Anthropic Offers Double-Digit Yield for a $60 Billion AI-Chip Financing
A reported $60 billion debt package would turn Anthropic’s five-year chip leases into one of the largest tests yet of private credit’s appetite for AI infrastructure.
Anthropic is asking private-credit investors to accept a new kind of exposure to the artificial-intelligence boom. It is not equity in a model developer, but debt supported by the developer’s obligation to lease computing capacity for five years.
The Financial Times reported on 17 October that a financing code-named Project North Star could raise about $60 billion for roughly 1.5 gigawatts of computing built around Google-designed tensor processing units. Banks had begun syndicating the package earlier in October; the genuinely new information is the price Blackstone is being offered for the risk. The proposed structure includes an $18 billion junior tranche and about $42 billion of senior debt. Blackstone funds would provide roughly half of the junior capital—about $9 billion—in exchange for a total return above 10%, according to people cited by the FT.
Reuters separately reported the central terms but said it could not independently verify them. Anthropic and Blackstone have not publicly confirmed the package. The figures should therefore be treated as source-reported negotiations, not a completed financing.
Even with that qualification, the structure matters because it shows how AI infrastructure is migrating from corporate capital budgets into the machinery of securitisation and private credit. The junior debt would reportedly be backed by Anthropic’s five-year lease payments. The senior piece would sit against chips supplied through Broadcom and could later be refinanced in the institutional bond market under Rule 144A.
That arrangement divides the risk. Hardware and data-centre assets provide tangible collateral, but the cash flow ultimately depends on Anthropic continuing to need—and pay for—large amounts of compute. Investors are therefore underwriting both equipment value and the commercial durability of a model developer whose spending is growing faster than the history available to assess it.
The proposed yield signals that this is not being priced like ordinary investment-grade infrastructure. The FT compared the offer with a roughly 7% yield on the ICE BofA BB US High Yield Index. More than 10% on the junior tranche compensates investors for technology obsolescence, customer concentration, refinancing risk and the possibility that model economics change before the five-year leases expire. KBRA reportedly rates Anthropic one notch below investment grade, reinforcing the distinction between a high-profile borrower and a low-risk credit.
The senior and junior tranches do not carry identical exposure. Broadcom is expected to guarantee the $42 billion senior portion, while the junior investors would rely more directly on Anthropic’s credit and lease performance. That subordination makes the Blackstone commitment a particularly useful market signal. It reveals the price sophisticated investors demand when a large AI customer, rather than a highly rated chip supplier, carries more of the downside.
The comparison with Anthropic’s earlier Project Big Sky financing is instructive. That $4.4 billion deal was reportedly priced at about 8.5%. North Star is more than ten times larger and offers a higher return, suggesting that scale is increasing rather than eliminating the risk premium. It also arrives while listed AI-infrastructure borrowers have faced sharper scrutiny over leverage, contracted revenue and the useful life of computing assets.
For Blackstone, the proposal would extend a strategy already built around data centres, power and digital infrastructure. For Anthropic, debt can preserve equity ownership while matching payments to contracted computing use. But it also creates fixed obligations that remain even if model prices fall, usage shifts to different hardware or revenue growth disappoints.
The package could influence how banks, private-credit funds and insurers assess the next wave of AI projects. If investors accept the economics, other model companies may be able to finance chips through special-purpose structures instead of carrying the full cost on their own balance sheets. If demand is weak or the transaction is repriced, it would be a warning that capital providers are reaching their limit after several years of increasingly ambitious AI commitments.
Why it matters
AI investment is no longer only a technology story. It is becoming a credit-market story in which future model revenue, chip values and long leases are used to support tens of billions of dollars of borrowing. Project North Star would be a defining test of whether private markets can distribute that risk without obscuring it.
The immediate beneficiaries would be Anthropic, its compute suppliers and the infrastructure owners that receive the capital. The exposed stakeholders would include private-credit investors, insurers and any future bondholders asked to refinance the senior debt. The unresolved questions are basic but consequential. They include whether the financing closes at the reported size, how much protection investors receive, who bears hardware residual-value risk and whether the leases remain economic through the full term.
Sources: Financial Times · Reuters