Meta is negotiating a potential $10 billion deal with Anthropic to lease AI computing power over two years, according to the New York Times.
This agreement could mark a significant revenue stream for Meta amidst its vast AI infrastructure spending, while providing Anthropic with much-needed compute resources without building its own data centers.
Anthropic proposed the deal in June, envisioning monthly payments with an early-exit clause for both sides, reflecting flexible terms in a nascent partnership.
Yet, the scale is relatively modest: this $10 billion deal amounts to about one-third of Anthropic’s $45 billion three-year compute contract with SpaceX signed earlier this year.
The ongoing compute arms race among AI firms pushes companies like Meta, Google, and Microsoft to invest billions in global data center expansions, stirring investor skepticism over whether such spending will generate adequate returns.
The urgency to secure compute capacity is clear: Anthropic requires extensive resources to train sophisticated models, while Meta is transforming from a model laggard to a competitor with frontier AI capabilities rated near A-minus to B-tier by experts such as Theo Jaffee from MTS.
For Meta, renting compute to a direct AI competitor is unconventional but strategically sensible. It can monetize excess hardware amid a $145 billion AI budget for 2026, more than double last year’s $72 billion.
Meta’s CEO Mark Zuckerberg has acknowledged that the company’s AI investments have yet to pay off fully, making such deals key for offsetting infrastructure costs and appeasing shareholders concerned about heavy expenditures.
Both parties have declined to comment publicly, and talks remain at an early stage with no guarantee of closure.
This deal, if finalized, will highlight how AI infrastructure is becoming a critical battleground where tech giants balance internal innovation with collaborative compute leasing to manage risk and costs.
This article is informational and not financial advice.
