Meta reportedly in talks with Anthropic on a potential $10 billion AI compute lease

Meta is reportedly discussing a major arrangement with Anthropic that would involve leasing AI computing capacity, according to a report carried by Yahoo Finance. The proposed deal is described as potentially reaching $10 billion in value, a scale that, if finalized, would make it one of the larger infrastructure commitments in the still-forming market for model training and deployment resources.

The report frames the concept as a compute lease, meaning Anthropic would be able to access Meta’s AI-focused infrastructure rather than building and operating all of the needed hardware and systems itself. In practical terms, such arrangements typically target the bottleneck that constrains AI development, namely access to sufficient compute capacity, including specialized processors and the networking and storage needed to run large training workloads.

For Meta, any compute-leasing partnership would be a continuation of its strategy to turn its technology stack and data-center buildout into a broader set of AI offerings and demand channels. While Meta is publicly known for building and deploying AI models across its advertising and social products, third-party demand for compute can also diversify monetization beyond its own model training cycles and internal usage.

For Anthropic, the appeal of a compute lease would be speed and scale. Frontier model development is expensive and time-sensitive, and even when an organization has access to capital and in-house engineers, it still depends on obtaining enough compute capacity on predictable terms. A large long-duration compute commitment, as implied by the reported $10 billion scale, would likely be aimed at reducing uncertainty around availability and scheduling of critical workloads.

The report does not, in the information provided here, spell out the exact structure of the arrangement. It does not confirm where the compute would run, whether it would be delivered through Meta’s existing cloud or a dedicated capacity program, or how the contract would allocate costs, performance metrics, or timing of capacity deliveries.

Meta also has not, in the material reflected in this review request, issued a public confirmation or a formal statement about talks with Anthropic. Without a corporate announcement or regulatory filing, key elements such as contract duration, commercial terms, and any governance or exclusivity conditions remain unverified.

In the wider industry, deals like this are increasingly common as AI companies seek reliable access to power-hungry data-center capacity and specialized chips. Even when model builders design their own architectures, they still rely on external or shared infrastructure, and competition among providers can shift from pure model quality to throughput, latency, and the reliability of compute delivery.