A Surprising Duo in the AI Arms Race
Anthropic has started very preliminary conversations with Meta regarding the rental of computing resources from the social media giant.
The discussions highlight the difficulty AI companies encounter in obtaining Nvidia chips. Anthropic has been making sizable commitments with other companies to lock in access to AI hardware. Just weeks prior, Anthropic struck a comparable agreement with Elon Musk's SpaceX to tap into the computing power of its Colossus 1 data center.
Why Meta Might Become a Cloud Landlord
Meta has hinted at this move before. Back in May, Meta CEO Mark Zuckerberg stated that the social media firm was looking into providing cloud computing services. And last October, Zuckerberg said, "if we have compute that they could buy from us at some premium to what we've bought it at."
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
Meta has been spending heavily on AI infrastructure. By 2026, the company's capital expenditures - including what it spends on AI infrastructure - could reach as much as $145 billion. Dave Brown, a former longtime top executive at Amazon Web Services, is expected to join Meta, according to CNBC.
What the Deal Says About the AI Chip Squeeze
Usage caps are applied by Anthropic on its most sophisticated models, including Fable. These restrictions stem partly from the massive computing power required to run such models and partly from the ongoing scarcity of essential chips. Even a well-funded AI lab cannot simply buy all the Nvidia chips it wants.
So the lab is getting creative. It is locking up capacity at SpaceX, at Meta. These deals are not just about money. They are about guaranteeing that when a model needs to run, the hardware is there.
The shortage of advanced semiconductors has become a bottleneck for AI development globally. Nvidia's latest Blackwell chips are sold out for months in advance, and cloud hyperscalers have locked in long-term contracts. For a company like Anthropic, which relies on Nvidia's hardware to train and deploy its models, securing compute capacity is as critical as the software itself. This is why Anthropic is pursuing multiple non-traditional partners, including SpaceX and Meta, rather than relying solely on major cloud providers.
The challenge reflects a broader industry reality: the supply of advanced chips cannot keep pace with surging demand from every major tech firm. Cloud providers like Microsoft, Amazon, and Google have already reserved vast amounts of Nvidia's next-generation hardware, leaving independent AI labs scrambling for scraps. By striking side deals with companies that have spare compute - like Meta and SpaceX - Anthropic is effectively buying itself a place in line, ensuring its models can be trained and deployed even if the traditional cloud market tightens further.
This potential collaboration underscores a broader trend where companies with large data centers are becoming alternative compute providers. As the AI arms race intensifies, the ability to secure hardware has become a strategic advantage. For Meta, leasing out spare capacity could help offset its massive infrastructure investments, while for Anthropic, it offers a vital alternative to the dominant cloud platforms.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
