AI Vendor Financing: Broadcom Backs a $60B Anthropic Deal, Eyes $50B for OpenAI

The companies selling AI chips are now arranging the loans their customers use to buy them. On Monday 5 October 2026 Bank of America, Citigroup and Morgan Stanley began syndicating what the Financial Times called "a record $60 billion debt package" for Anthropic's chips, with $42 billion of it backed by Broadcom, Investing.com reported. Two days later the Wall Street Journal reported that Broadcom was arranging more than $50 billion of financing for the custom chips it is building with OpenAI, per Yahoo Finance. Around the same time, Reuters reported that lenders were asking Nvidia for stronger guarantees before backing its $500 billion customer-financing plan. This is the vendor financing the IMF and the Bank of England had just described as a stability risk.

What happened

Anthropic. The $60 billion package funds Anthropic's lease of Google chips due for delivery in 2027. Lease payments begin only after the hardware is fully deployed. It has two layers. The first is $42 billion of senior secured loans resting on Broadcom's A-minus credit rating. The second is $18 billion of junior debt with no Broadcom guarantee, which leaves lenders directly exposed to Anthropic's credit. That tranche may wait until after Anthropic's planned IPO. Blackstone committed about $9 billion and is helping to place the rest. This builds on an earlier structure that Forkast dates to June: Reuters reported that Broadcom backstopped more than 80% of a $35 billion financing for Anthropic's compute.

OpenAI. According to the WSJ report of 7 October, Broadcom approached Apollo and Blackstone about financing chips under OpenAI's Nexus programme, part of a plan to deploy 10 gigawatts of custom accelerators by the end of 2029. The talks were early, and nobody confirmed them. Fortune, citing Bloomberg, put the figure at about $30 billion rather than $50 billion, as Forkast noted. The same report said Oracle was in talks with Apollo and Goldman Sachs on a chip financing deal of its own.

Nvidia. On 10 August Nvidia signed memorandums of understanding with Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs and KKR to mobilise more than $500 billion of third-party capital for its customers, CNBC reported. In early October, Reuters reported (as republished by CTech) that banks and credit managers wanted more than Nvidia had offered. Nvidia had said some deals could carry a residual-value guarantee of no more than 25%. Lenders wanted guarantees on every deal, or revenue backing from investment-grade customers, plus higher rates and bigger cushions.

The numbers

Why it matters

Vendor financing is not new, and it is not automatically a bad sign. Aircraft and telecom equipment have been sold this way for decades. What changes the risk is who carries it. When the supplier guarantees the debt, the supplier's sales and its credit exposure become the same bet. Broadcom's senior tranche is priced off Broadcom's rating, not Anthropic's. The IMF's 13 October blog described exactly this channel: stress at the company receiving the money cuts its revenue forecasts and valuation, and that weakens the debt tied to the deal. The full background is in my write-up of the IMF and Bank of England warnings.

The argument with lenders is really about how long a GPU lasts. Banks typically underwrite GPUs over three to four years, Reuters reported, while Nvidia says its top GPUs can earn revenue for about a decade. "The creditor community does not subscribe to long average lives for these assets," said Brian Gelfand, co-head of global credit at TCW. Broadcom's own chief executive, Hock Tan, said in mid-October that the investor rush into anything AI looked like "casino gambling", Reuters reported.

What sceptics say

The defence of these structures is that they are secured, priced and increasingly conservative. Nvidia's spokesperson told Reuters that "AI compute is a productive, durable and fungible asset that can support long-term financing." The $8.5 billion CoreWeave loan got an investment-grade rating because Meta's contracted payments stand behind it, not because of the chips alone. The pushback from lenders that Reuters described is itself evidence that discipline exists. Anthropic's junior tranche is being sold on Anthropic's own credit, and underwriters may wait for the IPO to price it. That is how a healthy market behaves.

My take, as opinion: the structures look sensible one deal at a time. What worries me is the total. A few suppliers are arranging financing for a handful of labs whose revenue commitments to each other add up to trillions. If one lab misses its numbers, the stress shows up in several suppliers' guarantees at once.

What it means for engineering and DevOps teams

• Know which lab your stack really depends on. If your product runs on one frontier model, your supply chain now includes that lab's debt. A second model provider behind a gateway is a cheap hedge.
• Read GPU contracts like a lender. Who owns the hardware, what happens if the lessor is refinanced or sold, and how quickly can you move your workloads and data? The answers matter more when the hardware sits in a special-purpose vehicle.
• Use the depreciation debate in your own models. If you buy GPUs, plan on the three to four year life the banks assume, not a decade. Teams running self-hosted inference should do the break-even maths on the shorter life.

FAQ

What is AI vendor financing?

It is when a chip or infrastructure supplier helps arrange, guarantee or provide the debt its customers use to buy its products. Broadcom backing most of Anthropic's $60 billion chip-lease package and Nvidia's $500 billion financing platform are the largest examples in 2026.

Is the Broadcom and OpenAI financing confirmed?

No. As of the October reports, the talks were described as early. The WSJ reported more than $50 billion and Bloomberg about $30 billion, and neither company has confirmed a deal.

Why are lenders cautious about GPU-backed loans?

There is little history on how long top GPUs keep their value. Banks underwrite them over three to four years, while Nvidia argues for up to a decade. Lenders want guarantees or investment-grade customer contracts to cover the gap.

Related: GPU Clouds in 2026: H100s at $2.59 an Hour, $104B of CoreWeave Backlog on $35B of Debt, and a Six-Year Bet on Silicon, Seats Are Dying. What the 2026 SaaS Repricing Looks Like From the Compute Layer and Self-Hosting LLMs vs API Costs in 2026: The 62% Utilisation Break-Even.