Sixty Billion Dollars of AI Debt Enters Syndication

--- title: Sixty Billion Dollars of AI Debt Enters Syndication slug: sixty-billion-dollars-of-ai-debt-enters-syndication meta_title: The $60B AI Chip Financing Is Now Being Sold meta_description: Banks began syndicating a $60 billion debt package to fund Anthropic's lease of Google TPUs, with Broadcom guaranteeing $42 billion of it. category: news tags: Broadcom,Anthropic,Google TPU,AI debt,syndication,Blackstone,chip financing,BofA,data centers,risk premium ---
Bank of America, Citigroup and Morgan Stanley began distributing a $60 billion debt package to other lenders this week, according to reporting from the Financial Times. The money funds Anthropic's lease of semiconductors developed by Google and co-designed with Broadcom. It is the largest chip financing transaction on record.
The structure matters more than the headline number, because it tells you who is carrying the risk and where it lands.

How the package splits
Roughly $42 billion in senior secured loans, backed by Broadcom, entered syndication first. Broadcom's A-minus credit rating lets lenders place that debt with a wider investor base, through private placements or the investment-grade bond market, which in turn lowers Anthropic's borrowing cost.
Another $18 billion in junior debt will follow, without Broadcom guarantees. Blackstone has already committed about $9 billion to that tranche and will participate in syndicating the rest.
The proceeds are earmarked for Anthropic's 2027 chip orders. Lease payments begin only after the chips are delivered, which keeps the cash flow aligned with the hardware rather than the contract signature.
The supplier became the lender
Anthropic's IPO prospectus shows Broadcom agreed to lend the lab up to $42 billion, and the instruments can convert into Anthropic shares. The tenants stack up: Broadcom co-designs the chips with Google, leases the equipment, and now provides the financing.
Anthropic has committed to a five-year, $125.2 billion lease for tensor processing unit capacity. Broadcom's convertible financing covers roughly a third of it. The partnership also gives Anthropic access to multi-gigawatt next-generation TPU capacity starting in 2027.
Broadcom projects AI semiconductor revenue of about $115 billion in fiscal 2027 and $230 billion in fiscal 2028. Anthropic is expected to become the largest customer in Broadcom's chip-design business next year.
The filing flags the obvious conflict: Broadcom's decisions on pricing and hardware availability could affect Anthropic's ability to get the compute it needs.
Where the risk concentrates
BofA Securities analyst Tom Curcuruto estimates Broadcom's AI financing vehicles could support around $370 billion in senior debt by mid-2029 at a 20-gigawatt scale, including roughly $150 billion of new issuance in 2027.
Broadcom takes on real exposure by backstopping lease obligations for some customers. Its first transaction puts it on the hook for as much as $29 billion.
Robert Leitao of Rothschild & Co. summarized the bet plainly: the financing assumes a very small number of companies will generate enough revenue to support it. Seaport's Jay Goldberg noted that Nvidia has already used its balance sheet to pull chip demand forward, and Broadcom is following the same playbook.
The earlier Broadcom transaction provides the template. A $35 billion financing with Apollo and Blackstone backed its 20-gigawatt "AI XPV" platform, aimed at helping companies including Anthropic and OpenAI secure computing capacity. The current deal is the same structure at roughly double the size, which suggests the constraint on AI infrastructure is increasingly the ability to borrow against it rather than the ability to build it.
There is a reasonable defense of the arrangement. An order that cannot be financed is not an order, and lease payments that begin only after delivery align the cash flow with the hardware. The concern is concentration: if the financing rests on a small number of customers generating enough revenue, then a slowdown at any one of them propagates through a lender that is also a supplier.
Why the risk premium matters
Investors have spent recent months demanding higher risk premiums on loans to companies spending trillions on advanced models, on the theory that the capital outlay may not convert to profit on a useful timeline. This syndication is the first large test of whether those demands actually change terms.
The mechanics of the two tranches tell you how the market is being asked to price that doubt. The senior portion carries Broadcom's credit rating, which means lenders are underwriting the chip designer's balance sheet as much as the AI lab's business. The junior portion carries no guarantee and is where the real risk question lives. Blackstone's roughly $9 billion commitment to that tranche is an early answer, and the spread on the remaining $9 billion will be the more informative number.
Broadcom's stock moved up 4.11 percent to $377.40 on the reporting day, which suggests the market read the syndication as progress rather than as a warning. The arrangement also shows how a supplier benefits from pulling demand forward: the loan makes an order possible that the customer could not otherwise finance on its own.
The signal for everyone else
The immediate market reaction was positive. Broadcom shares traded up 4.11 percent at $377.40 at the time of reporting.
The structural lesson is different. Vendor, lessor, lender and possible shareholder can now be the same firm, which means AI infrastructure contracts should be read the way companies read related-party financing. Three questions apply to any of them: who gets paid if the project slips, who can accelerate the debt, and whether your own cloud or chip commitments sit inside a similar circle.
Broadcom's quarterly report also notes Anthropic could issue up to $42 billion in convertible notes to Broadcom to fund lease payments. Cash was deposited into a restricted account for Broadcom's benefit in April 2026, and certain defaults could accelerate a large share of the lease while limiting use of the facility to cover it.
None of this makes the compute unnecessary. It does mean the cost of a GPU-hour is now partly a credit question. Investors have been asking for higher risk premiums on AI lending, and this syndication is the first large test of whether they get them.
What a lease-backed market means for buyers
There is a second-order effect that procurement teams should watch. If chip access increasingly depends on lease structures backed by a supplier's balance sheet, then the terms available to a smaller buyer will look different from the terms available to a lab the size of Anthropic. A company without investment-grade credit cannot borrow against future compute the same way.
The Treasury angle is the same story from the other side. When a hardware vendor guarantees a customer's lease obligations, the vendor's own risk profile becomes a factor in the customer's access to financing. That is a normal pattern in aircraft and shipping finance, and it is new to AI hardware.
For the industry, the practical question is whether the model converts demand into deployments at the pace the financial projections assume. Broadcom's revenue guidance for fiscal 2027 and 2028 depends on it. So does the ability to keep selling capacity that has not yet been built.
Leitao's framing stays the most useful summary: the whole structure rests on a small number of companies producing enough revenue to service it. That is a bet about product adoption as much as about credit, and the syndication is how the market prices it.
Related articles
The Memory Chip Is Now the Bottleneck in AI Compute
The logic roadmap is public and predictable. The memory roadmap is gated by packaging yields, a years-long talent base and three companies' capital plans.
AI Music Got a License. Read What It Actually Covers.
Cheap tracks still exist. What is disappearing is the assumption that a track generated from a prompt is free of claims, which was never true and is now demonstrably false.
Hengdian's Studios Are Empty While Asia's AI Film Pipeline Runs
Hengdian's empty stages are the evidence that the industry has already placed its bet, whether or not the audience agrees.
Google Bought 890 Megawatts of Nuclear to Feed Its Data Centers
The compute is available. The electricity is the thing that has to be arranged, years in advance, the way a supply chain is arranged.