The deal on the table
Anthropic is offering Blackstone a yield of more than 10 per cent to help finance its latest purchase of artificial-intelligence chips, the Financial Times reported on Friday. According to the FT, funds managed by Blackstone would invest $9 billion in debt backed by Anthropic's five-year lease payments on the chips. Reuters, which carried the FT's story, said it could not immediately verify the details — and neither company has publicly confirmed the talks.
The structure described is familiar from the industry's recent fundraising playbook. Rather than buying chips outright, an AI lab leases them from a special-purpose vehicle — a separate corporate entity that owns the hardware and carries the debt. The vehicle's bonds are then repaid from the lab's promised lease payments, while the chips themselves serve as collateral.
This would not be Anthropic's first trip to the private-credit counter. Earlier this year the company arranged tens of billions of dollars in financing with Apollo and Blackstone to fund computing capacity, and the new talks appear to be a larger sequel — reportedly funding roughly one-and-a-half gigawatts of Google's tensor processing units, the custom AI chips that have become one of Nvidia's main rivals in the data centre.
What is striking is the price. A double-digit yield on a technology deal in 2026 tells a story well beyond one company's balance sheet.
Why Anthropic is raising debt for chips
AI laboratories live or die by compute. Training and serving large language models requires vast clusters of specialised processors, and demand for those chips has outstripped what even the best-funded companies can buy with cash on hand. Valuations in the tens of billions do not translate into spare billions for hardware — especially when the hardware depreciates almost as fast as it is installed.
Debt backed by lease payments is the industry's workaround. The lease payments match the cost of the chips to the revenues they generate: Anthropic pays for the compute as it uses it, lenders collect a contracted stream, and the hardware sits as security behind the loans. Done well, it lets a fast-growing lab scale its fleet without diluting shareholders or pausing its training runs.
It is also a matter of ambition. Anthropic's leadership has discussed building tens of gigawatts of computing capacity within a few years — a target that implies hardware spending on a scale equity markets alone cannot comfortably fund. Borrowing against contracted payments is, in theory, the cheaper and faster route.
But the model has a weak point, and it is the same one now haunting the whole AI infrastructure market: the collateral is chips that lose value quickly, and repayment depends on Anthropic's revenues growing fast enough to cover the lease. If AI revenues disappoint, lenders are left holding hardware worth far less than the loans secured against it.
What a 10 per cent yield really says about AI credit risk
In credit markets, yield is a price tag on risk. A borrower judged safe pays little; a borrower judged shaky pays a lot. A yield above 10 per cent — in the neighbourhood of what high-yield or distressed borrowers pay — is the market's verdict that lending to Anthropic, against five years of lease payments, is not a safe loan.
Some of that reflects the deal's structure. Blackstone's $9 billion is reportedly earmarked for the junior tranche: the slice of the debt that absorbs losses first if things go wrong. Junior lenders are paid after senior lenders, so they always demand extra compensation. The senior slice, paid first and cushioned by support tied to the hardware's residual value, costs far less. The gap between the two is the market's price for Anthropic-specific risk.
And that risk is real. When Anthropic closed its earlier financing, its credit profile was described as sitting one notch below investment grade — respectable for a fast-growing technology company, but not the kind of rating that commands cheap money. A double-digit junior yield maps neatly onto that profile: lenders want to be paid for the chance that the AI boom's revenues do not arrive on schedule.
The pricing also fits a broader pattern. Even mighty Meta has had to offer more than 7 per cent a year on recent borrowing, while riskier cloud data-centre specialists have gone above 9 per cent, according to market analysis of the AI borrowing binge. Against that yardstick, 10 per cent-plus for a junior tranche of Anthropic chip debt is not an outlier — it is the going rate for AI risk in late 2026.
A market under strain: the 'Darwin moment'
The Blackstone talks are landing in a credit market that has turned hostile to AI borrowers. Just this week, the FT reported that investors are dumping bonds issued to fund some of the largest US data-centre projects, pushing debt backing facilities such as 'Hyperion' and 'Project Excalibur' — where Meta, Nvidia and Oracle are expected tenants — well below par.
The pain is sharpest among the so-called neoclouds, the specialised cloud providers — Nvidia-backed CoreWeave is the best known — that have been the most aggressive in financing the AI build-out with borrowed money. 'Data centre bonds are having their Darwin moment, where the weakest credits struggle to survive,' Ludovic Subran, chief investment officer at Germany's Allianz, told the FT, adding that widening credit spreads after power and construction delays showed data-centre debt had been priced too tightly for the execution risk.
The stress is feeding back into equities. Nvidia, which has invested tens of billions of dollars across neoclouds and AI start-ups, shed more than $300 billion in market value over the past week — a reminder that even the AI trade's biggest winners are exposed to its financing chain.
Zoom out and the scale of the borrowing is extraordinary. From next to nothing in 2024, technology-sector borrowing has reached roughly $500 billion in the nine months since January, as Google, Meta, Amazon, Microsoft and their peers raise debt to pay for chips, servers and power. AI now accounts for an estimated quarter of all corporate bond issuance, up from 4 per cent two years ago. In inflation-adjusted terms, the AI sector is on course to borrow more this year than America's cable operators did to build the internet — or its railroads did in the 19th century.
What happens next
The negotiations are not final, and terms could still shift. But the direction of travel is clear: AI companies are becoming some of the biggest borrowers on the planet, and credit markets are starting to charge them like it.
For Anthropic, closing the Blackstone deal would unlock the next chunk of its compute build-out — and set a benchmark yield for every AI lab that follows it to the debt market. For Blackstone, a double-digit coupon on debt backed by five years of payments from one of the AI race's frontrunners is the kind of trade private credit was built for: high reward, if the revenues arrive.
The risk is what the bond market is already signalling elsewhere. If AI revenues grow more slowly than the debt piles assumed, today's junior tranches could become tomorrow's restructurings — and the 'Darwin moment' will have found its casualties. Investors watching the AI trade may find useful context on our markets page, where we track the sector's swings.
For now, the message from the FT's reporting is simple: the AI boom's money is getting more expensive. The open question is whether the returns will be big enough to cover it.
FAQs
What exactly is Anthropic offering Blackstone?
According to the Financial Times, Anthropic is offering Blackstone-managed funds a yield of more than 10 per cent on a $9 billion investment in the junior tranche of debt tied to its latest chip purchase. The debt would be backed by Anthropic's lease payments on the chips over five years. The deal is still being negotiated and has not been confirmed by either company, and Reuters said it could not independently verify the report.
Why is Anthropic borrowing instead of raising more equity?
Scale is the main reason: the company's compute ambitions imply hardware spending far beyond what equity rounds can comfortably fund, and borrowing avoids diluting existing shareholders. Leases also match payments to usage, and a special-purpose vehicle keeps the debt off Anthropic's own balance sheet. With AI-linked equities volatile — Nvidia lost over $300 billion in market value in a single week — debt can look like the steadier route, provided lenders keep showing up.
What is a junior tranche, and why does it pay more?
Debt deals are sliced into layers. Senior lenders are paid first if the borrower runs into trouble, so they accept lower yields. Junior lenders sit below them and absorb the first losses, which is why Blackstone's slice commands a double-digit return. In this deal the senior layer is additionally cushioned by support tied to the chips' residual value — effectively a backstop linked to what the hardware would be worth second-hand.
What does the 'Darwin moment' mean for ordinary investors?
It means credit markets are repricing AI risk in real time: bonds funding big data centres have fallen below par and neocloud debt is trading as stressed. That matters beyond Wall Street because pension funds and bond index trackers hold this corporate debt, and because financing stress can delay data-centre build-outs — which in turn hits the chipmakers and cloud stocks many portfolios own.
Could the Blackstone deal fall through?
It is possible — the talks are described as ongoing and terms may move. But Anthropic has precedent on its side, having already closed a large private-credit financing earlier this year with Apollo and Blackstone. If this round failed, the company would need to find the money elsewhere at similar or higher cost, since lenders across the market are now charging AI borrowers a premium.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by FT. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
