ECONOMY STUDIOIn the week of October 12, 2026, 20 of the most valuable AI and chip stocks lost more than $1 trillion in value1. On October 14, a hotter-than-expected inflation report sent the Nasdaq Composite down 1.88%2. Investors worried that higher borrowing costs could weigh on the huge spending needed to build AI infrastructure2.
If you own an S&P 500 index fund, this is your money too3. Nvidia is about 8% of the index, about $8 of every $100 in the fund, and chipmakers have contributed about 70% of the index’s gains in 20263.
So is this a boom, or money going round in a circle? Nvidia’s own filings help answer it.
The demand is real
A bubble is not simply a product nobody uses. ChatGPT has more than 900 million weekly users4. OpenAI started 2026 with $20 billion of annualized revenue and told investors almost $50 billion for September, less than the roughly $70 billion run rate it had signaled earlier5.
Even a skeptic agrees: Broadcom’s chief executive Hock Tan, who likened some AI investors to gamblers in a casino, said OpenAI and Anthropic have shown they can produce meaningful revenue6. Nvidia’s revenue was $96.2 billion in the quarter ended July 26, 2026, up 106%7. For every $100 of revenue a year earlier, it now takes in about $2068.
The loop in Nvidia’s own filing
Nvidia sells its chips to AI clouds, companies that rent out computing power. In its 10-Q, it says it commits to cloud service agreements with them, $36 billion of commitments, typically six years long9. Nvidia also owns about 10% of CoreWeave, one of those clouds10, and in 2025 it agreed to buy CoreWeave’s unsold capacity, an order with an initial value of $6.3 billion, through April 203211. It has guaranteed up to $3.5 billion of AI clouds’ data center lease obligations12.
Then there is OpenAI. Nvidia agreed to invest $30 billion in OpenAI’s funding round13. OpenAI rents computing power from the clouds, the clouds buy Nvidia chips, and Nvidia rents some of that capacity back9. Follow a dollar around, and part of each sale started as Nvidia’s own money.
Nvidia’s stakes in private companies were $47.9 billion on July 26, 2026, against $3.8 billion a year earlier14.
When those stakes rise in value, Nvidia books a gain: $23.7 billion in the six months to July, against $2.1 billion a year earlier12.
What the BIS found
The Bank for International Settlements, the central banks’ own bank, calls such ties circular15. Across deals from 2021 to 2025, $55.20 of every $100 invested in AI firms came from other AI firms, by deal value15. Of the deal value between AI firms, 46.4% involved firms that were also supplier and customer15. And 73% of the circular investment relationships come from firms that supply computing power or infrastructure15.
In U.S. business generally, a supplier owns a stake in its customer in about 3.3% of relationships; in AI computing and cloud infrastructure, 15.2% of customer-supplier relationships also have financing ties16.
The BIS gives two reasons. Suppliers can see their customers’ growth sooner than other lenders can, and customers use the money to secure scarce inputs17. It also names the catch: part of the supplier’s revenue growth reflects its own investment, not organic demand, and an investor that is also a supplier is exposed to the same counterparty twice17.
Who pays until the customers do
In 2025, OpenAI took in $13.1 billion and burned $8 billion of cash4. That is about $161 spent for every $100 of revenue, so about $61 of it came from investors and lenders rather than customers18.
OpenAI told investors it aims to spend about $600 billion on computing power by 2030, and it projects more than $280 billion of revenue that year4. The spending comes first; the revenue is a forecast.
That is where Ohio comes in. In August 2026, Nvidia agreed to guarantees capped at $105 billion for a data center campus in Pike County, Ohio, where an OpenAI affiliate leases the buildout for 20 years; in exchange, the site will exclusively host Nvidia AI infrastructure19. The guarantees end if OpenAI reaches a satisfactory credit rating19. CNBC had reported talks of a backstop of up to $250 billion before it was signed at $105 billion, and Nvidia is also investing $1.5 billion in the builder20.
Why last week
Money got more expensive. September consumer prices were up 3.7% on a year earlier21. The 10-year Treasury yield first closed above 5% on September 16 and reached 5.31% on October 522.
And the boom runs more and more on debt. Goldman Sachs estimates that AI-related financing is nearly one-quarter of gross U.S. investment-grade bond issuance23. Two days before the report, Hock Tan said capital, not chips, is now the biggest constraint on AI6.
The weakest link moved first. CoreWeave has about $50 billion of debt and has become a focus of worries that the data center boom is fueled by credit10. After comments from Nvidia’s chief executive, who said he does not see a need to backstop CoreWeave, its shares closed 5.94% higher on October 151,10.
Nvidia’s answer is that this is no loop at all: in his words, AI compute is revenue, so a data center can be financed like a toll road or power plant that produces cash flows24. Analyst Dan Ives, who still calls demand robust, put it as “rates are a financing question, not a demand question”2.
We have seen this loop before
In the late 1990s, equipment makers such as Lucent and Nortel financed network operators so the operators could buy their equipment. When the operators’ revenue failed to appear, the makers lost both the loans and the sales25.
The Nasdaq peaked at 5,048.62 in March 2000 and was down 77.9% by October 200226.
There is a difference that cuts both ways. The fibre laid in that boom lasted decades, while the newest AI chips have a shelf life of only a few years27.
What to watch
None of this says what to do. It says what analysts watch: cuts to Big Tech’s spending plans, canceled contracts, and falling prices to rent AI chips are the three warning signs of real trouble2. In August, Nvidia teamed up with six of Wall Street’s biggest firms to raise more than $500 billion for AI infrastructure, mostly from insurance and retirement capital23.
So, is it a bubble? A bubble is not about whether people use the product. It is about who pays until the customers pay enough. Of every $100 invested in AI firms, $55.20 came from other AI firms15. The boom is real, and part of the bill is being paid by the people selling it.
