We Asked AI To Assess Stocks’ Fate If an AI Bubble Pops

The answers we received about which stocks might be safer were clear and confident. But they came from AI systems that predict text.

The clear and confident answers from ChatGPT and other large language models (LLMs) have convinced many people that these systems will soon reach, or have already reached, artificial general intelligence. This means being able to do any thinking task as well as or better than a human.

Long ago, Warren Buffett warned that excitement about major new inventions can create investment bubbles. Examples include railroads, airplanes, and the Internet. In these bubbles, stock prices rise far beyond what companies are likely to earn in the future. Buffett especially warned about what he called the “worst kind of business”: one that needs to keep spending large amounts of money, makes little or no profit, and has no lasting advantage over its competitors. That description seems to fit LLMs.

Consider OpenAI, the company that started the current excitement about LLMs. Its CEO, Sam Altman, often makes headlines with his comments. OpenAI has never reported a profit. In 2026, people inside the company predicted that it would go through a four-year “valley of death,” losing a total of $278 billion before becoming profitable in 2030. Given Altman's constant optimism, these figures may actually underestimate how much money OpenAI will lose before reaching that point. An independent study by HSBC estimated that OpenAI's total losses would reach nearly $500 billion by 2030. Even so, Altman has raised hundreds of billions of dollars and hopes the company will be worth more than $1 trillion when it sells shares to the public. However, its recently delayed IPO suggests that some investors may have doubts.

What should worried investors do? One possible strategy is to move their money into stocks that might be less affected if the AI bubble bursts. An unusual way to find these stocks is to ask several LLMs. We gave the following question to the latest versions of ChatGPT, Gemini, Claude, Grok, and Kimi K3:

How exposed are each of these companies on a scale of 1 to 10 to the AI bubble popping: Berkshire Hathaway, JP Morgan Chase, Home Depot, Alphabet, Unilever, American Express, JNJ, BroadCom, AMD, IBM, TSMC, SMCI, Micron, Apple, Costco, Microsoft, Amazon, Walmart, Meta, Nvidia?

These 20 stocks include the 10 companies that people admired most in Fortune's 2026 survey: Apple, Microsoft, Amazon, Nvidia, JPMorgan Chase, Berkshire Hathaway, Costco, Alphabet, Walmart, and American Express. We added 10 other companies that might produce a wider range of answers: AMD, BroadCom, Home Depot, IBM, JNJ, Meta, Micron, SMCI, TSMC, and Unilever. Table 1 shows the exposure scores given by the LLMs:

Table 1 Exposure to AI Bubble Popping, from 1 (lowest) to 10 (highest)

The results are mostly similar across the five LLMs. This is not surprising because they were probably trained on very similar collections of information. Their answers also generally make sense. This is a big improvement over earlier mistakes, such as inventing false references to support claims that Russia had sent dozens of bears into space.

There were some important differences. ChatGPT gave Meta a score of 6, while Gemini gave it 8.5. For JP Morgan, ChatGPT gave a score of 2, while Claude gave it 4. Also, LLMs do not always give the same answer. They predict text in a way that involves randomness, so asking the same question several times could produce different results.

We also found examples of the poor-quality answers that LLMs sometimes produce. In particular, their explanations focused mainly on how companies might lose revenue directly. They paid very little attention to how damage to the wider economy and stock market could affect the prices of all 20 stocks.

What happens if we use beta coefficients?

Beta coefficients are commonly used to measure how much a stock's price tends to move when the overall stock market moves. For example, if the market falls by 20%, a stock with a beta of 1 is expected to fall by 20%. A stock with a beta of 1.5 is expected to fall by 30%, while a stock with a beta of 0.5 is expected to fall by 10%. The last column of Table 2 gives the estimated beta values for these 20 stocks.

Notice that Amazon has an average exposure score of 6.3 in Table 1 but a high beta of 1.50. Microsoft has a slightly higher average exposure score of 6.9 but a much lower beta of 1.04.

We are not big fans of beta (for example, here). However, beta shows why we need to consider the effects of an AI crash on the whole economy, not just on companies that work directly with AI.

For example, the LLMs gave Home Depot an average exposure score of just 2.1. ChatGPT explained its score of 2 by saying: “Its housing, renovation, and consumer demand drivers dominate its direct AI exposure.” In other words, Home Depot depends more on the housing market, home improvements, and consumer spending than on AI.

But when we separately asked ChatGPT, “Would an economic recession have a large or small effect on Home Depot?,” it answered, “An economic recession would likely have a large negative effect on Home Depot.” Yet ChatGPT and the other LLMs failed to consider that an AI bubble bursting could cause serious damage to the economy as a whole, which would also hurt Home Depot.

ChatGPT made a similar mistake with JP Morgan. It gave the bank a low exposure score of 2, explaining: “AI-related market losses could affect sentiment and clients, but banking earnings are driven by many other factors.” However, when we separately asked how a recession would affect the bank, ChatGPT gave this answer:

An economic recession would likely have a large effect on JPMorgan Chase (JPM). Although JPMorgan is one of the strongest and most diversified banks in the United States, it is still highly exposed to economic conditions.

Home Depot's and JP Morgan's performance during the Great Recession and the COVID-19 market crashes had little to do with their direct exposure to subprime mortgages or COVID-19. The same thing would be true if the AI bubble burst.

What about other macroeconomic factors?

A good analysis of how well these 20 stocks might hold up should also look at other factors affecting the economy. These include the effects of lower interest rates. The Federal Reserve is likely to try to protect the stock market and the economy by cutting interest rates if an AI bubble bursts.

A good analysis should also examine information about each individual company, such as the figures in Table 2. When the Nifty-50 stock market bubble burst in the early 1970s, companies with the highest price-to-earnings (P/E) ratios generally did the worst. The same thing is likely to happen if the AI bubble bursts.

Two other useful measures are free cash flow and total yield. Free cash flow shows how much money a company has left after paying for its normal operations and investments. Total yield combines dividends and share buybacks. Both measures can help show whether a company is financially strong enough to survive major disruptions. They should be included in the analysis:

Table 2 Some Fundamentals

The main problem with the LLM answers is that they left out many important factors. A useful analysis would need careful thought and detailed financial models. That is much more than we can expect from a system that simply predicts text, no matter how clear and confident its answers may sound.