What history tells us about an AI bubble

Manhattan-sized data centers set to power AI social video fluff. Soaring valuations for startups with TBD products. If you’re looking for anecdotal signs of an AI bubble, there are plenty.

In fact, three out of three historians that we asked agree: The economy is indeed in the grips of a major bubble around this technology. We called up these experts to better understand how explosions of investor speculation around the Next Big Thing tend to play out—why they happen, how they burst, and what comes after.

“History suggests that it’s extremely likely that, yes, we are in an AI bubble right now,” David Sicilia, associate professor of economic history at the University of Maryland, told us.

Tech bubble ingredients

While the dot-com boom might be the most reached-for analogy to the current moment, the history of business and technology is sprinkled with these sorts of tech-fueled investor frenzies, from railways in the 1840s to radio in the 1920s.

“It’s not inevitable that bubbles are going to develop around just any old new technology,” Lee Vinsel, associate professor of tech and society at Virginia Tech, said. “What seems more inevitable is seeing bubbles develop around technologies that are dubbed in the public understanding as ‘transformative’ from the get-go.”

The harder it is to gauge the potential impact of a technology, the more likely it is that markets won’t correctly price it. And the future of generative AI—which might be a near-superintelligence entity or essentially useless, depending on who you ask—is full of question marks.

Bubbles “arise from a poor estimate of the prospects of a venture. Some overshoot—and also some undershoot—is to be expected,” Andrew Odlyzko, a University of Minnesota mathematician who researches tech bubbles, told us. “That said, the general tendency of the human race is to be optimistic. So, generally speaking, people tend to err on the upside, which is why you have these bubbles that collapse.”

Not your mom’s dot-com

While the dot-com bubble had a clear rollercoaster arc—a steady build then a precipitous drop—the AI bubble burst will likely be messier, Sicilia said. There are many different lanes of investors hoping that GenAI adoption develops in distinct ways, he said.

“Because of the multifaceted nature of AI potential, I think that it’s not going to be probably as clean a pattern,” Sicilia said. “So this really could go on for quite a while, and you might have certain pieces of it that don’t pan out, or that turn out to be just not very promising, and then the enthusiasm evaporates, whereas others might have a longer time horizon.”

Vinsel said the magnitude of the AI bubble is likely bigger overall. “It’s gone past that,” he said. “I’ve seen things that show it’s more out of line with expected profits than the dot-com bubble was.”

The signs

There are a few other indicators setting off alarm bells for experts familiar with previous boom-bust cycles. One is the amount of “investment capital sloshing around,” Sicilia said, including a new class of less-experienced investors getting involved.

Odlyzko said the growing number of stakeholders points to a bubble, too. “What you’re seeing is all sorts of deals from other people coming in: the electric utilities gearing up to build out special financing for data centers, energy networks, and other things,” he said. “The numbers that people are talking about for this investment are getting huge.”

He mentioned this month’s somewhat circular deal between OpenAI and Nvidia had analysts and investors fretting about a bubble. Vinsel pointed to a recent $300 billion cloud deal between OpenAI and Oracle.

As for when the bubble might pop, it’s difficult to tell when it’s happening, experts said. “I have a feeling that, in retrospect, we’re already in the air-going-out moment,” Vinsel said.

The aftermath

You may have noticed that the biggest historic bubbles mentioned have not been around technologies that are unfamiliar today—radio, railways, the internet. Even if the AI bubble ruptures, that doesn’t mean the tech is necessarily going away.

Odlyzko compared it to the mid-19th century railway mania, when the bubble popped but the technology kept growing at a slower rate. “It collapsed. There was lots of wailing and gnashing of teeth among investors and so on,” he said. “But traffic kept growing.”

Some even argue, Vinsel said, that bubbles are important to lay the foundation for a later tech transformation: They allow for a level of investment that is not purely rational from a cost-benefit perspective for investors involved, but can lay the groundwork for important societal change.

“The dot-com bust happened, but we have Amazon,” Vinsel said. “We have these firms that were products of the moment that really have created lasting change for better and worse in our economy.”