Nvidia remains one of the world's most profitable companies and is not facing an immediate debt problem. Its role is changing, however. The company is no longer only selling chips. It is offering credit support to firms that buy them, guaranteeing some partner obligations and discussing much larger arrangements. The cost of insuring Nvidia's debt has consequently reached a record. What is confirmed, what is still only a proposal and why should shareholders care?
Nvidia is not close to insolvency. It produces enormous profits, holds substantial liquid assets and its bonds still carry very low default risk.
The market is asking a different question. How much risk is Nvidia taking from customers that need its chips but do not have enough capital to build data centers themselves?
When a supplier helps a customer obtain financing, it can sell more products. If the customer succeeds, both sides benefit. If the customer later struggles, the supplier may lose future orders and also suffer losses on financing or guarantees.
That new type of exposure is beginning to appear in the price of credit protection on Nvidia.
Futuro Invest previously described circular transactions in Nvidia: Jak se dělá iluze růstu. Nvidia invests in companies that subsequently spend money on Nvidia chips. That is not proof of fraud. It can, however, make it harder to separate natural demand from demand supported by the supplier's own capital.
Three important developments have followed:
This article is therefore an update, not a repeat. Financial support is moving from equity investments into loans, guarantees and long-term obligations.
At the beginning of July, Nvidia announced a new business model for smaller AI cloud operators. Partners build infrastructure using Nvidia chips and rent capacity to other customers. Nvidia helps through credit support.
Nvidia receives ordinary hardware revenue and a share of the revenue generated by the supported cloud capacity.
In simple terms, Nvidia does not merely sell expensive chips and walk away. It helps a customer fund the purchase and then participates in what the customer earns from renting the chips.
The benefit is clear. More data centers can be built and Nvidia can sell more systems. The drawback is equally clear. If servers are underused or the operator cannot pay, part of the problem can return to Nvidia.
In its latest quarterly filing, Nvidia said it had entered agreements to guarantee partners' facility lease obligations if those partners default.
Maximum gross exposure was $3.5 billion. Partners had placed $712 million in escrow, reducing Nvidia's potential loss. The guarantees run for five to seven years and Nvidia received warrants in return, giving it upside if the supported firms increase in value.
The amount is not existential for a company as profitable as Nvidia. The principle matters. Nvidia is officially becoming not only a supplier but also a guarantor and provider of credit support.
Nvidia also has a commercial paper program with capacity of up to $25 billion. It had nothing outstanding at the end of April. The program is not a problem by itself, but it gives Nvidia the ability to obtain short-term financing quickly.
The largest headlines concern reported talks with OpenAI.
According to The Wall Street Journal, Bloomberg and other media, Nvidia is considering a guarantee of up to $250 billion that could help finance a massive data center project in Ohio. OpenAI would lease computing capacity and Nvidia's backing could reassure lenders.
Separately, Nvidia is reportedly discussing financing as much as $350 billion of Nvidia chip purchases for OpenAI.
These numbers must not be presented as completed obligations. The discussions are reported, the terms are not public, the final amounts could be much lower and the transactions might never happen.
The distinction is crucial. The $3.5 billion of lease guarantees is a disclosed fact in Nvidia's filing. The $250 billion and $350 billion figures are media reports about possible future arrangements.
Imagine a manufacturer of construction equipment. A customer wants ten excavators but the bank will not lend enough. The manufacturer guarantees the loan, offers instalments or invests directly in the customer. The manufacturer can then sell the excavators and book revenue.
If the customer wins enough contracts, the structure works. If business does not arrive, the customer stops paying and the manufacturer may bear part of the loss.
The AI version is much larger. Chips, buildings, cooling and electricity cost tens or hundreds of billions of dollars. The customer is often not an established profitable firm, but a young cloud operator or AI laboratory that is growing quickly while consuming cash.
Vendor financing is not automatically bad. Investors should still ask whether sales reflect real end-customer demand or mainly exist because the supplier helped pay for the transaction.
Investor Michael Burry returned to the subject on July 23. He highlighted analysis by the Bank for International Settlements and warned about circular financing in the AI buildout.
A simplified circle may look like this:
Burry's argument is not proof that Nvidia's revenue is false. It is a warning that more transactions may be mutually dependent. If one weak link stops paying, the problem can reach the investor, chip supplier, data center owner and lender at the same time.
The BIS described a similar structure. A data center may be owned by a special-purpose company financed by private lenders. A hyperscaler holds a minority stake, signs a long lease or minimum capacity commitment, and sometimes provides a guarantee. Most debt is not reported on the hyperscaler's balance sheet, but the economic risk has not disappeared.
A credit default swap, or CDS, is protection against a company failing to repay debt. The buyer pays a regular premium and receives protection if the borrower defaults.
When CDS becomes more expensive, the market is demanding more compensation for the same protection. Investors perceive more risk or more uncertainty.
Nvidia's five-year CDS reached a record 82 basis points on July 27. One basis point is one hundredth of one percentage point. A spread of 82 basis points is roughly an annual premium of 0.82 percent of the protected amount.
In a simplified example, protecting $10 million of debt would cost about $82,000 per year. Actual CDS settlement is more complex, but the example shows the scale.
An 82-basis-point spread does not mean the market expects Nvidia to fail soon. CDS on genuinely distressed companies can trade in the hundreds or thousands of basis points.
The significance is that this was the highest level since the contract began active trading in November 2025. It rose 14 basis points in one day. The credit market quickly repriced new information.
Credit investors also look at different risks from ordinary shareholders. Equity investors focus on profit growth and chip prices. Bond investors ask what happens in a stress scenario and who absorbs losses on long guarantees.
Nvidia's earnings and share price can rise while credit protection also becomes more expensive. The two outcomes are not contradictory.
Nvidia is not the only AI-linked borrower under scrutiny. Oracle's five-year CDS reached about 215 basis points and S&P cut its rating to BBB minus, the lowest investment-grade level.
That is still not a distressed rating. It does mean investors demand much more to insure Oracle's debt than Nvidia's.
Oracle is investing heavily in data centers and signing long contracts with large AI customers. If servers are used and customers pay, debt can be productive. If projects are delayed or a major customer reduces spending, the fixed costs remain.
The message from the credit market is simple. Revenue growth is no longer enough. Debt, guarantees, leases and customers' ability to pay matter increasingly.
There are two opposing forces for shareholders.
On the positive side, Nvidia is using financial strength to expand its market. More customers can access chips, more data centers can be built, and Nvidia may receive both hardware sales and recurring revenue.
On the negative side, some customer risk moves back to Nvidia. If guarantees become large, Nvidia's value will depend not only on the number of chips sold but also on whether supported customers can pay for many years.
That can change the valuation investors assign to the stock. A high-margin manufacturer is valued differently from a business that also acts as an investor, lessor or guarantor. As financial obligations grow, investors will pay more attention to the balance sheet and less exclusively to revenue growth.
If this model expands, it can affect several markets at once.
First, lenders will demand higher interest from weaker data center operators. Without a strong partner's guarantee, financing may become too expensive.
Second, debt may move into project companies and private credit funds. It may not appear in the technology company's headline debt, even though that company has signed a long lease, minimum purchase agreement or guarantee.
Third, the chain becomes interconnected. An AI firm needs compute. A cloud operator needs a data center. A project company needs a loan. The lender wants a guarantee. Nvidia wants to sell chips. When demand grows, every link supports the next. When growth stops, the same links can accelerate losses.
This is not direct credit risk for US Treasuries. However, massive corporate issuance and private credit compete for investor capital. The more capital AI infrastructure needs, the higher the yield investors may demand from other corporate borrowers.
Favorable: AI demand grows faster than capacity. Supported data centers remain full, customers pay and Nvidia earns both sales and recurring revenue. Guarantees are rarely used and prove to be an efficient way to expand the market.
Middle path: AI continues to grow, but some projects have lower utilization. Nvidia must support some partners for longer, CDS remains wider and investors value its obligations more conservatively. Profit grows, but the stock's valuation multiple may fall.
Stress: Construction runs far ahead of demand. Weaker operators stop paying, guarantees are activated and some data centers remain empty. Nvidia faces weaker orders, investment losses and payments on assumed obligations at the same time.
No single headline or number will settle the argument. The combination of larger guarantees, more expensive credit protection and customers' real ability to pay will matter most.
Nvidia remains an exceptionally strong company. Its strength is precisely what allows it to support customers in ways that weaker manufacturers cannot.
That does not make every transaction risk-free. The more Nvidia helps finance purchases of its own products, the less clear the boundary becomes between natural demand and demand created by vendor support.
The record CDS spread is not evidence of an approaching collapse. It is a warning light on the dashboard. The credit market is telling investors to watch not only revenue growth, but also who is financing the AI race and who will pay if expected revenue does not arrive.
Note: This article is for informational purposes only and does not constitute investment advice. Investing in financial markets involves risk, and investors should conduct their own analysis before making any investment decision.