Are We in an AI Bubble? William Bernstein's Four Signs and Why Investors Should Stay Humble

In the space of two weeks this June, SpaceX went public under the ticker SPCX, and both Anthropic and OpenAI confirmed that they had confidentially filed to do the same.¹¹ Anthropic had just raised money at a reported $965 billion valuation; OpenAI’s last private round valued it near $852 billion.¹² Add a stock market where a handful of AI-linked companies drive most of the returns, and the word that keeps surfacing is bubble.

I do not think anyone can know, right now, whether we are in an AI bubble. I would go further: I doubt a bubble can be reliably identified while it is happening. We get clarity in hindsight, rarely in the moment. So this guide is not a market call. It is about a tool that survives that uncertainty: William Bernstein’s four signs of a bubble, and how a do-it-yourself investor should actually use them.

The short version: Bernstein’s four signs are a useful behavioral checklist and a poor trading system. They can tell you the social environment around an asset has turned dangerous. They cannot tell you when, or whether, the price will fall. Treat them as a prompt to control your own risk, not as a signal to sell, short, or go to cash.

Bernstein’s four signs

William Bernstein, a neurologist turned financial historian and author of The Four Pillars of Investing, offers four warning signs that a market has tipped into speculative mania. What makes them distinctive is that none of them is about valuation. They are sociological. They describe how people around you behave.¹

1. The speculative asset becomes everyday chatter. People with no investing background start talking about it at dinner, at the gym, in the group chat.
2. People leave stable jobs to speculate. Day trading or flipping the hot asset starts to look like a viable career.
3. Skepticism is met with hostility. Question the story and you are dismissed as clueless, bitter, or “just not getting it.”
4. Extreme predictions become common. Forecasts of prices quadrupling, rather than the ordinary moves markets usually deliver, start to sound reasonable.

Bernstein is explicit about the limits of his own checklist. These signs tell you that the psychology has become dangerous. They do not pinpoint when the bubble bursts.¹ That caveat matters, and most four-signs explainers skip past it.

Are the four signs “factually driven”?

Partly, but not in the way a quant would mean it. The four signs are not a validated factor model. They are qualitative warning signs drawn from financial history and crowd behavior. That does not make them useless. It means they are observable symptoms, not a diagnostic test with a sensitivity and a specificity.

Where the academic evidence is strongest is in the machinery underneath the signs: investor sentiment, return extrapolation, the limits of arbitrage, and crash risk after extreme run-ups. Robin Greenwood and Andrei Shleifer, studying decades of survey data, found that the returns investors expect are extrapolative. Expectations rise after past gains and when the market is already expensive, exactly when model-based expected returns are moving the other way.⁴ That is sign one and sign four with a regression behind them.

Do investment bubbles exist?

Yes in a practical, historical sense. Harder to prove in a strict real-time econometric sense.

The cleanest evidence comes from the laboratory, because researchers can define fundamental value directly. In the classic Smith, Suchanek, and Williams experiments, traders bought and sold an asset whose true payout was known in advance. Prices still ran far above fundamentals and then crashed: 14 of 22 experimental markets produced bubbles. The bubbles faded as the same traders gained experience.³ When you can pin down what something is worth and prices still detach, “bubble” is not just a figure of speech.

Real markets are messier but point the same way. Robin Greenwood, Andrei Shleifer, and Yang You examined sharp U.S. industry run-ups in a paper pointedly titled Bubbles for Fama. Their finding cuts both ways, which is why it is worth citing carefully. A sharp price run-up did not reliably predict low average future returns, a point for the skeptics. But it did predict a sharply higher probability of a crash, defined as a 40% drawdown within two years. As a sector’s two-year return relative to the market rose from 50% to 100%, the crash probability climbed from about 20% to 53%; at a 150% run-up it reached roughly 80%.²

So a run-up is not a verdict, but it is a real shift in the odds. The same paper adds the inconvenient detail every would-be bubble-trader learns the hard way: prices often kept rising before they crashed.²

Why my skepticism is academically defensible

The claim that bubbles are easier to verify in hindsight than to identify in the present is close to the academic mainstream. Refet Gürkaynak’s survey of econometric bubble tests reached a blunt conclusion: detection cannot be achieved with a satisfactory degree of certainty, because for each study that finds a bubble, another fits the same data without one. The tests cannot cleanly separate a bubble from time-varying fundamentals.⁵

Eugene Fama’s objection is definitional and predictive. He does not argue that prices always look right. His point is that to be useful, a “bubble” claim needs a systematic way to identify when the episode ends, and he argues no one has shown they can do that reliably. Calling tops in hindsight is easy; doing it in advance is the part that has not held up statistically.⁶

The skeptical view should not curdle into “bubbles are a myth,” though. Behavioral finance gives concrete mechanisms for why speculative episodes form and persist. Andrei Shleifer and Robert Vishny’s work on the limits of arbitrage explains why mispricing need not vanish just because smart investors notice it: real arbitrage requires capital, carries risk, and becomes least effective precisely when prices have diverged furthest from fundamentals, because that is when arbitrageurs face redemptions and are forced to liquidate.⁷

Calling a bubble and profiting from the call are different problems

Even a correct diagnosis is hard to monetize. You can be early. You can be right on direction and lose money shorting. You can sell too soon, trigger taxes, and fail to get back in.

The cleanest illustration comes from the people best equipped to bet against a bubble. Markus Brunnermeier and Stefan Nagel studied hedge fund holdings through the dot-com era and found that funds did not lean against the bubble. They were heavily invested in technology stocks and rode the run-up. This was not naivety: on a stock-by-stock basis they reduced positions in names that were about to fall, capturing much of the upside while sidestepping part of the collapse.⁸ Sophisticated investors who clearly saw the speculation chose to ride it, carefully, rather than fight it. That is the opposite of the “spot the bubble, go to cash” instinct.

Carlota Perez: bubbles often form around real technologies

If Bernstein asks what a speculative crowd looks like, the economist Carlota Perez asks why big technology bubbles keep appearing around genuine technological revolutions. Her 2002 book, Technological Revolutions and Financial Capital, is the best antidote to the lazy idea that a bubble means the technology is fake.¹⁰

Perez argues that capitalism advances through recurring great surges of development, each lasting about half a century and built on a new technological revolution: the Industrial Revolution from 1771, steam and railways from 1829, steel and heavy engineering from 1875, oil and the automobile and mass production from 1908, and information and telecommunications from 1971. Each surge splits into two halves. An installation period, when financial capital rushes into the new technology, builds its infrastructure, and drives a speculative frenzy, gives way through a turning point of crisis and crash to a deployment period, the “golden age” in which production capital spreads the technology across the whole economy.¹⁰

The unsettling part of her argument is that the bubble is partly functional. The frenzy overfunds the new infrastructure, and after the paper wealth is wiped out, the canals, railways, and fiber-optic cable remain, ready for the deployment that follows. The speculative episodes even take the name of the infrastructure being built: canal mania, railway mania, the internet bubble.¹⁰ The technology was real every time. Many of the investors were still ruined.

For a DIY investor, that reframes the question toward price. Even if this technology changes the world, how much of that future is already in the stock, and are you buying the part of the value chain that captures it? A general-purpose technology can hand most of its gains to consumers, workers, suppliers, and later entrants rather than to the hyped early stocks. Eli Ofek and Matthew Richardson’s study of the dot-com crash documents how high valuations, optimistic buyers, and short-sale constraints inflated internet stocks, and how the eventual wave of selling, triggered by expiring lockups, brought them down.⁹ The internet won. A lot of internet stocks did not.

Perez is careful, and so should we be, about treating this as a clock. She calls her framework a stylized narrative with approximate dates and no clean breaks between phases, a historical lens for understanding the shape of a cycle rather than a tool for timing trades.¹⁰

A timely test: AI, SpaceX, and the mega-IPO moment

Which brings us back to June 2026. The AI and mega-IPO cycle is a useful place to practice bubble thinking, precisely because no one can honestly declare in real time that “AI is a bubble.” AI may turn out to be one of the most important technologies of our lifetime. SpaceX, OpenAI, Anthropic, the chipmakers, and the cloud build-out may all create enormous real value. None of that settles whether a given investor-facing opportunity is attractive at its current price.

Run the four signs as questions rather than conclusions. Is the theme everyday chatter? Are people reorganizing careers and portfolios around it? Are skeptics treated as fools? Are extreme forecasts becoming normal? These questions measure whether the environment still tolerates skepticism at all, which tells you more than whether any individual skeptic turns out to be right.

Mega-IPOs are worth singling out because they convert a private-market narrative into public-market participation, which is to say they manufacture attention and scarcity at the same time. And the structure favors the people who least need the help. The SEC’s investor education site states plainly that investing in an IPO is, by its nature, risky and speculative, that brokerage firms must consider whether an IPO is appropriate given your income, net worth, objectives, holdings, and risk tolerance, and that underwriters typically steer “hot” IPO shares to their most valued clients, so individuals often struggle to get them.¹⁴ Vanguard’s education makes the same point about the asset itself: IPOs carry single-stock risk, have limited public trading history, and can swing sharply in the days and weeks after the offering.¹⁵ Two companion guides go deeper: Should You Buy IPO Stocks? on the academic return evidence, and Mega-IPOs and Your Index Funds on how much of a SpaceX or OpenAI listing your VTI or VOO actually absorbs.

The SpaceX listing put the asymmetry on display. Reuters reported that retail investors ended up with about 20% of the offering, while long-term institutions took roughly 70%, and that brokers such as Fidelity, Robinhood, SoFi, and E*TRADE restricted small investors from selling for 15 to 30 days, even as some large funds faced no such anti-flipping limit and could sell straight into the open.¹¹ You can hold a strong view that SpaceX is an extraordinary company and still notice that the terms of access were not the same for everyone.

The aim here is to notice when an investment story becomes so compelling that ordinary risk controls start to feel optional, well before anyone could prove that AI, SpaceX, OpenAI, or Anthropic are bubbles. That feeling is what Bernstein’s signs are built to catch, and no valuation number captures it.

What this means for DIY investors

The goal here is narrow: keep a bubble environment from breaking your plan. That reframes who should care, and how much.

It matters most if you are:

• Concentrated in a hot asset, sector, or single stock.
• Holding significant employer stock or RSUs tied to the same theme as your job.
• Adding thematic ETFs, hot IPOs, or private deals on top of that.
• Using leverage or margin to chase the story.
• Near retirement, where a deep drawdown early in withdrawals does lasting damage through sequence-of-returns risk.

It matters far less if you:

• Already hold a globally diversified portfolio.
• Rebalance mechanically rather than by conviction.
• Avoid leverage and large discretionary single-name bets.

For the second group, the bubble narrative is mostly a temptation to do something unnecessary. Keep contributing, rebalance when allocations drift, and do not let “it feels like a bubble” become a license to time the market.

For the first group, the defensible responses are unglamorous and they work: cap how much of your net worth rides on any one theme, diversify concentrated and employer-linked positions on a schedule, rebalance when a winner grows past its target, and refuse leverage on a story. Notice that none of these require you to predict anything. They are risk controls, not forecasts. Selling everything to cash, or shorting, asks you to be right about timing, which is the part even the hedge funds did not attempt.

Stress-test your own exposure

The tool below is a stress test rather than a bubble detector, because a detector would be fake precision. Instead of asking whether this is a bubble, it asks the question you can actually answer: if your hot theme fell 30%, 50%, or 80%, what happens to your plan? Set your real numbers below, including any AI or tech exposure you already carry inside index funds and any overlap with your job.

Key takeaways

• Bernstein’s four signs are a behavioral checklist, not a valuation model or a sell signal. They flag dangerous psychology, not timing.
• Bubbles are real enough to damage investors. Lab experiments and industry run-ups both show prices detaching from fundamentals, with run-ups sharply raising crash risk.
• They are also hard enough to identify in real time that most investors should not trade on the call. Detection is statistically uncertain, and timing is worse.
• Carlota Perez’s history shows bubbles often form around real technologies. A world-changing technology can still be a poor investment at the wrong price.
• The AI and mega-IPO moment is a live test of the signs, not proof of a bubble. IPOs amplify attention and scarcity and rarely favor the individual investor.
• The reliable response is risk control, not prediction: cap single-theme exposure, diversify employer stock, rebalance, and avoid leverage.

Frequently asked questions

What are William Bernstein’s four signs of a bubble?

That the speculative asset becomes everyday chatter, that people leave stable jobs to speculate, that skeptics are met with hostility, and that extreme price predictions become common. They are sociological signs about crowd behavior, not valuation measures, and Bernstein notes they do not predict when a bubble will burst.¹

Can you predict a market bubble in advance?

Not reliably. Surveys of econometric bubble tests find that detection cannot be done with confidence, because models that find a bubble can usually be matched by models that explain the data without one.⁵ Sharp run-ups do raise the measured probability of a crash, but prices often keep climbing first, which makes betting against them costly even when the diagnosis is right.²

Is AI a bubble in 2026?

Nobody can answer that honestly in real time, which is the point of this guide. AI may be genuinely transformative and still contain pockets of overvaluation. The question that actually helps is how much of your own net worth depends on AI expectations, and whether your plan survives a deep drawdown in the most speculative names even if the technology succeeds.

Should I sell my stocks if I think we are in a bubble?

Going fully to cash or shorting requires you to be right about timing, which even sophisticated hedge funds avoided during the dot-com bubble; they rode it carefully instead.⁸ The more defensible moves are to cap single-theme exposure, diversify concentrated or employer stock, rebalance to your targets, and avoid leverage. Those control risk without requiring a forecast.

Related guides

• Concentration Risk: When to Sell Your Company Stock turns “cap your single-theme exposure” into concrete math, including how to measure concentration and run the sell-vs-hold decision.
• RSU Tax Strategy for Equity Compensation covers the case where your job and your portfolio are exposed to the same theme, and how to diversify employer stock tax-efficiently.
• The Problem With Buy the Dip applies the same skepticism toward market timing from the other direction: why waiting for the crash usually costs more than it saves.
• Should You Buy IPO Stocks? covers the academic evidence on IPO returns and why the “ground floor” is so often the exit.
• Mega-IPOs and Your Index Funds explains how free float decides what SpaceX, OpenAI, and Anthropic actually mean for VTI, VOO, and QQQ.

Sources

1. Bernstein, W. Revisiting “The Four Pillars of Investing” (Morningstar interview), and The Four Pillars of Investing, 2nd ed. (McGraw-Hill, 2023). The four sociological signs of a bubble and the caveat that they do not predict timing.
2. Greenwood, R., Shleifer, A., & You, Y. (2019). Bubbles for Fama. Journal of Financial Economics, 131(1), 20-43. Sharp run-ups do not reliably predict low average returns but sharply raise crash probability (about 20% to 53% to 80% as two-year run-ups rise from 50% to 100% to 150%).
3. Smith, V.L., Suchanek, G.L., & Williams, A.W. (1988). Bubbles, Crashes, and Endogenous Expectations in Experimental Spot Asset Markets. Econometrica, 56(5), 1119-1151. 14 of 22 markets bubbled; bubbles diminished with trader experience.
4. Greenwood, R., & Shleifer, A. (2014). Expectations of Returns and Expected Returns. Review of Financial Studies, 27(3), 714-746. Investor expectations are extrapolative and negatively correlated with model-based expected returns.
5. Gürkaynak, R.S. (2008). Econometric Tests of Asset Price Bubbles: Taking Stock. Journal of Economic Surveys, 22(1), 166-186 (FEDS 2005-04). Bubble detection cannot be achieved with a satisfactory degree of certainty.
6. Fama, E.F. Are Markets Efficient? Chicago Booth Review (Fama-Thaler discussion). A usable bubble claim requires a systematic way to predict the end, which Fama argues has not been demonstrated.
7. Shleifer, A., & Vishny, R.W. (1997). The Limits of Arbitrage. Journal of Finance, 52(1), 35-55. Arbitrage requires capital, is risky, and becomes ineffective when prices diverge far from fundamentals.
8. Brunnermeier, M.K., & Nagel, S. (2004). Hedge Funds and the Technology Bubble. Journal of Finance, 59(5), 2013-2040. Hedge funds rode the tech bubble and trimmed positions stock-by-stock rather than leaning against it.
9. Ofek, E., & Richardson, M. (2003). DotCom Mania: The Rise and Fall of Internet Stock Prices. Journal of Finance, 58(3), 1113-1137. High valuations, optimistic investors, short-sale constraints, and lockup expirations.
10. Perez, C. (2002). Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages (Edward Elgar). Great surges of development, installation vs. deployment, financial vs. production capital, and bubbles that form around real new infrastructure.
11. Valle, S., & Wang, E. (2026). Retail investors face tighter limits than funds in SpaceX IPO flipping. Reuters, June 15, 2026. SPCX listing; retail took about 20% of the offering; 15-to-30-day anti-flipping limits on retail brokers.
12. Anthropic. Confidential draft registration statement (Form S-1) (June 1, 2026) and Series H announcement ($65B raised at a $965B post-money valuation, May 28, 2026). OpenAI’s last confirmed private valuation was about $852B (March 2026); reported IPO targets beyond that are speculative.
13. U.S. Securities and Exchange Commission. Initial Public Offerings: Eligibility to Get Shares and Why Individuals Have Difficulty Getting Shares. Investor.gov. IPOs are risky and speculative; suitability factors; allocation favors valued clients.
14. Vanguard. What’s an IPO? Risks, rewards, and how to get started. Single-stock risk, limited public trading history, and sharp price moves around the offering.