By Joseph Adinolfi
Oracle’s decision to tap the debt markets has helped revive scrutiny of AI spending. Wall Street analysts highlighted key risks in reports published this week.
A growing number of strategists and hedge-fund founders seemed downbeat on the AI trade this week.
It might not seem like much based on the action in markets, but doubts about the durability of the artificial-intelligence trade were having a moment on Wall Street this week.
Several high-profile analysts published reports focused on potential risks to the AI narrative. Recent deals involving Oracle Corp. (ORCL) and OpenAI, as well as OpenAI and Nvidia Corp. (NVDA), revived concerns about circular financing in AI, as well as where money for these new data-center commitments would come from. For the first time, companies appeared willing to take on more debt to get in on the action: Just look at Oracle’s “unusual” jumbo bond deal from earlier this week.
Creeping doubts appeared to weigh on stocks, causing major indexes to finish the week just shy of record territory as several AI-related names came under pressure. Still, investors’ eagerness to buy on dips once again helped to limit losses, said Mark Hackett, chief market strategist at Nationwide.
At one point on Friday, the Nasdaq Composite COMP was on track to finish lower for a fourth straight day. That would have been its longest string of daily declines since April. The Nasdaq managed to finish the session in the green; however, it still snapped a three-week winning streak.
‘At some point, this has to translate to the bottom line’
Steve Sosnick, chief strategist at Interactive Brokers, discussed parallels between the AI spending boom and the rush to lay fiber-optic cable that fueled the telecom bubble in the late 1990s. That previous episode ended in a handful of high-profile bankruptcies and scandals.
In commentary published earlier this week, Sosnick highlighted what some have described as the dangerously insular nature of AI-related spending. The hyperscalers – Microsoft Corp. (MSFT), Amazon.com Inc. (AMZN), Meta Platforms Inc. (META) and Alphabet Inc. (GOOGL) – are Nvidia Corp.’s biggest customers. Nvidia itself is a huge customer of CoreWeave Inc. (CRWV) and Super Micro Computer Inc. (SMCI).
Super Micro’s shares have been on a roller-coaster run over the past two years and have yet to return to their record peak from early 2024. Meanwhile, CoreWeave has seen the wave of demand that followed its initial public offering fade, as its shares have retreated from their June peak, FactSet data showed.
While the internet-era bandwidth expansion eventually proved profitable, the money didn’t arrive in time to save companies like Global Crossing, which filed for bankruptcy protection in 2002, Sosnick noted.
“We’ve seen immense returns generated from the cash that has been invested, not necessarily in terms of bottom-line results at companies, but the market has seen the benefits,” Sosnick told MarketWatch in an interview. “But at some point, this has to translate into the bottom line.”
An arms race
Sosnick has not been the only one on Wall Street to raise questions about the relentless spending. J.P. Morgan Asset Management’s Michael Cembalest laid out the impact that the AI boom has had on U.S. markets and the economy in a report published on Wednesday.
Then he highlighted potential constraints for the AI build-out, like the aggressive need for power, which is already raising electricity prices for consumers in certain markets.
Since ChatGPT launched in November 2022, a group of 41 stocks associated with artificial intelligence have driven 75% of the S&P 500’s advance, according to Cembalest. They have accounted for 80% of the corporate-earnings growth and 90% of the growth in capital spending, he said.
But emerging financial risks also factored into the discussion. Until the Oracle-OpenAI deal, the AI spending boom had been mostly financed by a handful of companies like Microsoft, Meta, Alphabet and Amazon, which could afford to pay for it out of their cash flows. But by borrowing an immense sum in an unusually large bond issue this week, Oracle has broken this pattern – and in doing so has threatened to turn the AI data-center build-out into an arms race, with companies potentially rushing to borrow.
As for what might actually instigate a selloff, Cembalest said Chinese firms like Huawei and SMIC could succeed in building their own ecosystem for producing cutting-edge chips – loosening Nvidia’s grip on technological dominance. He sees this as the biggest risk facing the top-heavy U.S. equity market.
‘Kicking the tires on the bear case’
A team of equity analysts at Barclays said they remained optimistic about AI, but they appeared more willing to explore certain risks surrounding the theme in reports shared with clients.
The Barclays team pointed out that service-provider revenues have continued to grow at a healthy clip. Mentions of AI on earnings calls of S&P 1,500 firms have also risen, while more AI-related job postings emerged during the first half of the year.
However, the scale and speed of the data-center build-out have made “kicking the tires on the bear case” more worthwhile, the Barclays team said.
What if power constraints make it more difficult, or even impossible, to continue building? What if AI models stop improving, or newfound efficiencies lead to excess data-center capacity? That was the big concern during January’s DeepSeek-inspired selloff, and it remains front and center today.
“The lingering fear, then as now, is whether AI is due its ‘dark fiber’ moment, in which exponential efficiencies lead to critical underutilization of infrastructure assets,” the Barclays team, led by Venu Krishna, head of U.S. equity-linked strategies, said in its report.
The term “dark-fiber moment” is a reference to the telecom bubble, which occurred alongside a bubble in dot-com stocks in the late 1990s. Expectations for soaring internet traffic fueled a spending boom on fiber-optic cable. But these companies quickly discovered that much of what they had built would lay dormant, at least initially.
Cisco Systems Inc. (CSCO), which rode this trend to briefly become the most valuable company in the world, has never seen its split-adjusted market value return to its early 2000 peak.
The Barclays team also explored the risk that debt-fueled spending could supercharge the build-out, raising the risk that investors could be left holding the bag if many data centers ultimately end up becoming, essentially, stranded assets. Cembalest raised a similar concern in his report.
Even some from the hedge-fund world got in on the action. Citadel’s Ken Griffin told CNBC earlier this week that when it comes to AI, “there were obviously echoes of the dot-com bubble” in this moment. Greenlight Capital founder David Einhorn warned during a panel discussion that unchecked data-center spending could demolish enormous amounts of capital.
And in an article published in its Friday print edition, the Wall Street Journal questioned whether AI spending would ever pay off.
Beneath the surface, there are some signs that the AI trade is beginning to stall out. Nvidia, which designs the chips needed to train AI models, has seen its shares go nowhere for almost two months. On Friday, Nvidia finished at $178.19, on par with prices from early August, FactSet data showed. Although Nvidia was on track to finish the week with a slight gain, other Big Tech and semiconductor names, all closely associated with the AI trade, were set to decline.
On the technical side, chart readers could see signs that the AI trade was starting to look a bit stretched earlier this week. The Global X Artificial Intelligence & Technology ETF AIQ was the most overbought since its inception on Monday, when its 14-day relative-strength index hit 82.41. The only other time its RSI has been above 80 was June 15, 2023, when it peaked at 80.80, according to FactSet. The RSI is a popular momentum gauge used by market technicians in their analyses.
Doubts around AI aren’t entirely new. In August, none other than OpenAI CEO Sam Altman referred to the AI craze as a “bubble.” Some on Wall Street, most notably Jim Covello at Goldman Sachs Group (GS), expressed many of these same concerns more than a year ago. But Oracle’s decision to tap the debt markets to help finance its deal with OpenAI helped revive scrutiny this week, Sosnick said.
-Joseph Adinolfi
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
09-27-25 0919ET
Copyright (c) 2025 Dow Jones & Company, Inc.