Ever since OpenAI released ChatGPT in late 2022, Wall Street has compiled a long and ever-growing list of signals that the AI boom might be on the verge of collapse — and each time, investors have been left asking whether the end was finally at hand.

When AI chip giant Nvidia's market capitalization crossed the $2 trillion, $3 trillion, $4 trillion, and $5 trillion milestones, bears sounded the alarm at every turn. The borrowing spree by AI hyperscalers — the giant cloud operators anchoring the buildout — and their astronomical capital spending plans raised further red flags. In early 2025, the DeepSeek shock set off another panic, as the Chinese AI model called into question the need for all of that capex.

Also last year, OpenAI CEO Sam Altman's admission that an AI bubble exists and investor Michael Burry's bearish bets added to the unease. This year, the Iran war, the meltdown in the South Korean stock market, and "circular" funding deals among AI companies — arrangements in which firms simultaneously invest in and buy from one another — deepened the fears.

Yet earnings kept soaring, and stocks eventually rebounded to set record high after record high.

Of course, traders' persistent willingness to keep doubling down on AI could itself be a hallmark of a bubble. Many on Wall Street concede the good times will come to an end at some point; the only question is how long to ride the AI bets without leaving money on the table.

Now, though, there is a new reason to wonder whether it is time to cash out.

The current wave of AI anxiety centers on whether trillions of dollars in AI spending will pay off sufficiently — and whether the strain will drag down debt-laden companies such as CoreWeave. The neocloud computing firm, which rents out AI computing capacity from data centers packed with graphics-processing chips, recently watched its stock tumble while its credit default swaps — contracts whose prices track the perceived risk of a default on its debt — surged.

For Dan Greenhaus, strategist at Solus Alternative Asset Management, CoreWeave is not a canary in the coal mine signaling that an AI bust is imminent.

"How many canaries have we killed in the coal mine over the last three years as everybody has latched on to every new sign of weakness so to speak as evidence that the bull market was about to end only to see the broad market, both tech and otherwise, go on to make new highs?" he told CNBC on Friday. "So I don't know that I would read any more into what's happening now than I would have or did any of the other headlines."

To be sure, borrowing costs have jumped as the Iran war and oil prices sent bond yields to their highest levels in more than 20 years. The Federal Reserve is also hiking rates again to rein in stubborn inflation. That has drawn sharper scrutiny to the riskier corners of the AI boom, which has grown increasingly reliant on issuing debt via the bond market.

Greenhaus noted, however, that CoreWeave's unsecured debt has been especially targeted, calling the company "the riskier of the riskier AI buildout play." Meanwhile, earnings from AI chip suppliers like Taiwan Semiconductor indicate the broader demand remains robust.

At the same time, he does not expect companies across the AI ecosystem to be able to keep financing themselves indefinitely as rates move higher.

"I don't know whether this is the breaking point or another 25 or 50 basis points higher is the breaking point," Greenhaus explained. "But as rates move higher, the more speculative tails so to speak of the buildout are going to be called into question, and right now that's CoreWeave."

For its part, CoreWeave CEO Mike Intrator said markets can "misunderstand and act irrationally," and told employees in a memo that the company's deal backlog is more than enough to sustain its debt obligations.

Nvidia CEO Jensen Huang also came to CoreWeave's defense — the chipmaker counts itself as a shareholder in the company — saying it does not need a financial backstop.

"CoreWeave's strength is improving every day," he told Bloomberg. "I'm optimistic. Their cost of capital is actually decreasing."

Others on Wall Street also see strong demand for AI and view the recent selloff as yet another buying opportunity, arguing the fundamental AI story and its drivers remain intact.

Still, even eye-popping revenue projections may not be enough to underwrite sky-high spending. On Wednesday, Gartner warned that AI data center owners need to generate a cumulative $8.5 trillion in revenue through 2030 to cover capital spending, financing costs, and a 12% return.

But total enterprise and consumer AI revenue headed to data centers will amount to at most $3.6 trillion, leaving more than $4 trillion in asset value exposed to write-down risk — the forced reduction of an asset's book value — through 2032.

On Friday, analysts at Bank of America said in a note that the AI investment boom is following a classic capital cycle, which usually ends in disappointment "even though the underlying technology often lived up to expectations."

Typically, early profit margins tend to be high, driving overinvestment and excess capacity while also attracting more competition. In AI's case, competition has already ramped up, especially from Chinese open-weight models. On top of that, business spending on AI has started to dip, as higher token volumes fail to offset lower prices for tokens.

"This monetization disappointment in the face of intense competition is precisely the failure mode seen in past capital cycles," BofA added.

For now, the debate comes down to numbers like these — and to how far the Federal Reserve ultimately takes its rate hikes — as Wall Street weighs whether the latest warning is another head fake or, this time, the real thing.

This story was originally featured on Fortune.com.