Meta and Anthropic near $10B cloud deal as AI compute wars escalate

Meta is in advanced talks to rent massive AI computing capacity to Anthropic in a deal that could reach $10 billion over two years, according to the New York Times. It would be the social media giant’s first major compute-as-a-service customer — and a telling pivot for a company that’s been pouring cash into infrastructure faster than its own AI projects can consume it.

Mark Zuckerberg telegraphed this move months ago, telling investors that excess capacity could become a revenue stream if internal demand didn’t keep up. Looks like it hasn’t. Meta’s capex is staggering: up to $145 billion this year, mostly on AI. When you’re spending that kind of money, idle GPUs are a liability. Renting them out to Anthropic turns a potential sunk cost into recurring revenue. Smart, if the numbers hold.

Anthropic’s interest isn’t hard to decode. Claude Code demand has exploded, and the startup already inked a $45 billion deal with SpaceXAI for additional capacity. The company is hungry for compute and doesn’t want to be bottlenecked by infrastructure shortages while competing with OpenAI and Google. But here’s the wrinkle: Anthropic is simultaneously hiring ex-Google data center executives to build its own facilities. This rental arrangement smells like a bridge, not a destination.

What’s unsaid is whether this deal changes the competitive dynamics between the companies. Meta open-sources Llama; Anthropic keeps Claude tightly controlled. They compete for enterprise customers and developer mindshare. A $10 billion landlord-tenant relationship complicates that rivalry. I’m watching for whether the contract includes any restrictions on what Meta can do with the usage data it inevitably sees — or whether Anthropic even cares, given its long-term plan to move out.