Google Backstops $35 Billion Chip Deal to Keep Anthropic Running on Its TPUs

Google has agreed to backstop lease payments at five U.S. data centers, enabling a $35 billion chip financing deal that locks Anthropic into Google's custom TPU ecosystem for years to come, Bloomberg reported on June 9. The arrangement gives Anthropic — one of the world's most compute‑hungry AI labs — guaranteed access to purpose‑built chips, while Google secures a long‑term anchor tenant for its silicon business at a moment when AI infrastructure spending is the defining competitive metric.

The debt financing is being led by Apollo Global Management, in partnership with Blackstone and a syndicate of global banks, as part of a new Broadcom AI XPV Platform targeting over 20 gigawatts of compute capacity through 2028, Apollo announced in a press release. The initial $35 billion transaction directly supports Anthropic's previously announced expansion of more than 1GW of compute for training and inference starting mid‑2026.

The structure treats AI compute as an infrastructure asset class — similar to how real estate or aircraft are financed — keeping the debt off Anthropic's balance sheet. The $35 billion is split into three tranches: $6 billion in A1 notes, $25 billion in A2 notes, and $4.5 billion in B notes, according to AI Weekly, which summarized Bloomberg's original reporting.

Google's role is the backstop: if Anthropic defaults on lease payments at any of the five data center locations — in New York, Texas, Louisiana, and Indiana — Google covers the gap. Broadcom provides an additional residual‑value guarantee on the $31 billion senior tranche, meaning if Anthropic defaults and the used TPUs can't be resold for enough to cover the loan, Broadcom absorbs the shortfall. The guarantee effectively aligns the senior debt with Broadcom's investment‑grade credit rating rather than Anthropic's,.¹

This isn't Google's first swing at Anthropic. The search giant was one of Anthropic's earliest investors and has repeatedly bought equity in the AI firm, Bloomberg noted. A separate Google‑Anthropic cloud deal, announced in October 2025, is worth tens of billions and adds over a gigawatt of additional compute capacity, CNBC reported.

Google also planned to invest up to $40 billion in Anthropic through a combination of cash and performance‑linked targets — making it one of the largest corporate investments in an AI startup. The $35 billion chip backstop is layered on top of all of this, creating a relationship where Google is simultaneously Anthropic's investor, cloud provider, chip vendor, and now its lease guarantor.

The entanglement runs deep enough that it blurs the line between partner and competitor. Google's own Gemini models compete directly with Anthropic's Claude. Yet Google's cloud business benefits enormously from Anthropic's compute spend — Google Cloud's operating margin of roughly 20.7% means the Anthropic relationship could generate billions in high‑margin revenue, Investing.com reported.

Anthropic CFO Krishna Rao described the motivation in unusually direct terms: "This funding will help us serve the historic demand we are experiencing, stay at the research frontier, and bring Claude to more of the places where work happens,",¹ according to AI Weekly.

The emphasis on serving existing demand is significant. Most AI infrastructure deals are pitched as bets on future growth. Rao is framing this one as catch‑up — the compute is already spoken for by current customers. That changes the risk profile for lenders: the revenue is booked, not projected.

The deal also keeps Anthropic's equity story clean. By structuring the chip financing as off‑balance‑sheet debt — separate from the company's concurrent $65 billion Series H equity raise — Anthropic avoids diluting existing shareholders while still securing the compute it needs to scale. It's a financial engineering move that lets the startup have its cake and eat it too: billions in new infrastructure without the balance‑sheet weight.

For the developers building on Claude, this deal is a signal about reliability. When your API provider's compute is guaranteed by one of the world's largest companies, the chances of a capacity crunch affecting your application go down. Google's backstop means Anthropic's TPU supply isn't contingent on Anthropic's quarterly cash position.

But the lock‑in cuts both ways. Anthropic is now deeply tied to Google's custom silicon — TPUs, not the NVIDIA GPUs that most of the industry runs on. If Google's TPU roadmap falls behind, Anthropic's hands are tied by long‑term lease obligations. For builders, that means Claude's performance is increasingly a function of Google's chip strategy, not just Anthropic's model research.

The deal also reinforces a trend: AI compute is becoming a vertically integrated stack. Google makes the chips, finances the leases through Apollo and Blackstone, backstops the payments, and collects cloud revenue from the models running on them. The company that controls the silicon controls the economics for everyone downstream.

The structure has no historical precedent at this scale, and that creates blind spots. There is no large‑scale secondary market for used Google TPUs — ¹ that Broadcom's residual‑value guarantee may underestimate the illiquidity discount in a forced‑sale scenario.

Lease rates and pricing terms between Anthropic and the special‑purpose vehicle remain undisclosed. Broadcom's guarantee cap is also not public, so the full extent of its exposure in a severe AI hardware downturn is unknown.

And the concentration risk is real: Apollo and Blackstone investors face exposure to a single lessee (Anthropic), a single chip vendor (Google/Broadcom), and a nascent asset class with no historical default comparables. If Anthropic's revenue growth stalls — a scenario the company's own IPO filing suggests could happen before 2030 profitability — multiple layers of the financing structure get tested simultaneously.