Anthropic pays 10% to borrow $60B for AI chips

Anthropic pays 10% to borrow $60B for AI chips

Even the most valuable AI startup in the world is paying junk-bond yields to finance its computing habit. On October 16, 2026 (UTC), the Financial Times reported that Anthropic is offering Blackstone a yield above 10% to anchor a $60 billion debt financing package — codenamed "Project North Star" — that will fund the purchase of roughly 1.5 gigawatts of Google-designed TPU chips. If completed, it would be the largest chip-financing deal in history, and a stark signal that the AI infrastructure boom is running headfirst into a credit crunch.

The deal comes exactly one week after CoreWeave's bonds crashed to 15% yields and its stock lost a quarter of its value in five days. The message from the credit markets is unambiguous: AI companies, no matter how prestigious, are going to pay dearly for the capital that powers their expansion.

Inside Project North Star

The $60 billion package is split into two tranches, and the structure tells you everything about who is taking the real risk.

The senior tranche gets its investment-grade rating from Broadcom, which co-develops Google's TPUs and has agreed to guarantee the residual value of the chips. That means if Anthropic defaults, Broadcom is on the hook for $42 billion worth of hardware. The junior tranche — where Blackstone is putting $9 billion — has no such protection. It rests entirely on Anthropic's ability to make five years of lease payments.

A 10% coupon may sound reasonable for a speculative-grade borrower, but context matters. The ICE BofA BB US High Yield Index currently yields roughly 7% for comparably rated companies. Anthropic is paying at least 300 basis points above that benchmark. And four months ago, in a similar transaction called "Project Big Sky," Anthropic raised $4.4 billion in junior debt at 8.5%. The jump from 8.5% to 10%+ in just four months — on a deal 13 times larger — tells you which direction the market is moving.

Why it matters: the AI credit bubble has a price tag

Project North Star is not just an Anthropic story. It is the clearest evidence yet that the entire AI industry's expansion is being financed on increasingly expensive terms, and that the cost of capital is now a competitive weapon — or a liability.

Consider the arithmetic. Anthropic is borrowing $60 billion at a blended cost of roughly 8–9% (weighted between the senior and junior tranches). On $60 billion, that's $4.8–5.4 billion in annual interest expense alone. For context, Anthropic's entire Q3 2026 revenue was $16.8 billion. The company is now spending roughly a third of its quarterly revenue — every quarter, forever, until the debt is repaid — just to service the interest on this one financing package.

This is the part nobody talks about when they quote AI revenue growth. The headline numbers — $16.8 billion in Q3, up 45% quarter-over-quarter — look spectacular. But they don't account for the fact that Anthropic has signed $125.2 billion in TPU lease commitments over the next five years, and is financing those commitments at double-digit yields. The revenue is real. So is the debt.

The comparison with CoreWeave is illuminating. CoreWeave's 2032 bonds trade at 15% yields and its CDS implies a 60%+ five-year default probability. Anthropic's junior debt prices at 10%+ with a BB+ rating from KBRA. The gap between the two — roughly 500 basis points — is the market's way of saying: Anthropic is a better credit than CoreWeave, but it is not an investment-grade credit. Not yet, anyway.

Blackstone is betting that changes. The firm has told its investors that Project North Star is a rare opportunity to earn double-digit yields on Anthropic debt before the company's expected IPO drives its rating to investment grade and its borrowing costs down. That's a reasonable thesis — if Anthropic can maintain its revenue growth, avoid a major safety incident, and navigate an IPO in a market that just wiped $2 trillion off semiconductor stocks in eight days.

The Broadcom problem nobody is discussing

Buried in the structure is a conflict of interest that deserves more attention. Broadcom is simultaneously:

1. The chip supplier — co-developing Google's TPUs and selling them into the deal
2. The lender — guaranteeing $42 billion in senior debt
3. The lessor — Anthropic leases the chips through Broadcom, which means Broadcom sets the pricing
4. Potential equity holder — Reuters reported on October 1, 2026 (UTC) that the $42 billion loan could convert into Anthropic equity

A company that controls the supply, the financing, and the pricing of the product it is financing has every incentive to keep the cycle going — even if the economics stop making sense for the borrower. This is not unique to Broadcom (Nvidia plays a similar role with CoreWeave), but it means the "market price" for AI compute is not being set by an arm's-length negotiation. It's being set by vertically integrated financing structures that blur the line between customer, supplier, and creditor.

What to watch

The next several weeks will determine whether Project North Star is a one-off or the start of a broader repricing of AI credit.

1. Junior tranche distribution — Blackstone took $9 billion of the $18 billion junior piece. If the remaining $9 billion struggles to find buyers at 10%, expect the yield to rise further. That would be the clearest signal that the AI debt window is closing.
2. Anthropic IPO timing — the company was targeting a November roadshow at a $1.5 trillion valuation (down from $2 trillion). If the IPO slips to 2027, Anthropic will need to keep financing at junk yields for longer.
3. Broadcom's exposure — $42 billion in guarantees is a massive contingent liability for a company with a roughly $700 billion market cap. If Anthropic's revenue growth slows, Broadcom's guarantee gets tested.
4. Competitor financing costs — OpenAI, xAI, and Mistral all need similar-scale chip financing. If they're forced to pay 10%+, the AI capex cycle slows across the board.
5. The 144A market — the debt becomes tradable in 2027. If it trades below par in secondary markets, that's the market telling us 10% wasn't enough compensation for the risk.

The AI revolution isn't running out of demand. It's running out of cheap capital. Project North Star, at 10% yields on $60 billion, is the price tag for that shift. And if the most valuable AI startup in the world is paying junk-bond yields, nobody in this industry gets to pretend financing is free anymore.