Meta Data Center Deal Could Put Anthropic on Its Cloud

• The reported Meta data center deal could be worth as much as $10 billion over two years, according to The New York Times.
• A Meta data center deal with Anthropic would give Meta a potential cloud revenue stream beyond its advertising business.
• Anthropic needs enormous computing capacity for Claude, while Meta is building data centers at a scale few companies can match.
• The talks remain early, but they point to a future in which AI infrastructure becomes a business in its own right.

Meta’s expensive AI buildout may finally have a customer

The reported Meta data center deal with Anthropic matters because it could turn one of Silicon Valley’s most expensive corporate habits into an actual business. Meta has spent years building and expanding the infrastructure needed to train and run AI models. Now, according to The New York Times, it is discussing a plan to lease some of that capacity to Anthropic, the company behind Claude.

The conversations are said to be early, so this is far from signed-and-sealed territory. But the potential scale is hard to ignore: the arrangement could be worth up to $10 billion over two years. For a company that still gets the overwhelming majority of its revenue from ads shown across Facebook, Instagram and WhatsApp, that would be a strikingly direct entry into cloud computing.

Frankly, Meta has earned some skepticism whenever it announces another giant AI infrastructure bill. Investors can understand spending to improve ad targeting, recommendation feeds, creator tools and consumer AI assistants. Spending tens of billions on facilities that may not produce a clear return? That requires more faith. Leasing capacity to Anthropic offers a much cleaner answer: some of those racks of GPUs could generate revenue while Meta pursues its own AI ambitions.

Meta chief executive Mark Zuckerberg has not exactly hidden that this possibility was on the table. During an earnings call last year, he said Meta receives requests for its data center capacity “almost every week” and described selling access as “an option” for the future. A deal with Anthropic would suggest the option is moving from an investor-call hypothetical to a serious operating plan.

Why the Meta data center deal makes sense for Anthropic

Anthropic is one of the handful of AI companies for which compute is not a back-office expense; it is the central constraint on the entire business. Training frontier models is punishingly costly, and serving Claude users compounds the problem. Every long conversation, code generation request and enterprise workflow consumes processing time somewhere.

That makes the reported Meta data center deal less strange than it initially sounds. Meta and Anthropic compete in the broad race to build powerful language models: Meta has its Llama family, while Anthropic sells Claude to consumers and businesses. Yet rivalry at the model layer does not prevent cooperation at the infrastructure layer. Airlines compete for passengers while using the same airports. AI companies may increasingly compete for users while renting the same physical computing capacity.

Anthropic has already shown it is willing to make enormous infrastructure commitments. The company has longstanding relationships with Amazon and Google, both of which provide cloud resources and have invested in Anthropic. The source reporting also points to a separate compute arrangement with SpaceXAI valued at roughly $45 billion over three years, a reminder that AI firms are now signing capacity agreements with price tags once reserved for national infrastructure projects.

If that kind of spending holds, Anthropic’s priority is simple: secure enough chips, power and data center space to keep Claude available and improving. A Meta data center deal could give it another supplier at a moment when top-tier AI compute remains scarce and deeply contested.

Meta would be entering a business with very different rules

For Meta, renting infrastructure to third parties would mean taking a step toward the territory dominated by Amazon Web Services, Microsoft Azure and Google Cloud. Those businesses were not built overnight. They rely on global data center footprints, enterprise sales teams, support contracts, security certifications, billing systems and the rather unglamorous ability to handle a customer’s crisis at 3 a.m.

That is why my read is that this would not immediately make Meta a general-purpose cloud provider. Meta does not need to challenge AWS for every startup’s storage bucket or every retailer’s database. It could instead sell high-value AI capacity to a small group of customers with enormous needs. Think of it less like opening a public department store and more like signing a few lucrative industrial supply contracts.

The Meta data center deal would also make strategic sense because Meta is already planning spending on a breathtaking scale. The company has said it expects capital expenditures of between $125 billion and $145 billion in 2026, driven largely by AI data centers and related infrastructure. That figure is so large it changes the question from “Can Meta afford this?” to “What else can Meta do with all this capacity?”

Renting a portion of it out offers one answer. It could smooth the economics of new facilities, improve utilization during periods when Meta’s own workloads do not consume every available resource, and give the company a more tangible story about returns on AI investment. The idea has an obvious catch, though: capacity sold to Anthropic is capacity Meta cannot immediately use for Llama, Meta AI or its advertising systems.

AI’s compute arms race is becoming a landlord business

The more revealing part of a Meta data center deal is what it says about the market taking shape around AI infrastructure. The largest technology companies are no longer simply building products on cloud platforms. They are buying chips, locking up electricity, financing construction and potentially subletting capacity to one another.

Nvidia remains pivotal because its GPUs are the scarce machinery everyone wants, but data center access is becoming nearly as important. A leading AI lab can have excellent researchers and a compelling model roadmap, then still hit a wall because it cannot obtain enough compute at the right time. That reality gives companies with giant facilities unusual negotiating power.

There are risks. Anthropic would have to weigh how much operational dependence it wants on a company whose open-weight Llama models compete for developer attention. Meta, meanwhile, would need airtight technical and commercial boundaries around a customer that has access to some of its most strategically valuable assets. Data isolation, availability guarantees, power allocation and hardware scheduling are not details to sort out later.

Even so, the reported Meta data center deal fits the direction of travel. The AI race is no longer only about whose chatbot gives the cleverest answer. It is increasingly about who owns the digital factories behind those answers — and whether they can keep those factories busy enough to justify the electric bill.

We will see whether Meta and Anthropic get from talks to a contract. If they do, the more interesting question may be who calls next. Zuckerberg said the requests arrive almost weekly, and in this market, empty GPU capacity is starting to look a lot like vacant beachfront property.