Is the artificial intelligence boom a bubble? There are growing reasons to argue it is: Valuations of AI companies dwarf their revenues, the cost of investment in chips and data centers remains exorbitant, and there are fears that some firms are self-dealing and artificially pumping the growth of the industry. On the other hand, the companies investing in AI are by far the world’s most profitable, and their advocates argue that those firms will transform our economies and how we think about the internet and payments. 

Economist Jared Bernstein, who chaired President Joe Biden’s Council of Economic Advisers, argues we are in bubble territory. I invited him on FP Live to make the case and hear out some of the counterpoints. Subscribers can watch the full discussion on the video box atop this page or on the FP Live podcast. What follows here is a lightly edited and condensed transcript. 

Ravi Agrawal: We’ve had two bubbles this century: the dot-com bust in 2001 and the housing crisis in 2008. Tell us why you think AI might become the third one.

Jared Bernstein: First of all, thank you for saying “might,” because one has to hedge such predictions. What Ryan Cummings and I argued in a recent New York Times op-ed was that AI certainly has a lot of bubbly characteristics. A bubble occurs when the level of investment in an asset is so high that investors stop believing they can get an adequate return on that investment in a reasonable time frame. So they pull out, and the highly elevated valuations start to deflate. 

The evidence, to answer your question, is that the valuation of the firms involved, often referred to as the “Magnificent 7” or “Mag 7,” is north of $20 trillion. [Ed: The Mag 7 includes Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla.] These are very high, very quick valuations. More to the point, the price-to-earnings ratio—meaning how the valuations of these companies are relative to their earnings—is quite bubbly and excessive. So that’s often a sign. 

A concrete example is that OpenAI, one of the premier AI firms in the Mag 7, said it’s going to invest a trillion dollars this year. Its revenue is something like $13 billion—very low revenue. Now obviously that’s one year, and one of the key points we make is that these returns are expected over many years. So if investors have the patience to wait it out, then perhaps the bubble will disinflate without causing much damage. 

But with the dot-com bubble, you had precisely what we’re describing: very high price-to-earnings ratios—higher than they are now, by some measures, but not that much higher—and investors got quite impatient. You started seeing some bankruptcies among the firms. The bubble deflated, and we had a recession.

RA: A few other things lead people to call this current moment a bubble. Mira Murati, formerly of OpenAI, founded a new start-up, Thinking Machines Lab, that was valued at $10 billion without having a real product or a plan. She was able to raise $2 billion immediately off of that. There are also growing concerns about financial opacity and quirky accounting practices. And lastly, people are scared off by circular investments: AI companies investing in other AI companies, which are investing in other AI companies. For example, Nvidia pledged to invest in OpenAI, but OpenAI will spend some of that money buying more chips from Nvidia. 

When you look at these practices, how much feels particular to AI and how much of it reminds you of the dot-com bubble?

JB: For all the nuances you just injected, it all comes down to revenue generation. Remember, if these companies fail to generate the kinds of revenues that justify their valuations, then the likelihood of a bubble grows and could grow quickly. The things you raised are precisely the sort that would potentially lead to that problem.

Take circular financing, which is a potentially problematic development. Imagine that Safeway lends me money. I go to Safeway and buy groceries. That might look good on its balance sheet for five minutes. But obviously, if that circle stays closed, you end up with exactly the problem that I don’t want to lose sight of: You don’t have enough outside investments flowing in. 

You also mentioned business models. I think AI is in a stronger place than the internet bubble back in the late 1990s and 2000s. Meta and Alphabet are not the same as Pets.com, the classic example of a company with an indefensible business model. These companies generate a lot of cash flow. Amazon sells gazillions of things to the world. Meta, Microsoft, and even Tesla all have cash flow. This is a common argument about why AI is not a bubble. 

But here’s the thing. If you actually isolate the cash flow that’s generated from AI versus their spending on AI, you see a very stark imbalance. Take Amazon. Take all the books, cat food, shirts, and sweaters out of the picture and look exclusively at AI. It’s investing around $100 billion a year and its revenues are in the single-digit billions. There’s a huge gap there. 

So I’m a little bit underwhelmed by the argument that says, “These companies still have a lot of cash flow.” If the AI components fail to generate the profitability that is required to meet the valuations, that part will bust. That doesn’t mean the companies will go out of business, but it means that a lot of investors will lose. We’ll have a bubble and all the economic implications that come with it.

One final point is that just because a technology is a bubble doesn’t mean it won’t be transformational. Take the internet. We had a dot-com bubble, no question about it. The unemployment rate went up about 2 percentage points when that bubble burst. A lot of people got hurt. But the internet has been a highly productive and transformational investment.

RA: Those are good points and caveats. One thing that gives AI boosters confidence is that some of the companies in the Mag 7 are the most profitable companies in the history of the world. They aren’t playing with borrowed money. This clearly isn’t the housing crisis. Their valuations are high, but they’re not that crazy compared to Oracle, which was trading at 100 times its earnings in 1999. 

These companies have so much money and are playing off their profits to move toward dominance in the future. They can afford to do so. That makes it sound less like a bubble.

JB: To some extent, that’s true. But let me put some numbers on the point I made a few minutes ago. These are calculations by my colleague Ryan Cummings. If you look at AI spending versus AI revenues for Microsoft, Amazon, Google, Meta, you’ll find that their capital expenditures, plus their R&D spending, range from an estimated $70 billion to $100 billion in 2025. The highest amount of revenue is Microsoft at $15 billion. The lowest is Meta, with $300 million of revenue.

If AI is a bubble and it turns out that over the next, say, five years, that gap doesn’t close enough, it doesn’t mean that Microsoft, Amazon, Google, Meta, etc. will go bankrupt. These are solid companies with many other lines of business. But it does mean that the valuations for their AI components will take a hit, and that’s specifically the bubble we’re talking about. Time will tell.

The second point is that their investments are coming from their own cash flow. They’re not really exposed to any of the squirrely debt financing that has been all too characteristic in some past bubbles. Over the past few quarters, the cash-to-total asset ratio among hyperscalers—really deep and heavy investors—has declined significantly. It was 29 percent at the end of 2021, and it’s 15 percent at the end of Q2 of this year. That means they’re now spending more borrowed money to build up data centers, buy chips, and all that. That’s still a pretty low ratio compared to the housing bubble. The damage done by the opaque securitization of that bubble is by no means replicated here. But there is a lot of opacity, and we’re starting to see more of it. So now I have to worry about that, too.

RA: I’ve also heard the argument that the bubble will never pop. A lot of the clients of, say, Microsoft and Amazon, are already using generative AI tools. Meta and Google are using AI to get better at selling ads. Once ChatGPT and Claude start charging for consumption, there are a lot of revenue pluses that could emerge in the coming years. 

But this is also a good time to ask: Are bubbles necessarily bad? Or, as some venture capitalists would say, are bubbles actually good because they align capital and talent in new trends, creating new businesses that change the world?

JB: First of all, I want to underscore that consumer use just really hasn’t been monetized that much yet. There’s something like 80 percent of people using ChatGPT here and there, and that’s just generating these really very low revenue streams.

RA: I have read, however, that AI-related revenue is up nine times in the past two years, for what it’s worth.

JB: Yes, but that’s from a very low base. OpenAI is talking about investing a trillion dollars this year with revenues around $15 billion. If that $15 billion was $2 billion before, then, yes, it’s more than seven times higher. But that gap between investment and revenue is what matters. 

But look, for a bubble not to happen in the AI case, you’d have to move from a consumer model to an enterprise model. Yes, consumers are using AI all over the place. But for businesses, the breakthrough has been somewhere around 10 percent. It would have to go well above that. But maybe it will. 

Now your more interesting, profound, and almost sort of philosophical question: Are bubbles good? Because bubbles represent excess speculation—and excess usually leads to problems in economies—I wouldn’t necessarily call them good. But they are a fixture of any capitalist economy with a financial market. (I should probably extend that to communist economies as well, because China recently had a real estate bubble, which led to real problems.)

They’re characteristic of economies, but they don’t have to be damaging if you can contain the damage to the financial system. If you can ensure that the systemic risk potentially caused by bubble investment is controlled, then it’s not necessarily a deep problem. Some investors will lose some money. That’s what happens with speculation. It’s bad for them. I don’t want to dismiss that hurt that a lot of people can feel. But it doesn’t bring down the rest of the economy and hit a bunch of innocent bystanders.

RA: If this is a bubble and if it bursts, who are the people who get hurt the most? We often talk about how we’re in an economy that feels like it’s making the rich richer and leaving the middle class stagnant over the past few years. Would a bursting of this bubble hurt mostly the rich? Or are there trickle-down effects that are also worrying?

JB: It’s both. The investors with the most to lose from these very high valuations of AI companies tend to come from the top of the income scale, of course. While stock market ownership is more dispersed than is commonly thought, with about 50 percent of households holding some stock, something like 90 percent of the market is held by the top 10 percent. If the market takes a 10, 15, 20 percent dive, then the heaviest investors will take the hit. 

However, they have the wealth to just hold on and come back. If you have the capital and wealth buffer to sustain the market losses from the bubble bursting, you’ll be okay. I feel your pain, but not nearly as much as the pain of those who are hurt by a recession, if that’s what the bubble generates. 

Your question really comes down to who is hurt by an economic downturn disproportionately. That is, of course, people in the bottom half. They’re the ones who tend to have higher unemployment rates. They have much less of an income buffer. I’m particularly concerned about them because the current administration has been really whacking away at the safety net and, if anything, making life more expensive for people in the bottom half. They’re the least able to take the blow.

RA: If you were still advising the president of the United States, what regulations would you push for to protect industry and businesses from any of the fallout that could happen? What should the government be doing?

JB: I’m pretty anti-interventionist when it comes to advice regarding specific companies. What I’ve always tried to do in my government work is to put in the adequate guardrails so financial markets don’t get over their skis and implode regularly, which unfortunately has been the case in recent years.

The dot-com bubble and the housing bubble were a function of precisely this dynamic of systemic risk, of excessive speculation infecting the rest of the economy. It’s why I argued we should pass Dodd-Frank when I was in government. We did. And it helped a lot. The main reason it helped is it forced companies to have more of a capital buffer so they would be insulated from wealth losses. These high valuations could evaporate without having to go to the government for a bailout, demanding resources from taxpayers, and wrecking the rest of the economy. So my advice would be to make sure the financial guardrails are maintained.

Unfortunately, they’re going the other direction, particularly in the area of crypto. The Genius Act and the incentive to speculate on stable coins and crypto is taking down guardrails. And so that’s where I would be nervous. I wouldn’t interfere in the Mag 7 directly. But I would carefully watch this debt story you raised a minute ago. Initially, they were funding most of their investments from cash flow. But more recently, some of that funding is also coming from debt, some of which is opaque and coming from private credit. So we don’t have eyes on it as we’d like to. 

RA: Let’s talk about the rest of the economy. Whether or not we’re in a bubble, are investments in AI crowding out growth in other areas? One parallel with the dot-com bubble is that in the late 1990s, it became more expensive to raise money for non-tech businesses. Do you have worries about the rest of the economy?

JB: The increase over the past year in returns for the S&P 500 is about 17 percent, which is very strong. If you look at what’s called the equal-weighted S&P—which discounts the very heavy $20 trillion market cap of the Mag 7 (and excludes privately traded firms like OpenAI)—you’ll find a return that’s about 7 percent. That isn’t terrible, but you’d have to multiply that by more than two to get to the weighted return. That’s a complicated way of saying that there’s the S&P 493 and the S&P 7. The S&P 7 is banging out huge historical returns, and the rest of the companies are just in a normal place. That’s not so bad.

I don’t think that the crowd-out is glaringly obvious. The price of capital is not that high. Interest rates are up somewhat for a variety of reasons, including the economic policies of the current administration, which have a lot of investors nervous and lead to a risk premium in the interest rate. But I don’t see companies having a great deal of problems raising the revenue they need. 

RA: I’ll keep broadening this out. What is your sense of how much tariffs are impacting inflation so far? And how much worse could it get?

JB: Some very good economists at Goldman Sachs have quantified that. They probably have some of the most careful models for measuring consumer pass-through, which is how much of the tariff is hitting consumers. Remember, there are all kinds of links in this chain, from exporter to consumer. There’s the exporter, there’s the shipper, there’s the importer, there’s the wholesaler, there’s the retailer. And then there’s a poor consumer at the end of the chain taking out her credit card to purchase something. Because she’s at the chain, the consumer takes the brunt of this. But thus far, according to estimates that look right to me, pass-through has been maybe 50 or 60 cents on the dollar.

The inflation indicator that’s most germane to the Federal Reserve is running at about 3 percent. It would probably be running at 2.5 percent without the tariffs. But that pass-through is not complete. There’s more to come. Firms have buffered some of the pass-through. Why is it 50 or 60 instead of 70, 80 or 90 cents on the dollar? Inventory front-running: A lot of firms built up their inventories to front-run the tariffs. And some margin compression: They took some of the hits out of profits because they don’t want to lose market share by raising prices too quickly. So those things look pretty played out. By the end of the year, I think tariffs will add about a point to inflation; I expect it to be about 3 percent instead of 2 percent. So that’s not through the roof, but it’s pretty bad.

RA: Let’s talk about the shutdown. What is your sense of the real damage to the economy with shutdowns in general? And with this one in particular, if it’s resolved in another week, what would we lose?

JB: We lose maybe $20 billion in GDP, but we gain it back pretty quickly. So if a family was going to go to a national park, but the park is closed, they suspend their vacation but don’t cancel it. More to the point, when salaries aren’t paid, you’ve got pent-up demand, so people consume those resources on the other side of the shutdown. Typically, that kind of bounce back has occurred.

But as you say, this shutdown is different and more worrisome, particularly because the administration is threatening maybe not to pay people back. I hope that’s an idle threat. That would be illegal. More lastingly would be the idea that they’re firing people instead of temporarily furloughing them. And if you really have hundreds of thousands of people losing their jobs during this shutdown, that would lead to higher unemployment on the other end in a way that we haven’t seen in prior shutdowns.

RA: When you were last in government, you were one of the architects of what became known as Bidenomics, often described as building from the bottom-up and the middle-out. Looking back now, why do you think that message didn’t land at the polls?

JB: These things come down to many factors, of course. The election was very unusual with the president dropping out 100 days before. And so there’s a political side of this.

But if I stay in my economics lane, I can answer your question pretty tersely with two words: price levels. We were just talking about inflation. I said it would be a real problem if inflation were 3% instead of 2%. And that’s true. But from the perspective of consumers, what matters most is affordability. This is a huge problem with the American public right now. Tariffs exacerbate this problem by making things more expensive. So you have an issue there.

For us, inflation peaked at 9%. It came down to two and a half percent by the time we were leaving. But nobody wanted to hear about lower inflation. They wanted lower prices. And that difference is often elusive to people. What’s the difference between inflation and the level of prices? Inflation is how much prices grow. So if you remember that eggs used to cost $2.50 a dozen and then they cost $7, and I tell you, “But inflation’s coming down so they’ll cost $7.20 next week instead of $7.50,” I’ve lost you. You want to know, “When am I getting my $2.50 back?” It’s the price level. It’s remembering what things used to cost and, damn it, wanting those prices back. 

Meanwhile you have this candidate, Trump, demonstrably peddling falsehoods in promising to lower prices. Just to take one example: coffee. Because of the 50 percent tariff on Brazil, coffee is up some 20 percent over the past year. That’s the increase in the price of coffee. Nobody seems to be bugging these people about that. I am scratching my head as to why, given the depth of the affordability crisis. Our fingerprints were partially on the increase in inflation. It was largely supply chain snarl-ups due to the pandemic. But that was a minority share. This coffee price, you know, that’s their work. Their fingerprints are all over that. Take that tariff away, that price comes down. And yet, I see very little focus on that. And it’s quite disconcerting.

RA: A significant difference between presidents Biden and Trump is how they’ve approached clean energy. Trump is clearly a “drill, baby, drill” president. What is your sense of how much of the Biden agenda will be undone because of that? And what does that mean for competition with China, which is expanding its lead on solar and wind and electric vehicles?

JB: First of all, let’s talk about electricity prices, which are rising at about 7 percent, so more than twice the overall rate of inflation.

RA: And some of that is because of AI just to bring it full circle.

JB: Good point. 

The core point of your question is that the country that builds out electricity capacity to supply that side of the equation will have a huge edge in the AI competition. And by cutting yourself off at the knees regarding renewable energy, we’re ceding that fight to China. It’s destructive. And it’s not clear why the Trump White House takes the position it does. Yes, they want to drill. I get that; frankly, we did, too. It was probably underreported, but we had record production of fossil fuels. I’m not saying that proudly, given my green sympathies, but I do believe in an all-of-the-above approach to increasing energy supply. We had record production of both fossil and carbon-based energy and renewables. If you plotted them in terms of their percentage increase on a scale, they both sat right on top of each other. 

Now Trump is really winding back many of the Inflation Reduction Act tax credits and efforts to build out our production capacity in renewables. And that is highly problematic.

RA: One critique I had of the Biden administration policies was that if given the choice between competing with China and fixing clean energy as a problem globally, the administration often chose competition with China. Obviously, there are national security reasons as well. 

JB: I agree with your critique.

China has surpassed us by far in terms of producing electric vehicles at a much lower cost than we do. Now, a huge amount of that is government subsidy. But it’s also getting a jump on production technology, particularly battery technology. And I thought we should, instead of trying to keep those cars out, have a controlled quota with caps. Import those EVs and reverse engineer them. Frankly, under the radar, some of that is taking place. There was a joint venture with Ford and a Chinese battery company to try what I just described.

I think there’s a way to engage in more cooperation on behalf of our environmental future and climate control. Particularly around technology exchange. Battery technology is a great example. I don’t think we get very far in a clean-energy revolution until we significantly improve our storage capacity, and batteries are key there. Wind, solar, all of that depends on storage and batteries. China’s ahead of us there, and we should learn from them.