Are We in an AI Bubble?

Overview

This episode asks whether the current AI-driven stock market boom is a bubble, focusing on the rapid rise of the S&P 500, the dominance of AI-linked companies, and Nvidia’s extraordinary market value.

The hosts explain why bubbles are difficult to identify in real time: they usually form around new, uncertain technologies, where wildly different stories about future value can coexist. Harvard Business School professor Robin Greenwood describes research that tries to detect bubbles statistically, using clues like high valuations, volatility, new share issuance, and accelerating stock prices.

The episode concludes that the AI boom shows some bubble-like signs, but not all of them. It may be an “early bubble,” but bubble detection remains imprecise. The second half turns to policy and macroeconomic consequences, asking whether bubbles should be resisted or cleaned up after they burst, and whether some bubbles can leave useful infrastructure or research behind.

Segment-by-Segment Summary

[00:30] The AI Bubble Question

[Fact] The hosts frame the central question as whether the economy is currently in a bubble.

[Fact] They note that the S&P 500 is up almost 50% over the last two years, with much of the growth tied to AI-related companies.

[Fact] Companies mentioned include Microsoft, Amazon, Meta, and Nvidia, with Nvidia described as the microchip company powering much of the AI boom.

[Fact] Nvidia’s stock price is described as having almost quadrupled over two years, making it the most valuable company in the world.

[01:53] What the Episode Will Examine

[Fact] The episode sets out to ask whether it is possible to tell if one is in a bubble and what a bubble would mean for the broader economy.

[Fact] The hosts say they will discuss recent research on bubble detection.

[Fact] They also introduce a theory that some bubbles may not be as frightening as they seem.

[03:12] Defining a Bubble

[Fact] Robin Greenwood defines a bubble as something irrationally valued relative to the value it delivers.

[Fact] The textbook idea presented is that people buy and sell something at prices far above what it is actually worth.

[Fact] The episode explains that bubbles often involve something new, exciting, or unknown, which makes true value hard to assess.

[Fact] AI is described as fertile ground for a possible bubble because no one knows how transformative it will be.

[05:21] Why Nvidia Is Hard to Value

[Fact] Investors are said to value Nvidia at $4.6 trillion, compared in the episode to 22 Disneys or five JPMorgan Chases.

[Fact] The hosts say it is difficult to know whether Nvidia is overvalued because its value depends on beliefs about whether AI chips will change the world.

[Fact] Greenwood says multiple narratives about Nvidia’s future can survive in the market because its value is hard to determine.

[06:15] Eugene Fama’s Challenge

[Fact] Eugene Fama is introduced as a Nobel Prize-winning economist known for the theory that markets are mostly efficient.

[Fact] In a previous Planet Money appearance, Fama argued that if markets work, bubbles should not be predictable.

[Fact] Fama says the word “bubble” bothers him and that evidence would require showing reliable prediction of when such things turn.

[Fact] Greenwood says Fama’s view helped motivate his team to try to statistically identify bubbles.

[08:04] The Historical Bubble Study

[Fact] Greenwood and colleagues examined nearly a century of U.S. stock market data.

[Fact] They looked for industries where stock prices doubled or more within two years.

[Fact] They found 40 examples, including electricity company stocks in the 1920s and health care stocks in the 1970s.

[Fact] About half of the cases did not crash afterward, while the other half did crash dramatically within a couple of years.

[09:20] Four Bubble Clues

[Fact] The researchers did not find one definitive marker of a bubble, but they found a constellation of clues that made bubbles somewhat predictable.

[Fact] The four clues were high valuations, volatility, issuance, and acceleration.

[Fact] High valuations mean stock prices are high relative to current earnings.

[Fact] Issuance means many new companies go public or existing companies sell new shares to public investors.

[Fact] Acceleration means stock prices are not only rising but rising faster and faster.

[11:13] Applying the Clues to AI

[Fact] Greenwood says the AI boom shows some high valuations, including Nvidia’s price-to-earnings ratio in the 40s, compared with an S&P 500 average closer to the 20s.

[Fact] The AI-related market also shows some increased day-to-day volatility.

[Fact] Greenwood says there has not been much new stock issuance among major AI companies or private AI companies going public.

[Fact] He says AI-linked stocks have risen, but recently have not been rising faster and faster.

[Fact] His overall assessment is that the AI boom has many, but not all, signs needed to call it a bubble.

[12:47] “Early Bubble” and the Limits of Prediction

[Fact] Greenwood says that if he had to describe the current situation, he would call it an “early bubble.”

[Fact] He also emphasizes that the clues are not highly accurate.

[Fact] Looking back over past stock market spikes, the clues helped identify bubbles about 60% of the time.

[Fact] The episode says this is only a little better than a coin flip.

[13:52] A Truce With Fama

[Fact] One of Greenwood’s co-authors presented the “Bubbles for Fama” paper at the University of Chicago, where Fama was present.

[Fact] Greenwood says Fama was interested in the data but interpreted the observations differently.

[Fact] Greenwood characterizes the result as a truce rather than a clear refutation of Fama.

[Fact] The hosts conclude that detecting bubbles remains very difficult.

[15:14] Policy Debate: Lean Versus Clean

[Fact] The episode shifts to whether policymakers should do anything about bubbles if they can identify them.

[Fact] Gaudy Barlevy of the Chicago Fed is introduced, with the clarification that he is speaking for himself and not for the Fed.

[Fact] Barlevy describes the “lean versus clean” debate: whether the government should push back against a suspected bubble or wait and clean up after it bursts.

[Fact] The hosts say this debate became more prominent after the dot-com crash and the housing bubble.

[17:12] Dot-Com and Housing Bubble Consequences

[Fact] The dot-com bubble burst around 2000, many internet companies went bankrupt, and the Nasdaq fell 78%.

[Fact] The episode says the dot-com crash helped push the U.S. economy into recession.

[Fact] A later U.S. housing bubble also burst, contributing to the global financial crisis.

[Fact] These events made macroeconomists more interested in bubbles.

[18:10] How Bubbles Damage the Economy

[Fact] The episode says bubbles can hurt the economy when they pop, but the severity depends on how connected the bubble is to the broader economy.

[Fact] Important factors include who invested, how many workers are in the affected industry, and whether borrowing fueled the bubble.

[Fact] Barlevy says heavy borrowing can lead to defaults, bank losses, reduced lending, and a more severe recession.

[Fact] The housing bubble is given as an example because mortgage borrowing connected falling home prices to bank distress.

[19:15] Wasted Investment Before a Bubble Pops

[Fact] The hosts explain that bubbles can also hurt before they burst by directing money toward the wrong things.

[Fact] A toy example involving Labubus is used to illustrate speculative buying.

[Fact] The episode distinguishes between people buying something because they genuinely value it and people buying it because they expect to sell it later for more.

[Fact] The hosts say the problem with a speculative bubble is that society may end up producing too much of something people do not actually want.

[20:58] If the AI Bubble Pops

[Fact] The hosts ask how bad it would be if AI were a bubble and burst.

[Fact] They say investors would lose a lot of money, jobs could be lost, spending could fall, and effects could ripple outward.

[Fact] They cite one economist who told The Indicator that an AI crash could erase $35 trillion from the global economy.

[Fact] The episode says an AI crash may be less threatening than 2008 because AI companies are not borrowing directly from banks as much.

[Fact] The hosts also note uncertainty about how much private credit itself borrows from banks.

[22:00] Could AI Investment Still Be Useful?

[Fact] The hosts ask whether the current spending on AI training and data centers would be wasteful if AI disappoints.

[Fact] They compare AI infrastructure to dot-com-era fiber optic cables, much of which initially sat unused as “dark fiber.”

[Fact] The episode says that fiber later helped support broadband and today’s streaming video.

[Fact] The hosts suggest that unused AI data centers and computers may have other uses, unlike purely speculative collectibles.

[23:15] The Silver Lining Theory of Bubbles

[Fact] Some economists are described as arguing that not all bubbles are entirely bad and that some may even boost the economy.

[Fact] Barlevy does not fully accept the theory, but finds it interesting.

[Fact] The argument is that if society underinvests in something valuable, such as research and development, a bubble might push more money into that area.

[Fact] The episode explains that companies often underinvest in R&D because research can benefit competitors, making it a public-good problem.

[Fact] [Speculation] In the AI case, this theory implies that even a mistaken boom could leave behind useful infrastructure or research spillovers.

[25:02] Closing Jokes and Credits

[Fact] The hosts jokingly ask whether Dubai chocolate or podcasts are bubbles.

[Fact] They ask listeners to help grow the show by sharing Planet Money with a friend.

[Fact] The episode credits the producer, editor, fact-checker, engineers, executive producer, and hosts.

Podcast Review/Summary

This episode is valuable because it treats the AI bubble question as an empirical and policy problem rather than a simple market prediction. Its strongest section is the explanation of Greenwood’s bubble indicators, which gives listeners a concrete framework without pretending that the framework can deliver certainty.

The discussion is also useful because it separates two questions that are often blurred together: whether AI stocks are overvalued, and how damaging a possible crash would be for the broader economy. The comparison between bank-funded housing speculation and investor-funded AI spending is especially clarifying.

The episode’s limitation is that it cannot give a firm answer on whether AI is truly in a bubble. That is partly the point: the transcript repeatedly emphasizes uncertainty, weak prediction, and mixed signals. [Speculation] It is best suited for listeners who want an accessible economics framework for thinking about AI markets, rather than investment advice or a definitive forecast.