Why Circular Deals in the Artificial Intelligence Boom Make Investors Nervous

Technology stocks just had their worst week since June. Chipmakers led the decline as investors questioned how long companies can keep borrowing to build data centers for artificial intelligence while bond yields sit near their highest levels in about two decades, the Associated Press reported. Behind that worry is a feature of the boom that has drawn growing attention: money that moves in circles.

The pattern works like this. A company that sells the essential hardware, most prominently Nvidia, invests in the businesses that buy its chips. Those customers use the cash, along with borrowed money, to build data centers filled with the same hardware. The supplier books the sales, and its investment rises in value as the customer grows. In an analysis published Oct. 13 in The Conversation, the finance researchers Ruipeng Liu and Sagarika Mishra described Nvidia acting as investor, supplier and financier at once, with stakes in model developers such as OpenAI and Anthropic and in cloud providers such as CoreWeave, which rents out computing power built on Nvidia chips.

The authors cited a central bank analysis finding that more than half of investment in artificial intelligence firms came from other artificial intelligence firms, and noted that chipmakers have supplied a large share of the stock market’s gains this year. When so much money circulates among a handful of companies, trouble at one can spread quickly to the others. CoreWeave’s shares fell 15.5 percent on Oct. 14, during last week’s selloff, a drop the market research firm Trefis tied to concerns about the company’s debt.

History offers a warning. In the telecommunications boom of the late 1990s, equipment makers lent money to customers so they could buy more equipment. Far more fiber-optic cable was laid than the market needed, and investors lost heavily when the bubble burst. Liu and Mishra point out one difference that cuts the wrong way: that fiber stayed useful for decades, while advanced chips become outdated within a few years.

There is a strong case on the other side. The authors themselves write that none of this means the boom is fake. Demand for computing power is real, the companies involved earn large revenues, and circular investment can speed up construction that will be valuable if customers keep paying for artificial intelligence services. The question is timing: whether revenue arrives fast enough to cover the spending and the debt taken on to finance it.

That is the same question that moved markets last week. The AP reported that investors are worried companies that rely on debt markets may not be able to keep borrowing at the same pace with yields so high, and the Federal Reserve’s next moves will shape how expensive that borrowing becomes.

For ordinary savers, the stakes are not abstract. Retirement accounts and index funds hold large positions in the biggest technology companies, so a sharp repricing would reach well beyond Silicon Valley. The coming round of quarterly earnings reports will offer the next test of whether the spending is starting to pay off.