Google to backstop Anthropic’s costly AI chip leases in a roughly $35 billion financing arrangement

Alphabet’s Google is backing a major expansion of Anthropic’s AI infrastructure through a chip-and-data-center financing structure that could total about $35 billion, according to market reporting. The arrangement centers on Anthropic, the company behind the Claude family of artificial intelligence models, which is leasing powerful compute chips for use across multiple data centers.

Reports say Anthropic will lease high-end chips at five data centers as part of the scale-up, with Google agreeing to backstop lease payments at each location. In practical terms, a “backstop” functions like a guarantee that payments will still be covered even if the borrower’s cash flows fall short, reducing risk for lenders and other financing partners. That structure reportedly helps Anthropic obtain what amounts to a very large loan backed by the lease commitments.

Financial backing is described as involving private finance groups and chip-supply partners, with one report pointing to Apollo and Blackstone financing the expansion and a connection to Broadcom’s role in custom chips for the effort. Another market report also describes Google’s support as undergirding the financing tied to the chip leases, and says the deal’s data-center footprint has not been widely reported before.

While the overall figure of $35 billion is attributed to the financing size in the reporting, details on exact terms were not included in the accessible material. One additional report adds operational color, saying Google would supply tensor processing units, or TPUs, for the new data centers starting in 2027. TPUs are specialized hardware accelerators designed by Google to speed up machine-learning workloads, including the training and inference of large AI models.

For Google and Alphabet, the deal underscores the company’s broader strategy of positioning its cloud and AI compute stack as the foundation for top model builders. Google already provides machine-learning infrastructure used across the industry, and these types of long-term chip and capacity arrangements can also create sticky demand for cloud services, security, and ongoing compute needs.

More broadly, the reports reflect the competition among major tech companies, AI startups, and investors to lock in scarce computing capacity for the next generation of AI models. As hyperscale data-center construction has accelerated, so has the use of complex finance structures to spread the cost and manage risk around chip supply, power, and contracted capacity.

Still, several key specifics remain unclear from the publicly available excerpts. The reporting does not spell out the exact identity of the financing entity or the precise contract mechanics for the lease guarantees, nor does it state how much of the $35 billion figure is debt versus equity, or the duration and pricing of the chip leases. Google, Anthropic, and the financing partners were not quoted directly in the accessible excerpts, and Alphabet’s official communications were not included in the material provided.

What to watch next is whether Alphabet discloses more about the arrangement in investor communications or cloud partnership updates, and whether Anthropic provides further detail on the number of data centers, the timing of capacity ramp, and the expected compute scale tied to Claude. Those disclosures would help clarify how much incremental demand the arrangement creates for Google’s AI infrastructure and how quickly the expansion could translate into delivered model performance.