Meta Anthropic Compute Deal: When Rivals Become Landlords

Anthropic can’t build data centers fast enough to keep up with Claude’s demand — so it’s about to start renting spare computing power from Meta, a company it competes with every single day. The Meta Anthropic compute deal could be worth $10 billion. Here’s why AI’s biggest names are quietly becoming each other’s landlords.

Everything about the AI industry revolves around one thing: computing power. Right now, the world does not have enough of it, not even for an organization as financially robust as Anthropic, makers of Claude.

According to the CNBC report confirmed by Anthropic, Meta and Anthropic are in talks regarding a deal that could see Anthropic leasing AI compute from Meta’s data centers in an arrangement worth up to $10 billion over two years. The deal is yet to be signed and both organizations have remained silent about it until this point. However, if the deal happens, it will prove to be one of the most peculiar partnerships made in 2026’s AI boom, where two competing entities will be working together in order to provide infrastructure for each other.

The Details of the Meta Anthropic Compute Deal

Under the proposed deal, Anthropic will make payments in structured monthly payments over two years. Both parties will have the ability to terminate the deal prematurely if necessary. Talks on the matter are reported to have started quietly back in June, when Anthropic reached out to Meta with a proposal not the other way around.

The importance of this detail cannot be overstated. It shows us that Meta is not desperately trying to find a new revenue stream, it is Anthropic that is looking to procure whatever compute it can get.

Why Anthropic Needs Meta’s Computing Power 

The problem of demand is definitely not one that is bothering Anthropic. The problem lies in supply. Creation and maintenance of frontier AI models require huge arrays of specialized chips and power-hungry data centers and the shortage of such capability has become the main bottleneck of the whole industry. Usage restrictions imposed by Anthropic on its models occur because it does not have sufficient hardware to process all of its customers’ requests.

And it is not even Anthropic’s first such move. Just weeks before the talks with Meta began, Anthropic entered another deal this time with Elon Musk’s SpaceX. Under the agreement, Anthropic will be paying SpaceX approximately $1.25 billion per month close to $45 billion over the entire term of the contract for access to the Colossus data center in Memphis. The fact that even a company that is valued at approximately $1 trillion is willing to commit to such recurring payment solely to secure chips shows us everything we need to know about the tightness of compute market.

What Meta Gains From the Anthropic Compute Deal 

Of course, it is not only about recouping the expenses. Meta is set to spend as much as $145 billion on AI infrastructure in 2026 twice as much as the previous year and the investors are getting increasingly worried about whether this investment will ever yield dividends.

Rental of excess capacity solves both of these problems at once. It allows turning idle infrastructure into an immediate revenue source and helps Meta enter the cloud computing industry, competing directly with companies like CoreWeave and Nebius. Mark Zuckerberg mentioned in May that companies were approaching Meta almost every week to purchase access to Meta’s AI models and excess compute entering cloud computing business was definitely on the table for Meta.

The Actual Story: AI’s New “Frenemy” Economy

One aspect of the situation that has been overlooked by most media covering the deal is the following: this is not a story about one particular deal worth $10 billion. This is a story about a trend.

In the span of about two months, Anthropic has now twice turned from avoiding their rival to reliance on them first with SpaceX, whose CEO is known to have publicly criticized Anthropic and now potentially with Meta, whose own AI products compete directly with Claude. Such is the scarcity and price of compute, that ideological rivalry and market competition take a backseat to a more straightforward question: who has the spare chips right now?

This is quite a significant change in the actual way the AI industry operates behind the scenes. Publicly, AI laboratories are painted as engaged in an arms race against each other. Privately, they are becoming suppliers, landlords and even lenders to each other because no organization, no matter how well-funded, is able to build data centers fast enough by themselves.

What Will Happen Next

It is possible that nothing of the sort will happen. Both parties could decide not to finalize their talks, terms could change or the whole deal could simply not come to fruition. However, the very fact that talks are happening in public is already quite telling. Shares of Meta initially fell more than 5% on Friday amid a broader technology sector sell-off, then pared losses to close down 2% after news of the deal surfaced  clearly showing that the investors consider this as a promising revenue stream for Meta, not some random buzz.

TheTweaks Verdict

Whatever may become of this deal, it matters to us mostly as a signal. It shows us that computing power, not funding, not talent, not even model quality has become the single most limiting factor in AI. Companies are ready to pay millions to their rivals instead of waiting months or years for construction of their own capacity. Expect more such deals before compute shortage eases and expect the line between “competitors” and “suppliers” in AI to blur further in 2026.