Last October, Mark Zuckerberg mentioned almost in passing that companies kept asking Meta if they could buy computing capacity from it, at a premium. It sounded like a hypothetical. On July 17, the New York Times turned it into a very real story.

Meta and Anthropic are in early talks for a potential computing deal worth as much as $10 billion over two years, the New York Times reported, citing three people with knowledge of the discussions. Anthropic proposed the arrangement in June. Meta is reviewing it. Both companies declined to comment, and the talks are early enough that they may not result in a deal at all.

What the Meta Anthropic $10 billion AI compute deal says

The basic structure, as the Times described it, would have Anthropic paying Meta in monthly installments over two years with an option for either party to exit early. CNBC independently confirmed the talks, according to CNBC. CNN also confirmed the conversations but noted its source said any specific dollar figures in the reports are speculative.

More AI:

• Workers just sent AI companies an ultimatum
• Palantir CEO has a blunt verdict on OpenAI and Anthropic
• Elon Musk pulls no punches with AI rivals as Grok 4.5 debuts

Meta stock fell as much as 6% on July 17 before paring losses after the report came out, ending the day down about 2%.

The potential deal would be smaller than Anthropic’s existing arrangement with SpaceX, which signed a $45 billion, three-year compute deal in May giving Anthropic access to the Colossus 1 data center in Memphis. A Meta arrangement would layer on top of that, giving Anthropic yet another major source of GPU capacity.

What this says about Meta’s cloud computing ambitions

For Meta, the Anthropic talks are the clearest sign yet that the company is serious about entering the cloud computing business. Zuckerberg said in May that Meta was considering it as a way to show investors that its AI spending can generate revenue beyond advertising. The internal name for the effort is already circulating: Meta Compute.

The infrastructure is already being built. Meta is expected to spend as much as $145 billion on capital expenditures in 2026, more than double the $72 billion it spent last year, mostly on AI hardware and data centers.

The company cut 8,000 jobs in May while redirecting billions toward AI buildout. It also recently hired Dave Brown, a former senior executive at Amazon Web Services, a move that signals the cloud ambitions go beyond a single leasing deal.

“We hear from companies regularly that are asking if we have compute that they could buy from us at some premium to what we’ve bought it at,” Zuckerberg said in October 2025.

Anthropic is the first company publicly reported to be in discussions to do exactly that.

Michael/Getty Images

Why Anthropic keeps signing AI compute infrastructure deals

Anthropic’s compute appetite has been one of the defining stories of the AI industry in 2026. The company has placed usage limits on its most advanced models, including Claude Fable, because it doesn’t have enough processing capacity to run them without restrictions.

Revenue has been growing fast but the infrastructure needed to support that growth hasn’t kept pace. The SpaceX deal in May was one answer. A potential Meta arrangement would be another.

Anthropic is also IPO-bound, according to Reuters, with bankers setting up investor meetings ahead of a possible October listing. Going into a public market roadshow with multiple long-term compute partnerships locked in is a very different story than going in with a single supplier and a capacity constraint.

The pattern Anthropic is building looks less like a single vendor relationship and more like a distributed infrastructure network, which reduces dependency on any one partner and gives the company more negotiating leverage with each of them.

What the Meta Anthropic deal means for AI investors

The most interesting wrinkle in the reported deal is what it would make Meta. The company already builds and releases its own Llama AI models, which compete directly with Anthropic’s Claude. A compute lease would make Meta Anthropic’s infrastructure provider at the same time it’s competing with Anthropic’s product.

That sounds unusual but it’s already the norm in AI infrastructure. SpaceX sells GPU access to both Anthropic and Google. The compute shortage has made competitive boundaries largely irrelevant when it comes to infrastructure.

The companies that have capacity can charge a premium for it, and the companies that need capacity will pay, regardless of what the two sides are doing on the model side. For investors watching the broader AI trade, that matters. The infrastructure layer is becoming a business in its own right, separate from who wins the model race.

Meta monetizing its data centers while also competing in AI models is roughly what Amazon did with AWS: build infrastructure for yourself, then sell the excess to everyone else, including your competitors.

Related: Anthropic just made a move that changes the AI investing story