Apollo, Blackstone Pitch $36B Debt to Power Anthropic TPUs

Apollo Global Management and Blackstone are shopping a roughly $36 billion debt package to fund Anthropic’s purchase of Google tensor processing units (TPUs) for its next leg of model training, according to a report carried by Seeking Alpha citing Bloomberg. If completed at that size, the financing would be the largest single AI-infrastructure private credit deal struck to date — and it lands the same day Anthropic announced a separate $65 billion Series H equity round at a $965 billion post-money valuation.

The pairing of those two prints — $65B in equity plus a $36B asset-backed debt sleeve — captures the new shape of frontier-AI finance: equity is for the company, debt is for the chips. And the debt is increasingly coming from alternative-asset managers, not banks.

The deal at a glance

Item Detail
Borrower context Anthropic (private; $965B post-money valuation per its newsroom)
Lead arrangers Apollo Global Management, Blackstone
Size ~$36 billion (reported, not finalized)
Use of proceeds Purchase of Google TPUs for Anthropic’s training and inference infrastructure
Likely structure Asset-backed private credit (chips as collateral), syndicated across alternative-credit funds
Status In market; terms not officially disclosed by issuer or arrangers
Stock reaction (intraday) APO −2.12%; BX modestly lower
Source: Seeking Alpha citing Bloomberg report; Anthropic newsroom; intraday levels as of May 28, 2026.

Why private credit, not the banks

A $36 billion debt block of this risk profile — concentrated, single-name, and tied to AI hardware whose useful life is still being discovered — is precisely the kind of paper that no longer naturally fits a regulated bank balance sheet. Single-borrower concentration limits and capital charges for large technology-services exposures make this an expensive trade for the largest banks to warehouse on their own.

That gap is exactly what direct lenders were built for. Apollo reported $1.026 trillion of total AUM at the end of the first quarter, of which $834 billion sits in credit, per its investor relations disclosures. Blackstone’s credit-and-insurance platform and its real-assets group — the home for data-center and infrastructure financing — have been among the firm’s fastest-growing segments. Together, the two managers can underwrite a $36 billion deal, hold what they want for their own funds, and syndicate the rest into a deep bench of insurance-company partners hungry for higher-yielding, investment-grade-like paper.

Inside the Anthropic-Google plumbing

The financing has a clean operational logic. On May 18, 2026, Blackstone announced a joint venture with Google to launch a new TPU Cloud — purpose-built compute capacity sold to large AI customers under long-dated contracts. Anthropic has been one of Google Cloud’s anchor TPU tenants for years, and the new debt package looks designed to lock in chip supply for the next training cycle of Claude.

For Google, the structure is attractive: it monetizes TPU production at scale without taking direct customer credit risk. For Anthropic, debt-funded chip ownership is cheaper than indefinitely renting from a hyperscaler. And for Apollo and Blackstone, the assets — TPUs with defined depreciation schedules, attached to take-or-pay cloud contracts — look much more like aircraft leases or data-center financings than like venture debt. That is what makes them lendable at scale.

The macro backdrop: long-end yields near 5%

US Treasury yields, May 27, 2026 Bar chart of the 2-, 10-, and 30-year US Treasury constant-maturity yields as of May 27, 2026. US Treasury Yields — May 27, 2026 0% 2.5% 5% 4.00% 2-year 4.48% 10-year 5.01% 30-year Constant-maturity yields, Federal Reserve H.15
Source: Federal Reserve H.15, as of May 27, 2026.

The pricing backdrop is not benign. The 30-year Treasury closed Tuesday at 5.01%, and the 10-year at 4.48%, per the latest Federal Reserve H.15 release. Investment-grade infrastructure debt at those base rates plus a private-credit illiquidity premium suggests this paper likely prints in a high-single-digit to low-double-digit all-in yield — a level that pension funds and insurers will buy with both hands but that adds real cash-cost to Anthropic’s compute bill.

Stock-market read-through

Apollo shares fell about 2.1% intraday on the report, an unusual reaction for a manager that benefits from origination volume. The likely concern is single-name concentration: a $36 billion deal to one AI lab — even one valued at nearly $1 trillion — meaningfully raises Apollo’s exposure to the AI-capex cycle continuing. Blackstone traded lower in sympathy but with less drama, consistent with its smaller per-fund exposure and the diversification provided by its Google TPU JV announced earlier in May.

For public credit markets, the deal pulls a record-size piece of demand out of the syndicated loan and high-yield bond pipelines and routes it through private vehicles instead. Issuance calendars on the public side may stay lighter than they otherwise would, which has been a recurring tailwind for spreads on the bonds that do print.

What to watch next

Three signals will tell the market whether this deal is the start of a new pattern or a one-off:

  • Final size and pricing. If the deal clears at or above $36 billion with double-digit spreads to Treasuries, it sets a benchmark that every subsequent AI-infra financing will be priced against.
  • Collateral package. Disclosure on whether the chips themselves, the take-or-pay cloud contracts, or both serve as security will reveal how aggressively lenders are willing to price AI hardware depreciation risk.
  • Bank participation. Watch the Q2 earnings calls of the global investment banks for commentary on whether they sat out, financed in the warehouse, or syndicated alongside the alternative managers.

Either way, the May 28 print confirms the trend already in motion: AI’s biggest balance-sheet problem — funding the chips — is being solved in the private credit market, not on bank balance sheets.

Sources

Disclosure: This article was produced with AI assistance and reviewed before publication. It is for informational purposes only and is not investment advice.

Leave a Comment