Anthropic is reportedly lining up U.S. data-center capacity with Google-backed financing, a bet on scaling beyond cloud-only deployments

Anthropic, the AI lab backed by major technology investors, has reportedly begun signing preliminary agreements for U.S. data-center leases, totaling more than 1 gigawatt of power capacity, according to a report carried by Yahoo Finance that cites The Information.

The agreements, described as more than a dozen separate arrangements, would give Anthropic early control over where the computing capacity needed for training and running large language models will be located. That marks a shift from a model where AI startups depend primarily on third-party cloud providers’ capacity as demand ramps.

Crucially, the same report says Google is providing backing connected to the financing for these data-center commitments. While that does not necessarily mean Google is building the facilities, it suggests Google’s financial or commercial involvement is intended to reduce uncertainty for Anthropic as it scales infrastructure demands.

For Alphabet and its Google unit, the potential upside is straightforward but indirect. If Anthropic’s growth translates into heavier demand for data-center services, power, and connectivity, Alphabet can benefit through the financing structure and related commercial relationships, even if the hardware sits in Anthropic-leased space rather than purely within Google’s own cloud footprint.

For Anthropic, securing committed capacity can be a hedge against the reality that data center availability, grid interconnection, and power allocations are often bottlenecks for AI. Leasing deals at the facility level, even at a preliminary stage, can also help Anthropic plan around timelines for equipment installation and operational readiness.

The move fits a broader industry pattern. As generative AI adoption accelerates, companies increasingly treat infrastructure capacity, not just model development, as a competitive factor. That has pushed some AI labs to pursue direct capacity agreements while still retaining flexibility to use cloud services for peak demand or specific workloads.

What is not clear from the report is the final form of the arrangements. It is not specified which specific operators or facility providers are involved, what terms apply if leases are finalized or cancelled, or how much of Anthropic’s compute consumption would run inside these leased sites versus on existing cloud platforms.

Alphabet has not, in the reporting and material referenced here, publicly detailed the nature or size of its “backing,” and Anthropic did not disclose the agreements’ financial structure or schedule in a primary statement cited by the report. That leaves open questions about how binding the deals are and how quickly capacity will become usable.