A step-by-step guide to the AI bubble

Is it bursting? Here's how to tell, and what to do about it

Tech stocks have swooned in recent weeks. A reprieve seemed nigh on November 19, when chip giant Nvidia reported blowout earnings, demonstrating once again that there’s real revenue and profit in the artificial intelligence revolution.

But Nvidia isn’t the whole story, and a low-speed tech selloff still seems to be underway. Some seasoned investors are sounding the alarm on AI because they see warning signs that remind them of other bursting bubbles. One is irrational exuberance over a disruptive technology that might be getting ahead of itself. Another is the role of private lending to finance much of the spending, which is more opaque and possibly riskier than traditional financing.

The upshot could be a market downturn similar to the dot-com bust that began in 2000 and lasted three years. Investors overexposed to AI or to tech in general could endure sharp losses they never saw coming.

Here’s the lay of the land:

¶ Hundreds of firms across the economy are spending more than $100 billion per year on artificial intelligence systems, including power-hungry data centers, computing hardware and cloud infrastructure. This spending surge has almost single-handedly driven the stock-market boom of the last three years.

¶ Much of the borrowing that finances this buildout comes from private sources other than banks, such as private-equity firms, hedge funds and specialty finance companies. Some familiar Wall Street firms now have private credit arms, as well. This private credit is less regulated and less transparent than more traditional financing through banks. Some analysts see parallels to the reckless lending that inflated the subprime mortgage market in the early 2000s, and left many investors with PTSD that’s flaring now.

¶ It would be normal for many of the AI bets these companies are placing to fail. Breakthrough technologies often generate speculative manias in which a small number of players become stratospheric successes while many go bust.

¶ There’s sharp disagreement over whether AI will transform life as we know it or just amount to a glorified Google search. The market narrative, so far, has tilted toward transformation. But some prominent voices are now calling AI vastly overhyped.

¶ Stocks and tech stocks in particular are at extremely high valuations. A correction would be normal, but a selloff could deepen if it exposes a lot of undercapitalized firms that can’t survive routine market stress. Some bigwigs think we’re already starting to see that.

Here are the possible elements of a bursting AI bubble:

1. A souring AI narrative. The AI hype suggests bots and machines will soon be able to reason like humans and do much of the work people can do. But there are skeptics. Banking analyst Chris Whalen calls AI a “global marketing con.” He wrote recently that “most public companies are so invested in the false gospel of AI that they dare not even hint at the truth, namely that the vast majority of AI projects will never be profitable or even relevant.”

Yann LeCun, an AI progenitor who has been chief AI scientist at Meta, has been drawing converts to his unconventional view that the standard approach to training AI is a dead-end that will never yield anything near human-level intelligence. LeCun recently said he’s leaving Meta to start a firm developing a different type of AI.

2. Stretched valuations. Current stock values are extremely high, relative to earnings. Economist David Rosenberg of Rosenberg Research points out that by one measure, S&P 500 valuations are higher than during any bubble of the last century except for the peak of the dot-com boom in 1999. “We are in the stratosphere,” he says.

By Rosenberg’s estimation, stock prices hit bubble levels in June 2024. The typical bubble lasts 16 months before it bursts. Last month would have marked the 16-month point. For what it’s worth, the tech-laden NASDAQ index hit its last record high on October 29. It’s down nearly 8% since then. 🤔

3. Dodgy financing. There’s been about $500 billion in AI investment in the United States during the last 10 years. The pace of investment now is about $110 billion per year. Much of that is borrowed money, and much of that comes from private sources that face little regulation and don’t have to disclose their obligations the way banks do. Private lending is still a small portion of all lending, but it probably accounts for a much bigger chunk of riskier gambits on AI.

There’s nothing inherently wrong with private credit. But insiders say this is where the froth is forming, with private lenders seeking outsized returns by gambling on AI ventures that may never be profitable. Billionaire Jeff Gundlach, CEO of DoubleLine Capital, said recently that “in recent years, garbage lending has gone to these private markets.” Public markets, by contrast, have much better lending standards than they did during the subprime mortgage explosion, which led to the financial crash of 2008.

4. Cockroaches. Most people haven’t noticed, but a few firms have gone bust recently in surprisingly dramatic fashion, including subprime auto lender Tricolor Holdings, auto-parts company Frist Brands and Renovo Home Partners. Those firms relied on private credit for their financing, and they essentially collapsed with almost no warning. That suggests terrible oversight by their lenders, and possibly fraud.

“My antenna goes up when things like that happen,” JPMorgan Chase CEO Jamie Dimon said in October. “When you see one cockroach, there are probably more.” These failures were not AI companies, but many analysts point out that small failures in the early days of the subprime collapse in 2006 heralded much worse things to come.

5. Dot-com scale. What’s happening in AI right now is not on the scale of the subprime wipeout that led to the 2008 financial crash. That crisis metastasized throughout the banking system and cost Americans as much as $20 trillion in lost wealth, along with 9 million jobs. We’re not heading for another full-blown financial crisis.

An AI bust could resemble the dot-com bust, however, which is hardly reassuring. That extended selloff caused around $5 trillion in investor losses, which would be about $9 trillion today—roughly two Nvidias. Of course, the dot-com boom that preceded the bust did bring us Amazon, Google, Nvidia itself and many other titans of today’s digital economy. So you could argue that investor losses from 2000 to 2002 turned into gains for those who held on. Huge gains, for those who held long enough.

6. The AI defenders. Some important players think there’s no AI bubble to burst. Red-hot Nvidia is the biggest beneficiary of AI spending, and unlike many of those dot-com darlings of 25 years ago, it has gargantuan profits. Net income for the current year is likely to hit $114 billion, according to Capital IQ, with an eye-popping profit margin of 54%. “There’s been a lot of talk about an AI bubble,” CEO Jensen Huang said after the company reported quarterly earnings on November 19. “From our vantage point we see something very different.”

7. What should ordinary investors do? Certainly not panic. Gundlach says investors should lower their stock allocation to 40% of their portfolio, and most of that should be outside the United States. But that’s not realistic for many ordinary investors, who have to book capital gains and pay taxes if they sell assets to move money around. For anybody worried about a bubble, it’s more realistic to allocate new investments to safer assets. Investors should also be sure they know the risk of any newfangled products—such as private-credit vehicles—they may have heard about in financial media or from a broker.

Perspective matters, too. David Kotok, cofounder of Cumberland Advisors, suggests we’re in the “middle phase” of a cycle very similar to other transformative technologies. “The sequence in all of history is the same,” he says. “In the early phase, there’s rapid acceleration. The middle phase includes leverage. Too much leverage and that’s where we will see the pain.”

Eventually, he says, “we will have applications that will be extraordinary, and worthy of investment.” Right now, we just aren’t sure which ones, or how long it will take.