Apollo and Blackstone Lock In $35B Chip Deal to Fuel Anthropic's AI Expansion

Apollo Global Management and Blackstone have finalized a $35 billion debt financing package to buy Google's custom tensor processing units (TPUs) for Anthropic, Investing.com reported, citing Bloomberg. The transaction ranks among the largest private credit deals ever structured and signals that AI infrastructure financing has entered a new phase — one where Wall Street's biggest credit engines are directly underwriting chip purchases.

The structure uses a special‑purpose vehicle (SPV) that borrows the funds, receives an equity investment, buys the TPUs, and leases them to Anthropic. Anthropic plans to deploy the chips at data centers across New York, Texas, Louisiana, and Indiana. The lease payments are expected to service the debt, keeping the obligation off Anthropic's balance sheet entirely, according to Quartz.

The financing is divided into three tranches with different risk profiles. Roughly $6 billion of A1 notes carry investment‑grade private ratings and were sold to banks at a spread of just 1 percentage point over Treasuries. Another $24 billion in A2 notes priced with a 5.75% coupon. Both senior tranches benefit from a residual value support agreement from Broadcom, the chipmaker that co‑develops Google's TPUs.

A separate $4.5 billion B tranche, which does not have Broadcom's backstop, priced at an 8.5% coupon, Investing.com reported. Roughly half of the total debt was syndicated to outside investors.

The Broadcom backstop is the deal's keystone. If Anthropic defaults on lease payments and the chips are resold for less than the amount owed, Broadcom covers the shortfall for A1 and A2 noteholders. This effectively transfers credit risk from a high‑growth AI startup to a $500 billion semiconductor company, making the senior notes far more palatable to institutional investors. Broadcom CEO Hock Tan confirmed the arrangement, Investing.com reported.

This transaction represents a shift in how AI infrastructure gets built. Instead of tech companies raising equity or issuing their own debt to buy chips, a new financial layer is emerging — private credit funds underwrite the hardware, chipmakers backstop the risk, and AI labs lease what they need without loading up their balance sheets.

Broadcom CEO Hock Tan said earlier this week that the AI XPV platform is targeting more than 20 gigawatts of compute capacity through 2028, according to Investing.com. That's roughly equivalent to the power consumption of a small country — and signals that the AI buildout is nowhere near its peak.

The timing is notable. Anthropic recently completed a $65 billion funding round that valued the company at $965 billion — surpassing OpenAI's valuation for the first time — and filed confidentially for a U.S. IPO. The chip financing deal locks in compute capacity ahead of what could be a landmark public offering.

Anthropic's rapid climb to a $965 billion valuation, confirmed by GuruFocus via Yahoo Finance, reflects an AI market where compute is the primary constraint on growth. Claude, Anthropic's flagship model family, competes directly with OpenAI's GPT line and Google's Gemini. Each generation requires exponentially more training compute, and serving millions of users demands inference capacity at scale.

The Google TPU selection is strategic. While most of the industry builds on NVIDIA GPUs, Anthropic has deepened its Google relationship — Google is both an investor and the TPU supplier. Custom silicon can offer better price‑performance for specific workloads, and owning the chip supply chain through this financing structure gives Anthropic predictable capacity at a known cost.

For builders, the infrastructure story matters because it shapes which models are available, at what latency, and at what price. When a company locks in $35 billion in dedicated chips, it's a signal that they plan to serve a lot of inference — and compete aggressively on both capability and cost.

The special‑purpose vehicle structure is increasingly the standard for AI chip financing. The SPV buys the hardware, takes on the debt, and leases the chips to the AI company. This keeps the debt off the AI lab's balance sheet — important for a company about to go public — while giving lenders recourse to the physical hardware if things go wrong.

The Broadcom backstop adds another innovation. Normally, chip financing carries technology obsolescence risk: TPUs bought today could be worth far less in three years when newer chips arrive. Broadcom's agreement to cover any residual value shortfall effectively insures investors against this depreciation — but only on the senior tranches. B‑note holders take that risk directly, which is why they demand an 8.5% coupon.

This model is likely to be replicated. If Apollo and Blackstone can syndicate $35 billion for Anthropic, similar structures will emerge for other AI labs. The financialization of AI compute is accelerating — and it's creating opportunities for builders who understand how the infrastructure layer actually works.