Meta is reportedly negotiating a deal worth up to $10 billion over two years to lease computing power to Anthropic, according to sources cited by the New York Times. If completed, the arrangement would open a significant new business line for Meta while addressing Anthropics growing need for computational resources to train and run its frontier AI models.
Under the reported terms, Meta would provide Anthropic with access to its data center infrastructure, allowing the AI company to train its models on Metas vast computing network. This would help Anthropic alleviate what has become one of the most critical bottlenecks in the AI industry: access to sufficient high-performance computing capacity.
For Meta, the deal represents a strategic pivot toward monetizing its hardware investments. The company has spent tens of billions of dollars building out AI infrastructure, including data centers and custom silicon, to power its own AI research and products. Leasing excess capacity to other AI companies would generate revenue from those investments while strengthening ties with a leading AI research organization.
The negotiations come at a time when demand for AI computing power is outstripping supply across the industry. Companies like Anthropic, OpenAI, and Google DeepMind face intense competition for Nvidia GPUs and other specialized hardware needed to train large language models. Anthropic, in particular, has been aggressively expanding its compute capacity as it develops its next-generation models, including Fable 5 and Mythos.
Neither Meta nor Anthropic has publicly confirmed the negotiations. If finalized, the deal would be one of the largest compute leasing arrangements in the AI industry, reflecting the enormous scale of infrastructure required to compete at the frontier of AI development.
The reported talks also highlight a broader trend of technology companies finding new ways to extract value from their capital-intensive AI infrastructure investments beyond their own product roadmaps.
This article was adapted from BeInCrypto. Read the original here.
