Let’s talk about the supplier that lends its customer the money to buy its product, then books the signature as growth.
That sentence sounds like a scam. It isn’t. It’s a financing structure, it has a name, and it has already taken down one entire industry.
Vendor financing. The seller helps the buyer pay. In the late nineties it was Lucent, Nortel, Motorola — telecom equipment makers lending money to the little carriers who were buying their gear. Lucent alone was carrying something like eight billion dollars of customer financing. The equipment shipped, the revenue got booked, the quarter looked magnificent. Then the customers ran out of money, because the customers never really had money. They had a loan from the people selling them the thing. The revenue turned into a write-off, and the write-off turned into a decade nobody wants to talk about.
Now look at October 2026.
Broadcom has agreed to lend Anthropic up to forty-two billion dollars in convertible notes — notes that can turn into Anthropic shares — so Anthropic can pay Broadcom for Broadcom-designed chips. That’s roughly a third of a hundred and twenty-five billion dollar lease Anthropic has signed for those chips. And in Anthropic’s own draft prospectus, reported by Reuters on the first of October, the company warns that a supplier which is also its lender creates, quote, “potential conflicts of interest.”
That line is doing a lot of quiet work. It’s the legal way of saying: we know exactly what this looks like.
Nvidia is the original version of this. Thirty billion dollars deployed of a hundred billion committed into OpenAI. On top of that, a guarantee of a hundred and five billion dollars of lease payments, with maximum guarantee exposure disclosed at a hundred and eight and a half billion in its own filings. And the number I want you to sit with: ninety-nine billion dollars of equity on Nvidia’s books in the companies that buy its chips. Against ninety-six billion of revenue last quarter.
Read that again. The supplier owns more of its customers than it sells to them in a quarter.
Jensen Huang, at a conference in September, defending it: “It’s not circular because we put a little bit of money in, and a lot of money comes back.”
That is the whole thing, said out loud, by the man doing it, and nobody flinched.
Morgan Stanley is now counting more than three point one trillion dollars of guarantees, leases and other support that five hyperscalers, Nvidia and Broadcom have committed to other companies’ AI build-outs. Meta is guaranteeing about twenty-eight billion of value to the bondholders behind its Louisiana campus. Alphabet’s backstops on other people’s data-centre payments went from seventeen billion to forty-four billion in six months. The Bank of England’s Financial Policy Committee published a record in September warning that the rising indebtedness of AI firms, combined with opacity and circular arrangements, “could amplify losses if expectations disappoint.”
And the revenue underneath all of it is softer than the story. OpenAI’s run rate came in around fifty billion, not the sixty-eight billion that had been reported. Chip stocks fell the day that number got out.
Now — the obvious thing to do here is argue about whether it’s a bubble. Don’t.
That argument is a market-timing debate, and you are not a market timer. Nobody rings you before the repricing. The useful question is not “will it pop,” it’s “what structure am I standing inside, and what happens to my business when the terms change.” Because that one you can actually do something about.
Two things almost everyone is getting wrong.
The first: this is being covered as an investing story. It’s a procurement story. The price you pay for intelligence right now is not a market clearing price. It’s a market-share decision, funded by capital raised on a narrative, made by companies that need your usage more than they need your margin. Cheap tokens are not efficiency. They’re a subsidy with a clock on it — and the clock is always set by someone who has never met you.
The second: vendor financing looks like generosity at the point of sale and like ownership at the point of default. When your supplier is also your lender, they don’t just know your numbers. They know whether you can survive without them. That’s not a commercial relationship. That’s a position.
And here’s how it lands on a normal business, one that will never sign a billion-dollar lease.
You don’t get a letter explaining the repricing. You get a rate limit. You get a price change on a Tuesday with thirty days’ notice. You get a model retired with a migration guide written by someone who has never seen your workflow. You get a renewal with new terms and an account manager who is genuinely lovely and cannot help you. The migration cost always lands on the smallest customer, because the biggest ones had guarantees and indemnities written into their contracts by lawyers you can’t afford. You had a credit card and a good attitude.
That is how sovereignty gets lost, by the way. Not in a dramatic takeover. In a deal that looked generous while the subsidy was running, signed by someone who assumed the price was a fact.
So here are the three questions I’d put in front of any operator building on this stuff. Not a framework. Just the things I’d want answered before I bet payroll on a price.
Who funds this customer? If the money in your client’s account came from the same people who sold them the product, the growth you’re admiring is a position, not demand. That applies to your AI vendor. It also applies to your own pipeline, which is the uncomfortable part.
What happens to my business if this price triples in eighteen months? If the honest answer is “we’re fine,” you have a business. If the honest answer is “we’d have to rebuild the product,” you don’t have a business built on AI. You have a bet on someone else’s balance sheet.
And: what did I sign that I can’t carry out of the building? Not the subscription. The workflows, the prompts, the evaluation data, the record of why a decision got made. If the exit costs more than the entry saved, you didn’t buy a tool. You rented a dependency.
Then turn the same questions on your own deals, because this isn’t only a Big Tech disease. Every agency and every founder has felt the temptation to win a logo by stretching terms — invoice in ninety days, take a small equity slice, carry the build until their funding lands. It feels like hustle. Structurally it’s the same move at a smaller scale. If your client can only pay you with money you helped them raise, you’re not being paid. You’re being promised.
A discount is a price. A loan is a relationship where you now own their problem. Know which one you offered.
The principle worth keeping: a price is a promise about the future. When the people setting the price are financing the people paying it, that promise is being made with money that hasn’t been earned yet — and the bill, when it comes, gets sent to whoever has the least room to argue.
Build as if the subsidy ends. It will, and the notice will be short.