Meta is in talks to lease computing power to Anthropic in a potential deal worth up to $10 billion, according to a source familiar with the matter. Reuters reported the story on July 17. Anthropic proposed the deal, and Meta is considering it. The terms remain subject to change, and both companies declined to comment. CNN’s source described any specific numbers as speculative. The discussions may not result in a deal.
But if they do, the company building Llama becomes the landlord for the company building Claude. And that restructuring of the AI infrastructure market matters for every small operator buying API credits.
What’s Actually on the Table?
Anthropic would pay Meta in increments over the period.
The companies would be able to exit any agreement early. The New York Times said Meta would lease computing power, calling the deal a potential step toward a new AI business for Meta.
The talks are in the early stages.
AI Stock Wire confirmed the deal is not signed. CNN said the negotiations are still in early stages. The Chosun report said the negotiations could fall through. So treat every dollar figure as a ceiling, not a commitment.
What makes this structurally interesting is who leads the effort. AI Weekly reported that the deal is part of a nascent Meta Compute unit led by Santosh Janardhan, Daniel Gross, and Dina Powell McCormick. That’s not a skunkworks project. That’s a named business unit with senior operators attached. Meta is building a compute leasing operation, and Anthropic would be the anchor tenant.
Why Would Anthropic Pay a Competitor?
AI Stock Wire noted the unusual dynamic: Meta builds Llama models and Anthropic builds Claude. They compete for the same developers, the same company contracts, the same mindshare.
Anthropic paying Meta for compute means Anthropic needs infrastructure more than it needs competitive isolation.
That tells you something about the compute crunch. If you’re Anthropic and you’re negotiating a deal with the company training the model that directly rivals yours, you’ve exhausted the easier options. The alternative providers are either too expensive, too constrained, or too slow to deliver. The GPU shortage isn’t a temporary supply chain hiccup anymore.
It’s a structural bottleneck, and the companies with the most compute hold decisive power over everyone else.
For small operators, this is the signal to watch.
When the second-largest AI model company in the world can’t get enough compute without going to its biggest rival, the pricing power sits entirely with the infrastructure owners. Your API costs are downstream of these deals.
What This Means for Meta’s AI Strategy
Meta has spent aggressively on AI infrastructure. The New York Times framed this deal as a step toward a new AI business for Meta, and that framing matters. Meta’s core revenue is advertising.
AI compute leasing would be a fundamentally different line of business with distinct margins, separate customers, and other competitive dynamics.
The Meta Compute unit is the tell.
You don’t staff a unit with Janardhan, Gross, and McCormick to run a one-off deal. You staff it because you see a recurring revenue opportunity. If the Anthropic deal closes, expect Meta to open the doors to other AI companies. The model is straightforward: build more data center capacity than you need for your own models, then lease the excess to companies that can’t build their own.
This is the same playbook that turned AWS from Amazon’s cost center into its profit engine. The difference is that Amazon started with spare retail infrastructure. Meta is starting with purpose-built AI data centers. The economics could be even better.
What Small Operators Should Do About It
If you’re running a small agency or solo operation that depends on AI APIs, this consolidation matters.
The number of companies that control the underlying compute is shrinking. Meta, SpaceX, Google, and Microsoft are the serious players. Everyone else is renting from them.
Here’s what I’d do right now. First, don’t lock your entire stack to one provider. If your application runs on Claude, test it on Llama and GPT equivalents so you can switch when pricing changes. Second, track the infrastructure deals. When Meta Compute signs Anthropic, pricing for Claude API calls will reflect what Anthropic is paying Meta for compute. That cost flows downstream to you. Third, evaluate open-weight models for high-volume tasks. If you’re running millions of inference calls, the difference between a hosted API and a self-hosted open model compounds fast.
The deal isn’t signed.
The numbers are speculative. But the strategic direction is clear. The companies that own the compute are becoming the landlords of the AI economy. And everyone building on top is a tenant. Plan accordingly.
Sources: Reuters via Yahoo Finance, New York Times, CNN, AI Stock Wire, AI Weekly
