Alphabet is guaranteeing lease payments on five U.S. data centers totaling $35 billion in committed capacity, a financing structure that places Google as both equity investor and infrastructure underwriter for Anthropic's Claude model expansion. The arrangement covers facilities already under construction or recently operational, with Anthropic named as primary lessee and Google providing payment assurance to the facility owners.
The deal marks a shift from check-writing to balance-sheet commitment. Google has been an Anthropic equity participant since early 2023, contributing roughly $2 billion across two funding rounds, but this lease guarantee represents a different class of obligation — multi-year, non-cancelable, and tied to physical assets rather than valuation milestones. The five data centers are spread across Virginia, Ohio, and Texas, with power allocations ranging from 150 MW to 300 MW per site. Anthropic has publicly stated it requires 10x the compute capacity by late 2025 to sustain training roadmaps for Claude 4 and subsequent releases, a timeline that aligns with these facilities reaching full operational status.
The structure matters because it reveals how hyperscalers are managing AI capital allocation without triggering immediate GAAP expense recognition. By backstopping lease payments rather than purchasing compute outright, Google defers the cash outlay while locking Anthropic into its cloud ecosystem. The data centers will run Google Cloud infrastructure — TPU v5 pods and Axion ARM instances — ensuring that Anthropic's inference and training workloads remain within Google's margin stack. This is not altruism. It is vertical integration through financing.
The arrangement also signals that data center developers are no longer willing to bear tenant risk on AI-scale builds. The $35 billion figure likely reflects 12 to 15 years of lease commitments at current data center rental rates, which have climbed 40% since early 2023 in markets with reliable power access. By requiring a Google guarantee, the developers are pricing in the possibility that Anthropic's revenue growth does not keep pace with its infrastructure burn — a reasonable hedge given that Claude's commercial traction, while real, has not yet produced the cash flow to self-finance this scale of capacity.
Operators should track three follow-on developments. First, whether Microsoft structures a similar guarantee for OpenAI's pending data center expansion in Louisiana and Arizona, expected to be formalized by Q2 2025. Second, whether Anthropic begins offering dedicated cloud instances to enterprise customers by mid-2025, which would be the logical revenue path to justify this lease burden. Third, whether Google starts booking these lease guarantees as contingent liabilities in its 10-Q filings, which would surface in the March 2025 quarterly report.
The $35 billion is not a headline number. It is a 15-year run rate, and it places Google as the lender of last resort in a market where AI training costs are still climbing faster than AI product revenue.