The AI build-out has closed into a loop. The company that sells the chips is also, increasingly, financing the companies that buy them — directly, through equity and credit commitments, and indirectly, by backstopping the data-centre projects that create the demand. Nvidia has announced more than $540bn in such "circular financing" arrangements across 2026 (as reported; terms in several cases not final), and is separately reported to be in talks on a roughly $250bn backstop for OpenAI's Ohio data-centre campus and on a deal to help finance OpenAI's purchase of around $350bn in Nvidia chips. The pattern has now been named as a systemic risk by both the International Monetary Fund and the Bank for International Settlements — the two institutions whose job is precisely to see this class of problem early. The desk's read: the AI capability is real, but that is not the question the loop poses. The question is what the wider equity market and hyperscaler balance sheets look like if a meaningful share of the demand signal turns out to be the seller financing its own sales. We do not predict a crash — the crash thesis is speculation, and we label it so — but we do think the burden of proof has shifted onto whoever claims the demand is entirely organic.
Vendor financing is not fraud, and it is not new. A company that lends its customers the money to buy its product is doing something car dealers, aircraft makers and telecoms-equipment vendors have all done for a century. What makes the AI version worth a lead story is not the practice. It is the scale, the concentration, and the fact that the revenue it generates is being read by the entire market as evidence of independent, end-user demand — the one thing vendor financing cannot, by construction, prove.
1. The loop, mechanically
Strip away the deal names and the loop has four steps, and it turns in one direction.
Step one. Nvidia sells graphics processors to an AI company — say, a model lab or a cloud provider that rents compute to model labs. It books the sale as revenue. That revenue is the single most-watched number in the equity market, because it is treated as the thermometer of AI demand for everyone.
Step two. Nvidia commits capital to that same customer — or to the project the customer depends on. It takes an equity stake, extends or guarantees credit, or backstops the financing of the data centre the customer is building. The customer's ability to keep buying is now partly underwritten by the seller.
Step three. The customer uses that capital — or capital it raised from third parties on the strength of Nvidia's backing — to buy more Nvidia chips. The money that left Nvidia as an investment returns as revenue.
Step four. That revenue supports Nvidia's valuation and cash generation, which is what lets it fund the next customer. The loop closes, and turns again.
The critical feature is in step three: revenue that originates as the seller's own outlay is indistinguishable, on the income statement, from revenue that originates with an unrelated end-user who bought chips to run a profitable business. Both land in the same line. The accounting does not flag which dollar is which. Established That is why the loop is a measurement problem before it is a solvency problem: it degrades the reliability of the very number the market uses to judge whether AI demand is real.
2. The numbers, as reported
The scale is what moves this from a structural curiosity to a systemic question. Nvidia has announced more than $540bn in circular-financing arrangements over the course of 2026 — a figure that, as reported, excludes a potential further arrangement with OpenAI. Separately, it is reported to be in talks on a backstop of roughly $250bn for OpenAI's Ohio data-centre campus, and on a deal to help finance OpenAI's purchase of around $350bn in Nvidia chips. Assessed
Hold those numbers next to the thing they are supposed to describe. If a large share of the chip revenue the market is capitalising is being seeded by the chipmaker's own capital commitments, then the demand curve the whole complex is priced against is, to that extent, drawn by the seller. The desk is not asserting a specific fraction — that fraction is not disclosed and we will not invent it. The point is narrower and harder to dodge: at these magnitudes, the question of how much of AI demand is self-referential stops being rhetorical and becomes a number that matters to anyone holding the index.
3. Who is now saying it out loud
For most of the build-out, the circularity critique lived with short-sellers and sceptical columnists — a constituency easy to discount, because it is paid to be bearish. That is no longer where the warning is coming from.
The International Monetary Fund and the Bank for International Settlements have both flagged AI circular financing as a systemic downside risk. Established When the central bankers' central bank and the Fund name a private-sector financing structure as a stability concern, the analytical centre of gravity has moved. The claim no longer has to survive the objection "that's just the shorts talking."
The market-based tells point the same way. The cost of insuring Nvidia's debt against default — the price of its credit-default swaps — roughly doubled over two months, a move the investor Michael Burry highlighted as a signal that credit markets are repricing the linchpin's risk. Established Burry and the investor Mark Cuban issued paired warnings in late July; the broadcaster Jim Cramer likened the arrangements to dot-com-era vendor financing; and on 2 August 2026, NPR ran a mainstream treatment framed around a "$750bn" build-out that "critics call a bubble." Established
Here the desk draws a hard line the coverage often blurs. That the warnings were made, by these people, at this volume, is established fact. That the loop ends in a crash — a disorderly repricing that drags the wider market down — is not a fact. It is a forecast, and a contested one. Speculation
4. Second-order effects, which are the real story
The Navigator desk's standing instruction is to reach past the first-order question — is AI real? — to the second-order one: what does the loop do to the things attached to it? Three effects matter more than the headline.
The demand signal is degraded for everyone downstream. Hyperscalers set their own capital-expenditure budgets partly by reading the strength of AI demand off the market — off Nvidia's order book, off peers' commitments. If that signal is partly self-referential, then capex decisions across the largest balance sheets in the market are being calibrated against a thermometer the seller is warming with its own hand. Assessed This is how a measurement problem propagates into real spending.
Concentration turns a single-name risk into an index risk. Nvidia and a small cluster of hyperscalers account for an outsized share of major-index gains. When the linchpin is also the financier of its own customers, its idiosyncratic risk stops being idiosyncratic — it becomes the market's risk, because the market is the linchpin to an unusual degree. The doubling of Nvidia's CDS is the price of exactly that recognition beginning to form. Assessed
The failure mode is reflexive, not gradual. In an ordinary demand slump, orders soften and the vendor's revenue eases. In a loop, the same event runs in both directions at once: if end-user AI monetisation disappoints, the funded customers cannot pay, the seller's investments and receivables sour, and the revenue that supported the seller's valuation reverses — each leg pulling the others. That reflexivity is why institutions whose mandate is stability, not stock-picking, are paying attention. It is not that a downturn is certain. It is that, if one comes, the structure is built to amplify rather than absorb it. Assessed
None of this requires believing the AI capability is fake. It plainly is not. The models work, the compute is genuinely consumed, and some of the demand is unambiguously organic. The loop does not turn any of that off. What it does is make it impossible, from the outside, to tell the organic share from the seeded share using the one metric everyone relies on — and it wires the whole complex together so that a shock to any leg is transmitted to all of them.
Bottom line: The AI capability is real; the demand signal is compromised. When the seller of the shovels is also the banker to the diggers, the revenue line stops being clean evidence of independent demand — and at $540bn-plus, with the IMF and BIS now watching, that is a systemic question, not a valuation quibble. Watch the demand, not the price.
5. Prediction — logged in the Ledger
On or around Nvidia's next quarterly earnings release, scheduled for 26 August 2026, the disclosure of related-party and vendor-financing exposure will become a first-order theme of the report and the analyst call: the desk predicts that (a) at least one of the major reported arrangements — the OpenAI chip-purchase financing or the ~$250bn Ohio backstop — will be either confirmed, expanded, or explicitly addressed by management, and (b) the questions of customer concentration and self-financed demand will feature prominently in sell-side and press coverage of the print. This resolves as correct if the earnings materials or call transcript substantively address vendor/related-party financing and concentration; incorrect if the topic is absent or only glancingly mentioned.
Confidence: High (Assessed). The basis is that the circularity question has already reached the IMF, the BIS and mainstream press before the print, making it structurally impossible for management to avoid. The principal way this resolves wrong is a report so dominated by a headline revenue beat or miss that the financing structure is deferred — possible, but unlikely given the pre-print salience. Resolution date: Nvidia's fiscal-quarter earnings release and call, 26 August 2026, read against the accompanying financial statements and transcript.