The AI Stock Bubble Thesis – Why One MIT Economist Thinks We Are Exactly Where We Were in 1999

You almost instantly notice something when you pull up a chart showing the recent performance of the U.S. stock market. Fifty-five percent of the market’s growth can be attributed to ten companies, including Nvidia, Microsoft, Amazon, Broadcom, and six others riding the wave of AI enthusiasm. Not a widespread demonstration. Most boats are not lifted by a rising tide. The index is pulled upward by ten names that are concentrated at the top, while the majority of other names remain relatively flat. If anyone who was keeping an eye on the financial markets in 1999 recognizes that pattern, it’s most likely not a coincidence.

In a research note released in the summer of 2025, Apollo Global Management’s chief economist Torsten Sløk made a claim that alarmed a number of portfolio managers. He claimed that the overvaluation of AI stocks is even greater than that of dot-com stocks during their 1999 peak. Not overpriced either.

More. He drew comparisons to the valuation multiples that typified internet stocks in the months preceding the Nasdaq’s roughly 78 percent decline from its peak, concentrating on the same names that drove that concentration: Nvidia, Microsoft, and their cohort. The note went viral, creating the awkward silence that sometimes occurs in investment meetings when someone says something that is difficult to ignore.

Because it cuts both ways, marketers carefully consider the 1999 comparison. Everything was truly altered by the internet. The businesses that survived the crash—Amazon, for example—went on to produce profits that both justified and significantly outpaced the initial enthusiasm. The technology wasn’t the issue. The issue was that the stocks of companies that would not survive were priced as though they would, while the companies that would survive were priced as though their survival was assured and the timeline was shortened.

Many people still lose a lot of money when they are correct about the technology but incorrect about the valuation. In September 2025, GQG Partners succinctly stated that the belief in American economic dominance was the driving force behind both the dot-com bubble and the current AI boom. In the past, this belief has a tendency to be painful in the medium term and accurate in the long run.

Owen Lamont at Acadian Asset Management has presented the counterargument with greater rigor than most, and it is worth considering. His framework revolves around aggregate net equity issuance, which he refers to as the “Third Horseman of the Bubble Apocalypse.” The reasoning goes like this: corporate executives would be issuing new shares to profit from inflation if they truly thought their stock prices were inflated. From a corporate finance perspective, the best course of action when your stock is too high is to sell overpriced equity to willing buyers.

That’s exactly what occurred in 1999. The market saw more than 400 initial public offerings (IPOs), existing companies issued new shares at high prices, and eventually the supply of internet stocks outpaced demand, which contributed to a decline in prices. Over $250 billion had been issued overall by March 2000. Lamont points out that even if you were unfamiliar with the dot-com era, you could use the chart to determine the peak of the bubble.

The image appears differently today. Over the past year, corporations have bought back about $1 trillion in stock. Instead of selling new shares, companies are repurchasing their own. By 1999 standards, the IPO market is essentially quiet. Lamont’s conclusion that there are a trillion reasons why we are not yet in an AI bubble does not imply that valuations are fair or that there are no potential problems. It’s a more precise assertion: that the mechanism that has historically stopped bubbles isn’t working right now. Reading his analysis gives the impression that he is figuring out what’s lacking rather than completely discounting the issue.

It’s difficult to ignore the fact that both sides are genuinely in agreement on the fundamental facts when you watch this debate unfold in research notes and conference panels. The focus is genuine. The prices are high. There are historical similarities that should be investigated. The question of timing and mechanism is where they diverge, whether the market has more room to run before the dynamics that usually lead to these things start to materialize or whether the conditions for a correction are already in place due to the absence of an IPO wave.

On this point, history provides a frustrating precedent. Equity issuance peaked in 1989 and 1991 along with prices during the Japan bubble of the late 1980s. Between 1895 and 1897, the British Bicycle Mania of the 1890s gave rise to over 600 new businesses. In each instance, the issuance occurred, the correction ensued, and those who had been alerting others about it were vindicated at a time that did not benefit the investors who had already placed their bets.

AI may be sufficiently different from the internet era to warrant valuations that would have seemed concerning in any prior cycle. Although it is still up for debate, the productivity effects are real, the corporate adoption is quantifiable, and the compute infrastructure being built is real. It’s also possible that being correct about the price does not equate to being correct about the technology, as in 1999. Right now, it appears that the market is placing a lot of bets on the former while the latter question remains unanswered and is gaining attention.