Is There an AI Bubble, and Will It Pop? (Here's the Math)
What is an AI bubble burst?
Ask 10 people, and you'll get 11 opinions.
However, for the purposes of this post, we'll call it a significant, sustained drop in the valuations of leading AI-related companies, supported by official price data and major news outlets.
Now, let's see what people are most worried about.
Market concentration is at an all-time high
The first concern among many investors is the high concentration in the U.S. stock market.
It's currently at an all-time high, with the 7 biggest stocks (AI-exposed megacaps) comprising a whopping 33% of the S&P 500.
In a 500-company index, where only 7 account for a third, that surely can't be good, right?
Actually, economic research doesn't support this claim.
- Higher Stock Market Concentration Does Not Mean Higher Risk (Acadian, March 2024)
- Stock Market Concentration (Morgan Stanley, Jun 4, 2024)
- And a variety of other studies show there has been no historical correlation between market concentration and the probability of stock market dips.
However, this doesn't mean tail risks remain the same. With higher concentration, markets are more prone to sharper dips because of sudden "black swan" events.
Valuations are very high
It's well known that a high CAPE (Shiller cyclically adjusted P/E) has historically been a good indicator of lower-than-average returns going forward.
This has been shown by a variety of market research:
- Valuation Ratios and the Long-Run Stock Market Outlook (Campbell & Shiller)
- Equity Market Focus: Objective Expected Returns (AQR, 2025)
- Current Constituents CAPE (Research Affiliates, 2025)
This pattern has held many times across different markets: the "lost decade" of the US market after 1999, Japan after 1989, Canada in 2000, and so on.
However, "statistically more likely" does not mean "it will happen". And markets can sustain bull runs for a long time before an eventual decline.
The real problem: spending vs. revenue
If we look at the substance, the main problem with the current AI wave is simple:
Companies are spending huge amounts of money to build AI infrastructure and hire the best available AI talent, while not earning nearly enough to cover those costs.
Leaked industry reports support this claim: leading AI labs are burning billions of dollars and, as experts say, don't have a clear strategy for recouping them.
The strategy seems to be: "we'll go all-in on AI research, and then figure out how to make money".
That's why many AI tools are surprisingly cheap these days.
They are so cheap because they are subsidized by venture capital. You can build a working app for just a few dollars using Claude Code or OpenAI Codex because those tools have been paid for by AI investors. A mature, self-funding market would have priced them much higher than they are now.
How likely is a significant devaluation?
Companies are taking this high-stakes gamble not because they are dumb, but because they think they will win big. Of course, nobody knows the answer, and any estimates here would be probabilistic.
These estimates also depend on the time horizon. Looking at the next 2 years, the estimated chance of a "bubble" by the end of 2026 is roughly 20%, and by the end of 2027 it's 35-40%. There is probably no merit in looking even further, as predictions become increasingly uncertain the longer the time horizon.
So, the chances of it happening rather soon are there.
At the same time, there is likely a greater than 60% chance it won't happen by the end of 2027.
If that's true, we'll soon see even higher valuations and spending among the top AI companies.
What this means for you
People using AI tools are likely to continue being subsidized by venture capital in the coming years.
If you can leverage these tools for your work, now is the best time to do it.
Best,
Val