How to Know When the AI Boom Is About to Go Bust

Almost from the start of the artificial-intelligence boom, a steady chorus of bears has warned that it’s destined to end in tears—like the original dot-com bubble more than a quarter-century ago and the housing bubble of the mid-2000s.

Nearly four years after the public launch of ChatGPT, the AI boom is still going, the bulls are still dancing and investors are listening for any sign that the music is about to stop playing.

The list of things that could go wrong is long. There isn’t enough power for all the data centers under contract. There might not be enough paying customers to support all the data centers. The biggest AI-model developers could themselves be disrupted by upstarts. Rising interest rates could make borrowing costs prohibitive. And, most spectacularly, rogue AI agents might kill us all.

Yet new capital continues to pour into the AI trade and, as obvious as this point might be, that is where its biggest vulnerability lies.

Booms need continuous flows of fresh capital to perpetuate themselves. History shows they can start turning into busts the moment the window to raise new capital closes. Tipping points can happen suddenly, or as a slow-motion capitulation. But we’re not there yet.

Here is a signal to watch for in the meantime: See if the term “funding gap” returns to the market’s vernacular. Along with “burn rate,” funding gap became part of the financial lexicon in 2000, as the dot-com bubble was starting to deflate.

Back then, burn rate measured how much cash a money-losing startup was blowing through each quarter and how many months it had left before it ran out. It was popularized by a famous Barron’s article in March 2000 that helped prick the bubble. A funding gap meant a company still needed investors to keep supplying fresh cash until it became self-sustaining—and that the chance to raise new money had already vanished.

One poster child was Pets.com. In early 2000, Ernst & Young gave the company a clean audit opinion with no warning about its ability to continue as a going concern. That was in part because it seemed able to keep raising cash from investors no matter how much money it lost. The online retailer sold products for $5 that cost $10 to buy and $50 to market. The Nasdaq was reaching new highs.

Then the market crashed, the money ran out, and the company shut down in November 2000.

Similar concepts made a comeback during the financial crisis that began in 2007, except this time with large financial institutions. A key moment arrived in November 2007, when Freddie Mac slashed its dividend and sold $6 billion of preferred stock in a desperate bid to bolster its regulatory capital amid mounting mortgage losses.

What Freddie really needed was to raise billions of dollars in common equity, which absorbs losses more effectively than preferred equity does. If even a government-sponsored enterprise couldn’t do it, that signaled it would be difficult for almost any financial-services company to raise fresh common equity to restore its capital cushion.

Some equity offerings still got done in the months after that. But those weren’t enough to offset the torrent of losses. By September 2008, Fannie Mae and Freddie were placed into conservatorship, and the government was bailing out the country’s financial system.

Back to the present, the AI trade still has plenty of momentum, but early signs of nervousness have surfaced. OpenAI ruled out going public this year. Holtec Nuclear, which aims to supply small modular reactors to AI hyperscalers, this month paused its IPO plans. Anthropic, the Claude maker, pushed its planned IPO from October to November.

While those may be bumps in the road, they aren’t signaling that the capital window is closing, or at least not yet. Even if the companies don’t go public anytime soon, they still appear able to raise money in private markets. Holtec, for one, is already profitable, and has other established nuclear businesses.

But it’s hard to get a read on a company such as OpenAI or Anthropic, because their finances aren’t public and may not be for some time. All anybody on the outside seems to know is they are burning through lots of cash, with an insatiable need for more and no visible profits by conventional measures.

If OpenAI can rely on the likes of Nvidia NVDA 0.22%increase; up pointing triangle and SoftBank to keep funding it, that helps hold the AI ecosystem together in the hope that one day it can become self-sustaining. OpenAI’s ability to raise new money also means a company like Oracle ORCL -1.75%decrease; down pointing triangle, which is counting on OpenAI for hundreds of billions of dollars of future revenue, can continue to justify its own debt-fueled capital expenditures on data centers.

But if those circumstances should change, and the capital window narrows enough to create a funding gap for any of the world’s most important AI startups, the momentum behind the AI trade could quickly reverse. Once that happens, there is no telling how events may play out.

Chances are, the market will know that moment when it sees it.