Meta Platforms is in early talks to lease computing power to Anthropic in a potential arrangement worth up to $10 billion over two years, a deal structure that would signal a significant shift in how large AI infrastructure owners monetise compute capacity. The talks are preliminary and may not result in an agreement, but the proposal shows how fast-growing demand for AI workloads is creating new commercial models beyond traditional public cloud contracts.
Under the discussed structure, Anthropic would pay Meta monthly over a two-year period. The terms remain subject to change, and the companies would be able to exit an agreement early. Anthropic is understood to have proposed the arrangement in June, while Meta is still considering it. The discussions are complicated by the fact that Meta does not currently operate a conventional business selling computing power to third parties.
The potential transaction would give Meta a path to diversify revenue beyond advertising by turning part of its infrastructure base into a commercial compute business. It would also put the company closer to competition with newer AI infrastructure providers such as CoreWeave and Nebius, which have grown around the market’s demand for high-performance GPU clusters and AI-ready data-centre capacity.
The talks follow comments by Meta Chief Executive Mark Zuckerberg at the company’s shareholder meeting in May, where he said entry into cloud computing was “definitely on the table” and noted that firms were approaching Meta almost every week to buy access to its AI models or spare computing power. Meta has also been reported to be building a cloud business that could sell excess computing capacity and host AI models for developers.
For Anthropic, the possible arrangement would add to a broader compute procurement strategy. The company struck a deal in May with SpaceX to tap the full computing power of the Colossus 1 data centre in Memphis, Tennessee. The scale of the Meta discussions indicates how frontier model companies are diversifying infrastructure access as training, inference, coding-agent use and enterprise deployments place pressure on available capacity.
The development comes as AI compute is becoming a strategic market in its own right. Large technology companies are investing heavily in data centres, GPUs, networking and power contracts, while model developers are seeking guaranteed access to infrastructure that can support rapid product growth. For India-facing enterprise buyers, developers and IT services firms, these global infrastructure deals influence model availability, pricing, latency, platform competition and the economics of deploying advanced AI systems at scale.
