Full Breakdown
Meta Explores $10 B AI Computing Lease with Anthropic
8/10/2026, 10:51:04 AM
Core Deal Overview
Three people with knowledge of the discussions said Meta is in early-stage talks to lease computing power from its artificial-intelligence data centers to Anthropic. The proposed arrangement could be worth as much as $10 billion over two years, with Anthropic paying Meta in monthly installments. Both parties would retain the right to terminate the agreement early, according to the same sources.
Context of AI Compute Demand
Anthropic’s proposal, submitted in June, is roughly one-third the size of the $45 billion, three-year contract the startup signed with Elon Musk’s SpaceX in May, which translates to about $1.25 billion per month for compute capacity. The talks reflect a broader scramble among leading AI firms for additional compute resources, as tech giants such as Meta, Google and Microsoft pour “hundreds of billions of dollars” into building new data centers worldwide.
Financial and Operational Details
Meta’s chief executive, Mark Zuckerberg, has indicated the company will spend up to $145 billion this year on AI initiatives—more than double the $72 billion expended the previous year. A lease to Anthropic would represent a new revenue stream that could help offset the massive capital outlays required for the data-center expansion.
Official Stance and Market Implications
Both Meta and Anthropic declined to comment on the negotiations. Analysts note that investors have been questioning the scale of Meta’s data-center spending, and a successful lease could provide a tangible return on that investment. Wall Street observers have expressed concern about whether such large-scale infrastructure spending can be justified without clear revenue-generating outcomes.
Potential Impact on Meta’s Business
If finalized, the deal could open a distinct line of business for Meta, diversifying its income beyond advertising and potentially easing pressure from shareholders wary of the company’s heavy AI-related expenditures. The arrangement would also signal a growing market for “compute-as-a-service” offerings among AI startups seeking to accelerate model development.
