Meta logo and signage Image: Reuters
by VibecodedThis

Anthropic Is Now in Talks to Buy $10 Billion in Compute From Meta

Reuters reported Thursday that Anthropic is in early negotiations to lease GPU capacity from Meta in a deal worth up to $10 billion over two years. It would be the second major compute partnership Anthropic has announced in a month.

Share

Anthropic is in early talks to lease computing power from Meta in a deal that could reach $10 billion over two years, Reuters reported Thursday.

The arrangement would have Anthropic paying Meta in monthly installments for access to GPU capacity. Both sides would have the option to exit early. Anthropic proposed the deal in June; Meta is still evaluating it. Neither company has commented publicly.

Meta enters the compute-leasing business

Meta isn’t a cloud provider. It doesn’t have commercial infrastructure products like AWS, Azure, or Google Cloud. It builds data centers and AI clusters for its own models — Llama and derivatives — and for its own internal workloads across Facebook, Instagram, and WhatsApp.

But it has excess capacity, and it has been eyeing ways to monetize it. At Meta’s shareholder meeting in May, CEO Mark Zuckerberg said entering cloud computing was “definitely on the table.” He noted that companies were approaching Meta “almost every week” looking to buy access to spare compute or AI models.

That pipeline apparently includes Anthropic.

The deal structure would put Meta in competition with neocloud providers like CoreWeave and Nebius, which have built businesses on exactly this model: acquiring hardware and selling GPU time to AI companies that need more compute than they can build or buy directly.

The oddity here

This is a deal between companies that compete directly on AI. Meta builds and releases Llama, an open-source model family. Anthropic makes Claude, which competes with Llama across enterprise and developer use cases. If the compute deal closes, Meta would be a critical infrastructure supplier to a direct competitor.

That’s not unprecedented in tech — hyperscalers routinely provide cloud services to companies building products that compete with their own. But it’s unusual in AI at this particular moment, when model competition is still intense and the underlying compute is still genuinely scarce.

Anthropic’s second big compute deal this month

Four days earlier, Anthropic announced a compute deal with SpaceX, giving it access to the full capacity of the Colossus 1 data center in Memphis. SpaceX’s Grok division competes with Anthropic, too.

The pattern is consistent: Anthropic needs GPU time it can’t build fast enough internally, and it’s acquiring it from companies with spare capacity, even when those companies are also competitors. The SpaceX deal came bundled with a rate limit increase for Claude Code users.

If the Meta deal closes, it would dwarf the SpaceX arrangement in scale. Anthropic’s annualized revenue is reportedly around $47 billion. A $10 billion compute lease over two years represents significant but manageable infrastructure spending at that revenue base.


Sources: Reuters via Yahoo FinanceCNBC

Share