Nvidia Is Financing Its Own Customers

A chipmaker sells $100 of processors to a customer. To help that customer afford them, the chipmaker guarantees the loan. The sale books as revenue. The guarantee sits off to one side as a contingent obligation.

Do that at the scale of hundreds of billions of dollars and you have the structure now sitting underneath the largest company in the market. Circular financing in artificial intelligence means a supplier funding, guaranteeing, or investing in the very customers who buy its products, and Nvidia has been doing it in sizes that are difficult to picture.

The company has been linked to roughly $750 billion in such arrangements. Credit default swaps on its bonds, which function as insurance against a default, recorded their sharpest intraday rise since they began actively trading.

This matters to you even if you have never bought a single share, because index funds made nearly everyone a shareholder. Below is how the structure works, why reasonable analysts disagree about it, what the credit market is signalling, and how to work out your own exposure.

How Circular Financing Actually Works

Strip the jargon and the mechanism is simple, which is what makes it easy to miss.

In an ordinary sale, a customer raises money from a bank or investors, decides your product is worth buying, and pays. That purchase is independent evidence that demand exists, because somebody unconnected to you assessed the risk and funded it.

In a circular arrangement, the supplier supplies the funding too. It may invest equity in the customer, guarantee the customer's borrowing, or finance the purchase directly. The customer buys the chips. Revenue is recognized.

What has not happened is an independent party judging that the purchase makes sense. The supplier has validated its own demand, and the risk that the customer cannot ultimately pay has moved from the customer's balance sheet onto the supplier's, where it sits as a contingent obligation rather than a line of cost.

Nvidia's reported commitments illustrate the scale: a plan under consideration to guarantee $250 billion for an OpenAI data center project in southern Ohio, a separate arrangement to finance OpenAI's purchase of $350 billion in chips, and an SK Group partnership involving more than $500 billion of mutual business.

Fortune covered the original Nvidia and OpenAI arrangement and the bubble comparisons it drew when the structure first became visible.

The Case That This Is Fine

Give the other side a fair hearing, because the sceptical reading is not automatically correct and plenty of serious analysts reject it.

Vendor financing is ordinary commercial practice. Aircraft manufacturers finance airlines. Equipment makers finance contractors. Doing so is not evidence of fraud, and the arrangement makes obvious sense when the supplier understands the customer's economics better than a bank would.

Jensen Huang's argument is that Nvidia does not need these deals to sell chips, because demand already exceeds supply. On that reading, the company is backstopping OpenAI because the firm with the clearest view of compute demand has concluded the capacity will generate returns sufficient to service the financing. If you have the best information in the market, acting on it is rational rather than reckless.

The economist Noah Smith has argued a version of this at length in his assessment of AI's circular deals, pointing out that the structure only becomes dangerous if the underlying demand turns out to be imaginary.

That is the actual question. Not whether the arrangement is unusual, but whether the end demand is real.

What the Credit Market Is Saying

Equity investors and credit investors are looking at the same company and reaching different conclusions, which is usually worth noticing.

Credit default swaps on Nvidia bonds posted their largest intraday increase since they began actively trading. A credit default swap is insurance against a borrower defaulting, so a rising price means the market is charging more to insure Nvidia's debt.

Why does that matter when the share price is holding? Because the two markets answer different questions. Equity holders are asking how much the company will earn. Credit holders are asking whether it can meet its obligations in a bad outcome. Guarantees are precisely the kind of obligation that does nothing in a good scenario and everything in a bad one.

Credit markets have a decent track record of noticing structural risk before equity markets do, because their payoff is asymmetric. A bondholder gains nothing from spectacular growth and loses everything in a default, so they pay closer attention to the downside.

This sits alongside the debt-funded capital spending we examined in how AI capital expenditure shifted onto borrowed money. Circular financing is a different mechanism from borrowing, but the two compound each other.

Working Out Your Own Exposure

Most people reading this own Nvidia without having chosen to, and the amount is larger than they would guess.

The S&P 500 is weighted by market capitalization, so the biggest company gets the biggest allocation automatically. Take a straightforward retirement portfolio:

• Balance in an S&P 500 index fund: $100,000
• Allocation to a single company at a 7% index weight: $7,000
• Add the other large AI-linked names at a combined 25% of the index: $25,000

If that block fell 30% while the rest of the index held flat, the portfolio drops about $7,500, or 7.5%, from one theme. Most people describing themselves as diversified would not expect a quarter of their money to move together.

Run your own version of that calculation rather than taking the example. Open your fund's holdings page, find the top ten weights, and add up the ones exposed to AI infrastructure. The number is usually a surprise, which is the point we made in why an index fund is less diversified than it looks.

The hardware supply chain transmits this in the other direction too, into the prices you pay for devices, as we covered in the memory chip shortage pushing up device costs.

Frequently Asked Questions

What is circular financing in AI?

It is an arrangement where a supplier funds, guarantees, or invests in the customers buying its products. The supplier books revenue from the sale while the risk that the customer cannot pay sits on the supplier's books as a contingent obligation rather than as an independent credit judgment by a bank.

How much has Nvidia committed to these deals?

Reporting has tied the company to roughly $750 billion in such arrangements, including a plan under consideration to guarantee $250 billion for an OpenAI data center in Ohio, financing for $350 billion of chip purchases, and an SK Group partnership exceeding $500 billion of mutual business.

Is vendor financing illegal or improper?

No. It is standard practice in aircraft, heavy equipment, and telecoms. The concern is one of scale and concentration rather than legality: the structure makes reported demand harder to verify independently, and it only becomes dangerous if end demand proves weaker than assumed.

What are credit default swaps telling us?

Swaps on Nvidia bonds recorded their sharpest intraday rise since active trading began, meaning insuring that debt got more expensive. Credit markets focus on downside scenarios, so they often register structural risk earlier than equity markets.

How exposed am I through an index fund?

More than most people expect. The S&P 500 weights by market value, so the largest AI-linked companies can collectively account for around a quarter of the index. Check your fund's top ten holdings and total the AI infrastructure names to get your real figure.

Judge the Demand, Not the Headline

Circular financing is not proof of a bubble and it is not nothing. It is a structure that makes demand harder to verify from the outside, concentrated in the largest company in the market at a scale with few precedents.

The question that decides the outcome is whether the compute being financed gets used profitably. If it does, these deals look like a supplier backing its own judgment with better information than anyone else had. If it does not, the guarantees convert from footnotes into losses.

You cannot answer that question. You can find out how much of your retirement account is riding on it, which takes about five minutes and is the only part of this within your control.

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Always consult with a qualified financial advisor before making investment decisions. Past performance is not indicative of future results.