Meta and Anthropic are reportedly discussing a compute-lease arrangement that could reach $10 billion over two years

Meta Platforms is reportedly in talks with Anthropic about leasing computing power in an arrangement that could total as much as $10 billion over roughly two years, the New York Times reported, citing three people with knowledge of the discussions. The report, carried by Yahoo Finance, did not say when the talks began, whether an agreement has been reached, or which specific Meta infrastructure would be involved.

A compute lease is a contracting model in which one company provides access to processing capacity, while the customer secures the right to use that capacity for training and inference, the two main types of work required for modern artificial intelligence. For a provider like Meta, such deals can potentially translate expensive data-center buildouts into steadier revenue. For an AI developer like Anthropic, they can reduce the need to fully build out and operate every part of the required infrastructure in-house.

The New York Times report framed the discussions as still in progress, which matters because the terms of these arrangements often determine how much value actually transfers. Details typically include the pricing structure, the volume of capacity reserved, performance expectations, and how disputes are handled if capacity is constrained. None of those elements were disclosed in the report as summarized by Yahoo Finance.

Meta, for its part, has been investing heavily in data centers and AI infrastructure as it increases the scale of its models and related products. Anthropic, meanwhile, is one of the best known AI labs focused on building frontier models. In that context, a compute-lease path can become a way to manage timelines, costs, and supply for model development.

Market participants are likely to view a potential large compute deal as a announcement about how quickly demand for AI training and serving resources is moving from “capacity planning” into “capacity contracting.” Large customers often seek more predictable access to GPUs and data-center power, while large operators look for longer-term commitments that can help underwrite capital spending.

Even if the talks do progress, the public record may remain limited for a time. The report did not include a final price, a signature date, or any formal disclosure by either company. That leaves investors and customers without confirmation of whether the purported $10 billion figure is tied to guaranteed usage, optional capacity, or a broader package that could include related services.

The next point to watch is whether Meta or Anthropic provides any update through earnings commentary, regulatory filings, or official announcements. If a contract is signed, disclosures may appear indirectly in guidance tied to capital intensity, operating expenses, or changes in the mix of revenue tied to AI infrastructure services.