computing power, and that if Meta judged its infrastructure to be over-invested, "external sales are a perfectly viable option." In a Bloomberg interview last week, he said leasing data center space makes sense given industrywide constraints on AI computing.

The strategy mirrors one pioneered by SpaceX, which has inked multi-billion-dollar deals to lease data center capacity to Anthropic and Google. SpaceX's arrangement with Anthropic runs through May 2029, though either party can exit with 90 days' notice.

Meta's Cloud Ambitions

Meta is exploring a broader initiative internally called "Meta Compute" that would sell graphics processing unit capacity through an Infrastructure-as-a-Service model, according to industry sources. The company is also developing its own AI chips — code-named Iris, designed with Broadcom and manufactured by Taiwan Semiconductor Manufacturing Co. — with production slated to begin in September. Meta plans to launch a new chip roughly every six months through 2027.

The pivot creates tension with the neocloud companies that have been Meta's core GPU suppliers. Meta signed a long-term infrastructure contract with Nebius that could expand to as much as $27 billion, and inked multi-billion-dollar deals with CoreWeave. Those same companies now face the prospect of their largest customer becoming a direct competitor.

"This market is not a zero-sum game," Blackledge said. "Especially right now, this is a moment of absolute capacity shortage." He noted that even Google faces capacity constraints after doubling its capital expenditures this year.

Still, the transition will take time. Meta plans to bring about 7 gigawatts of computing capacity online this year and double that to 14 gigawatts in 2027. Its own AI models and vast service operations will consume the majority of that capacity, meaning external leasing will initially target only surplus compute.

"What Meta is trying to do is get firsthand experience in this business and understand the margins and costs," Bob O'Donnell, founder and chief analyst at TECHnalysis Research, said. "This is completely different from any business Meta has done before."

Meta shares trade at about 24 times earnings, a discount to Microsoft at 30 times and Amazon at 35 times, reflecting investor skepticism about the return on its unprecedented capital spending. A $10 billion compute lease with Anthropic would represent roughly 7% of Meta's projected 2026 capital expenditures — a modest but meaningful step toward proving that its data center build-out can generate external revenue.

This article is for informational purposes only and does not constitute investment advice.