Meta Platforms (META) is in early-stage negotiations to lease computing power from its vast data-center network to AI startup Anthropic, a transaction that could be worth up to $10 billion over two years and would mark the social-media giant’s formal entry into the cloud infrastructure business.
The talks, first reported by The New York Times and independently confirmed by several news outlets, were initiated by Anthropic in June. Under the proposed structure, Anthropic would make monthly payments to Meta over a two-year term, with either party retaining the right to exit the agreement early. People familiar with the discussions cautioned that the negotiations remain preliminary and may not result in a final deal.
Meta shares initially fell as much as 6% on July 17 before paring losses to close roughly 2% lower, underperforming a broader tech selloff. Some reports noted a brief intraday pop on the news before the decline resumed.
The potential agreement signals a strategic pivot for Meta, which has poured tens of billions of dollars into AI hardware and data centers but has yet to commercialize that infrastructure beyond its own advertising and product needs. CEO Mark Zuckerberg has publicly floated the idea of a cloud-computing business for more than a year.
“Almost every week, there are different companies that come to us from the outside asking us about computing power that they could buy from us at some premium to what we’ve bought it at,” Zuckerberg told investors in May. “We haven’t done that yet because we think we have a use for the compute. But obviously if we get to a point where we feel that we have overbuilt, then that is an option that we have.”
Internally, the initiative is already being referred to as “Meta Compute,” according to the Times. The company recently hired Dave Brown, a former senior executive at Amazon Web Services (AWS), a move widely interpreted as a signal that its cloud ambitions extend far beyond a single leasing arrangement.
Meta is on track to spend as much as $145 billion in capital expenditures in 2026, more than double the $72 billion it spent last year. The bulk of that outlay is directed at AI infrastructure, including next-generation data centers and Nvidia (NVDA) chips. The scale of the spending has drawn persistent questions from Wall Street about whether the company can generate adequate returns, particularly since Meta’s own AI models, such as Llama, compete in a crowded field that includes offerings from Anthropic, OpenAI, and Google.
For Anthropic, a deal with Meta would add a third major pillar to its rapidly expanding compute portfolio. In May, the company signed a three-year, $45 billion agreement with Elon Musk’s SpaceX to access the full computing power of the Colossus 1 data center in Memphis, Tennessee—a facility housing more than 220,000 Nvidia processors. That contract, structured at roughly $1.25 billion per month, dwarfs the proposed Meta arrangement, which would be about one-third the size. Anthropic has also inked an $18 million deal with cloud provider Akamai.
The flurry of compute-leasing activity underscores a fundamental reality of the current AI landscape: access to advanced chips and the power to run them remains severely constrained. Anthropic, the creator of the Claude family of models, has imposed usage limits on its most advanced systems, including a model referred to as Fable, because of capacity shortages. Securing additional infrastructure from Meta would give the company more headroom to train models, run inference, and support a growing base of enterprise and consumer users.
“AI companies are increasingly accepting the idea of partnering with competitors as computing power becomes extremely scarce,” the Times reported, describing the dynamic as a kind of “frenemy” symbiosis reshaping the industry.
The proposed deal also arrives as Anthropic prepares for a potential initial public offering that could come as early as October, according to some reports. Locking in additional capacity ahead of a market debut would strengthen its operational story for public investors.
For Meta, the transaction represents more than just a revenue opportunity. It would position the company as a direct competitor to a new class of “neocloud” providers such as CoreWeave (CRWV) and Nebius (NBIS), which have built businesses around renting GPU-accelerated infrastructure to AI developers. Unlike traditional hyperscalers—Amazon (AMZN), Microsoft (MSFT), and Google (GOOGL)—Meta has no existing cloud division, meaning it would be building a sales, support, and operations function from scratch.
That inexperience may be one reason the talks are described as complicated. “Meta does not have a business selling its computing power,” a source told Reuters, which independently confirmed the discussions. The company declined to comment officially, while Anthropic also declined to comment.
The financial contours of the potential deal, while eye-catching, come with significant caveats. The $10 billion figure represents a theoretical maximum over the full two-year term. Because the proposed structure allows for monthly payments and early termination by either party, the actual revenue realized could be far lower. The flexible terms suggest both sides are hedging against the rapid pace of change in AI infrastructure—where capacity shortages could ease, or new chip generations could alter the economics, within months.
Still, the symbolism is powerful. A deal between Meta and Anthropic would pair two companies that compete directly in the market for large language models—Meta with its open-source Llama models and Anthropic with its proprietary Claude systems—in a commercial relationship where one supplies the physical infrastructure the other needs to operate. It is a configuration that would have seemed unlikely just two years ago, when the major tech platforms were largely building AI capabilities in-house and treating their data centers as proprietary fortresses.
The talks also illustrate how the AI infrastructure buildout is creating unexpected business models. Zuckerberg acknowledged as early as October 2025 that companies were regularly inquiring about buying compute capacity from Meta at a premium. “We hear from companies regularly that are asking if we have compute that they could sell to them at some premium to what we’ve bought it at,” he said at the time. What sounded hypothetical then is now the subject of a formal negotiation.
Wall Street’s reaction to the news was mixed. While the prospect of a new revenue stream could eventually ease concerns about Meta’s capital spending, the initial stock decline suggested investors were focused on the near-term uncertainty. A $10 billion lease would cover only a fraction of Meta’s annual AI-related capex, and building a cloud business from the ground up carries its own execution risks.
Nevertheless, the direction of travel is clear. With AI model developers desperate for compute and hyperscalers sitting on increasingly massive infrastructure, the lines between competitor and supplier are blurring. The Meta-Anthropic talks may be the most prominent example yet of a trend that industry observers expect to accelerate: the emergence of a liquid, multi-party market for AI computing power in which even direct rivals become commercial partners.