Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal
The New York Times ● Covered by 8 sources
Anthropic pitched Meta a deal to rent its computing power, worth up to $10 billion over two years. Meta gets a new revenue stream from AI without having to win the model race itself.
Based on reporting by The New York Times — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic went to Meta in June with a pitch that sounds almost backwards at first glance: pay us to let us use your computers. The proposed arrangement could be worth as much as $10 billion spread across two years, structured with monthly payments and built-in exit ramps for either side to bail early if things don't pan out.
What makes this interesting is the role reversal. Meta has spent the last two years pouring tens of billions into GPUs, data centers, and custom silicon, all in service of building Llama into something that can go toe-to-toe with GPT and Claude. That bet hasn't fully paid off yet in terms of matching demand for Meta's own AI products. So instead of letting excess compute sit idle, Meta gets to become a landlord of sorts, renting out capacity to one of the very labs it's supposedly competing against.
For Anthropic, the calculus is straightforward: training and running frontier models eats compute faster than almost any single company can build for, and diversifying suppliers beyond Amazon and Google (its two big backers) reduces risk if any one partner hits capacity limits or changes terms. Ten billion dollars is a serious hedge against that kind of bottleneck.
The deal also says something about where the real money in AI might end up sitting. Model makers burn cash chasing capability, but the companies that own the physical infrastructure, the chips, the power contracts, the data center real estate, are positioned to profit almost regardless of who wins the model wars. Meta building out compute for its own ambitions and ending up as a computing landlord for a rival isn't a contradiction. It might just be the more reliable business.
My take — AI-written commentary, not fact-checked reporting
This is the tell that infrastructure, not model quality, is becoming the real chokepoint in AI, and I'd bet we see more of these frenemy compute deals as everyone realizes GPUs are the new oil pipelines. Meta hedging by renting to Anthropic while still chasing its own AGI dreams is smart, if a little awkward, and it's a reminder that open-weight ambitions and closed-model rivals can share a data center without anyone blinking.
Read more about this at: The New York Times