Anthropic in talks for $10 billion Meta AI computing lease

Anthropic has begun early discussions with Meta to lease up to $10 billion in artificial intelligence computing power over the next two years.

The proposed arrangement, initially forwarded by Anthropic in June, reveals the severe compute bottleneck still plaguing the artificial intelligence industry.

Under the early terms, Anthropic would make monthly payments to Meta, with both companies retaining the option to terminate the agreement before the full two years are up.

This possible deal with Meta comes on the heels of Anthropic’s recent efforts to expand its infrastructure. In May, the company signed a three-year, $45 billion agreement with SpaceX, giving it access to Nvidia processors at the Colossus 1 data center for $1.25 billion each month.

Even as Anthropic nears a $1 trillion valuation, the company still struggles to obtain enough hardware to keep up with growing demand for its Claude AI models.

A worldwide shortage of advanced Nvidia processors has forced Anthropic to limit how much users can access its most powerful models.

The rush to secure hardware is being felt across the technology industry. Google, for example, pays SpaceX about $920 million each month for processors and has begun to ration access to its Gemini model, even placing strict limits on Meta’s usage.

For Meta, the deal opens a profitable new revenue avenue. The company is set to invest an extraordinary $145 billion in capital expenditures in 2026, mainly to expand its artificial intelligence infrastructure.

Following the strategic cut of 8,000 jobs in May to fund this expansion, leasing surplus capacity to Anthropic would give CEO Mark Zuckerberg a highly profitable answer for anxious shareholders.

If the agreement goes through, it would set up an unusual situation in the industry. Meta, which is developing its own Llama models to compete with Claude, would become both a key competitor and a key infrastructure provider for Anthropic.