Bayesian Investor Blog

There’s a stock market bubble currently in progress. The key question is: which stocks are unsustainably high?

There are a few ways that the current AI boom could turn out to be a mistake (as in AI stocks dropping much below current levels; I’m not commenting here on whether AI will kill us all).

• Regulations outlaw most new AI development on safety grounds.
• Financial markets decide that frontier AI companies won’t make much profit (my counter-argument).
• Military action in Taiwan might cause an AI winter where lack of new GPUs slows progress dramatically for a couple of years.

I’m betting that if any of those causes happen, they’re more than two years away, and that the resulting crash will leave AI stocks above current levels.

Note that that I don’t take seriously any claims that AI is approaching technical limits. If AI hit a wall in 2025, it was bad news for the wall.

The most likely alternative to AI being in a bubble is that some large stocks in slow-growing industries are in a bubble. Not a stereotypical bubble, but a bubble that overconfidently predicts a normal economy.

My clearest example of a bubble-driven stock in this scenario is Walmart. Its stock doubled over the past two years, on a modest increase in profit margins. Its PE is over 40, which is normally appropriate only for companies with strong growth prospects. Yet its sales growth over the past 15 years has ranged from 1% to 6% per year. Per share numbers are maybe on track to grow at 8% as a result of share buybacks. That PE leaves little margin for error. It reflects overconfidence in market forecasts that real interest rates will remain low (roughly below 2% for 15 years, according to the TIPS market).

The paper Could Advanced AI Drive Explosive Economic Growth? suggests we shouldn’t be surprised if growth rates reach 30%. It’s likely that interest rates will rise roughly in line with economic growth expectations. That suggests a significant decline in PEs for stocks whose earnings growth lags the AI-related sectors.

Rising interest rates are also bad for banks. This is a more complex topic, and I’ll provide more detail in a future blog post. But banks have been betting, mostly as they have done for quite some time, that long-term interest rates will remain somewhat stable.

Banks have recently been getting away with paying interest on deposits that’s a bit less than economic growth rates and t-bill rates. But if t-bill rates rise to, say, 15%, banks will need to pay at least 10% or face large withdrawals. But they can support paying 10% when their income comes from 7% mortgages or 5% treasury bonds.

A version of such a reaction to rising interest rates, that played out more gradually and mildly than I expect with an AI boom, contributed to the Savings and Loan crisis of the 1980s.

It will probably take more than two years for interest rates to rise enough to pop such a bubble.

Past booms of sizes comparable to the current AI boom have invariably ended in financial bubbles, even when the underlying businesses were sound (e.g. railroads). So odds are that we’re headed for an AI financial bubble.

Keep in mind that warnings about bubbles reliably start before prices reach bubble levels. I remember concerns in 1993 among computer science students that Microsoft had become a more valuable company than General Motors. It may be hard to believe today, but for most of the 20th century it was considered standard wisdom that companies which had stood the test of time were wiser investments than inexperienced startups.

And don’t forget Greenspan’s 1996 warning about high stock prices. If you sold then, you would have watched as the market more than doubled over a period of a little more than three years.

See also my AI Winter post from 6 years ago about whether AI was in a bubble then.

Micron is still my favorite stock pick. Its impressive guidance caused analysts to raise 2026 earnings forecasts by 78%. That seems almost as dramatic as the NVIDIA’s May 2023 guidance which kicked off a 4x return over the next year. This is not the kind of surprise we see near the peak of a bubble.