Opinion | We Can Survive an A.I. Bust

If you’re the first one to predict a bubble, you probably were wrong because it went up a whole lot before it went down. And so getting the timing of these is just much, much harder than knowing that eventually there probably will be one. Is the A.I. economy a bubble? Are we just in the early moments of a technological revolution? “Nvidia’s stock price exploded from $14 in 2022 to over $180 today, turning early believers into millionaires.” “An artificial intelligence boom pushes investors to pour money into the stock market.” Or are we overextended and headed for a crash? And if we are watching a bubble inflate right now, what should the government— or for that matter, the individual investor— do about it? My guest today has a lot of bubble experience. He was an economic advisor in the Clinton White House when the dot-com boom imploded. And then he worked for President Obama in the aftermath of the housing bubble and the crash of 2008. Today, he’s a Harvard economist and a contributing writer for New York Times Opinion. Jason Furman, welcome to Interesting Times. Great to be with you. So, Jason, we’re going to talk about the economy as a whole and even the long-term future of American growth. But I want to start in the place where a lot of Americans’ thoughts about the economy start, which is with my own stock market portfolio. I had this moment a few months ago when OpenAI rolled out its latest version of ChatGPT, and there was a general disappointed reaction. And you had people who had been A.I. skeptics finally claiming vindication, saying, “Look, the A.I. revolution is not delivering as fast as was hoped.” And in that moment, I went in and looked at my very boring mix of mutual funds and index funds, looked at just how fast it had risen in the era of the A.I. boom and thought to myself, “Man, it seems like I’m invested in what might be a bubble.” And a lot of Americans have had this thought I think. In the months and weeks since then, you’ve had a lot of talk about how large A.I. and A.I. related companies loom in the stock market, and I want to start there. Just how big a part of growth right now is artificial intelligence and everything associated with it. A.I. is an enormous part of our macro economy right now, and there’s two ways to think about the economy. One is demand. That’s the amount of stuff you’re buying. The other is supply. How much you’re producing. Right now, A.I. is really showing up on the demand side. It’s building those data centers, buying those microchips. And by my estimate, in the first two quarters of this year, 92 percent of the increase in demand in the U.S. economy was due to just two categories in GDP— one called information processing systems and the other called software. Ultimately, though, what we’re really hoping for is that A.I. shows up on the supply side of the economy, actually helping us do more with less. And certainly some of that is happening. But there hasn’t been anything particularly special about productivity growth to date. So just to stay on that, the point you just made about how much of growth and expansion is driven by A.I. Now, that doesn’t mean that absent A.I., data center investments and everything else, economic growth would be zero, right? Yeah, that’s right. I don’t think we’d have as much growth absent A.I., but we’d have a different type of growth. So a bunch of that A.I. demand is coming in the form of imports because those chips are being produced abroad, which don’t add to GDP. Absent the A.I. boom, I would expect that we would have lower interest rates right now because so much capital is being siphoned up by the enormous demands in this sector. If that weren’t happening, there’d be a little bit less pressure on the economy. The Fed would be cutting rates faster. And what does that mean? We’d have more activity in other sectors like home building and manufacturing. So the A.I. boom is partly adding to the economy and partly crowding out other activities. And just as a really wild, rough guess, that might be 50/50. So half of growth due to the A.I. boom. And then the other part of the A.I. boom is crowding out what would have been other sources of growth that we could have had if we didn’t have this. And then something similar is happening in the stock market. Can you tell listeners, what are the Magnificent Seven? The Magnificent Seven are seven enormous tech companies, companies like Amazon and Microsoft that represent a very large fraction at this point of the S&P 500 and a very large, very large fraction of the increase in the S&P 500. So if one is wondering why the stock market has risen, you’re largely talking about why are the Mag Seven worth so much more today than they were a year and a half ago. And why are they worth so much more today than they were a year and a half ago. It’s because of all this expectation of what their profits will be in the future. Now there’s some differences between the different companies. A company like Apple is a little bit closer to buying a utility, where every year they’re going to come out with a new phone. Every year they’re going to sell a bunch of those phones and you buy it and you’re just guaranteeing that cash. So there’s a little bit less expectation of growth there. But for a company like Meta, a lot of their value is based on the expectation that they’re going to figure out something about A.I. and that they will also figure out how to make a profit out of that something from A.I. To justify these valuations need really both of those both the breakthroughs will happen, but also that the breakthroughs will generate a profit. But all of these companies generally are themselves profitable. The companies that look more like classic startups are the companies that are just doing A.I. research and development. And most of those are not publicly traded. So you’re not looking at ridiculously high stock market valuations for these companies. But they do have extremely high valuations in some sense. Can you talk a little bit about those companies. Yeah OpenAI, for example, is worth hundreds of billions of dollars, but it’s pretty hard to buy a share in it. It’s not publicly traded. That’s just the valuation you get when they get new investors from venture and elsewhere. One sense that’s enormous. I mean, Goldman Sachs has been built up over a century, and OpenAI, which was built up over the last decade, is more valuable than Goldman Sachs. On the other hand, 10 percent of the world is using OpenAI is using ChatGPT. And that’s just amazing. Five years ago, basically 0 percent of the world was using it, and now 10 percent of the world is using it. Now, a much smaller percentage of the world is actually paying for ChatGPT. And that’s the big question is 1 is 6, 7 and 8. Going to be a continued things like iPhones have become where there’s small design changes and the camera gets better. Or are going to be, profound changes in the way that 3 and 4 were. We don’t know the answer to that. And then second, and this gets back to the issue I was talking about before. It’s not just that you need the innovation, it’s that you also need to be able to profit from the innovation. And if large language models become like a commodity where there’s three of them that are absolutely amazing, but they’re basically all the same, it’s very hard to price a commodity at something higher than the marginal cost of delivering that service. And so they won’t make a profit. They won’t justify their valuations if they can’t figure out how to build what’s often called a moat, a way of delivering something unique, sticky people, once they’re in the Apple ecosystem, they tend to stay in the Apple ecosystem. Is it going to be the case that if you’re in the ChatGPT ecosystem, you can stay there. How much of this I know this is the impossible question, right. For the economist. But how much of this is just vibes. Psychology definitely is a big driver in markets. There’s something enormous going on here. This is not pets.com. Pets.com was part of the bubble. It had barely any revenue, barely any business plan. And yet had a pretty high valuation. These are companies that already have quite high revenue, enormous numbers of customers, enormous upside. And then how do you quantify all of this. Well, you add the genuinely large thing with some vibes. And maybe that’s where we are today. And in terms of historical comparisons, is there any model from American history or modern history of a single form of technological investment playing this kind of role in the economy or in an economic expansion. So on the demand side, railroads were probably larger at various points in time in the 19th century, where just enormous amounts of the economy was being devoted to laying track as well as building railroads. We’re at roughly a similar scale of where we were in the late 1990s and early 2000, with the fiber build out broadband. That was an actual real economy, demand side activity. And there’s probably some other things in between that have been at similar scale. So we’ve often had, whether it’s the automobile or electricity or airplanes or the personal computer. One thing that was playing a disproportionate role in many of those cases, that one thing did turn out to be a bubble, but not in every one of those cases. Well, and the pattern. So with railroads and I guess you could say the same thing with the internet, right. The pattern seemed to be there was a bubble in the sense that people got overextended building out infrastructure before it had a path to profitability. But in the end, they were right to build the infrastructure. Yeah So by the way, on railroad, you actually want to say railroad bubbles because it burst over and over again in the United States, in the UK and continental Europe. And my guess is in lots of other countries around the world. So they just kept making the same mistake of overbuilding track, maybe building duplicative track. But yeah, when it all settled out, the railroad was enormously transformative. The internet was enormously transformative. Radio is another thing, which was a bubble that was enormously transformative. And I don’t have an exact tally, but I think there probably have been more bubbles of things that were actually real, actually big, than these just totally fantastical, unproductive ones. O.K, so play play bubbles advocate for me right now. If you wanted to make the case that this is what we’re looking at right now, that A.I. is a railroad style productive bubble where the tech is real, but we’re just overinvested and overbuilt, what would that argument look like. Well, first I’d look at the market as a whole. And Robert Shiller, who won a Nobel Prize for his work on ways in which markets could turn irrational, developed a concept called the cyclically adjusted price earnings ratio, or CP. The Shiller Cape right now stands at about 40, which says the price of a stock is 40 times the inflation adjusted average earnings over the last decade. That 40 is the second highest that the Shiller measure has ever gotten, and it goes back about 150 years. The first highest was where it got in early 2000 right before the tech bubble burst. So the basic standard, first thing that financial market people, economists look at to assess the value of the stock market right now is screaming that it is sky high in a way that has never lasted before. So that would be number one. Number two would then be to dig into certain companies and go through just what would have to happen to justify their valuations. And if it’s a really small startup to say their revenue is going to double every year for the next decade. Fine, that definitely happened sometimes. But when you already are a big established company and you’re being priced a little bit more like a startup what’s the plausibility of that. When it requires both the technology to work and you to need to figure out how to profit from that technology. You mentioned earlier that you don’t think we’re seeing in productivity data and other statistics, evidence that A.I. uptake, the use of A.I., programming or whatever else is having a fundamentally transformative effect yet yet and the yet is a really important part. If you’re a business and you go out and hire 20 people to figure out how to integrate A.I. into your small business or medium sized business or large business, whatever it is. And they’re all out there trying to figure out how the chain of stores that you run or the chain of restaurants or something can use I the people you hire. If they don’t figure it out right away, they actually show up in the data as lower productivity, because you basically have more people working in that business, and it’s not producing a higher output. Now, that doesn’t mean it’s a mistake to hire those people. They may well figure it out. And five years from now, you can replace all sorts of people or get all sorts of higher profits or whatever it is, and the productivity will show up. But economics, this is called a J curve, where sometimes you go down before you go up. And I think that’s happening in some companies right now that in a sense, A.I. is actually reducing their productivity because they’re busy figuring out how to use it, but they haven’t yet figured out how to use it. So in some sense, it’s not that surprising to me that we’re not seeing the productivity growth from A.I. yet. I do expect that we’ll see some but it is an open question as to how much. We’ll see. And that would be, again, in the argument that for a bubble, right, you would say that if you have overextension, extraordinary overextension of investment, and you’re in the downward part of the Jay curve, that makes a bubble scenario more likely. Yeah, I think that would be in the case for a bubble. And productivity isn’t the only thing that matters. We actually had more productivity growth from 2000 to 2005 than we did from 1995 to 2000. So even after the bubble burst, even after this investment was collapsing, productivity growth was actually very, very strong. It just wasn’t nearly strong enough to justify the way in which those companies were valued in the year 2000. I should also say, as an economist, productivity growth actually is almost everything I care about. It tells you what the size of your economy is. It tells you, on average, what wages will be or possibility for the future. So to me, that’s what I’m most focused about and care the most about. But definitely for the stock market, it’s just one input. And what about just to be more anecdotal, there’s been a fair amount of coverage in the last few weeks of these deals where effectively the A.I. companies are paying each other and increasing each other’s valuations through deals with one another, where one company agrees to buy another company’s chips and in return it gets I guess, shares in that company. I may be misrepresenting this slightly, but get shares in that company, and then its decision to purchase chips from that company drives that company’s share price higher. So the money that it uses to buy the chips, it’s effectively getting from the increase in the share price of the company it’s investing in. Does that to you seem like the kind of thing that happens in bubble environments where it’s companies, hyping each other up. Or is that more, I guess, more just what you would see normally in an environment where a bunch of companies are working together closely and are growing quickly. So like everything here, unfortunately there’s two sides to this, and I wish I could come down for you firmly on one side. No, you don’t. I’m going to ask you for the case. The case against a bubble in a moment. On this being a bubble, it’s the opacity of these arrangements that would make one the most nervous. And also, I guess, the circularity of them. There was an old phrase that in a gold rush, the way you could guarantee a profit is being the person that sold the picks and shovels to the miners. And the idea was the person went off to find gold. Maybe they found it and got rich. Maybe they found nothing and ended up poor, but you were guaranteed money if you sold them the picks and shovels. Well, right now, instead of selling them the picks and shovels, you’re in some sense lending them the picks and shovels and telling them that you’ll be repaid if they actually strike gold. So NVIDIA would be the one with the picks and shovels, and OpenAI would be the one going off looking for gold. Now NVIDIA is making is making the chips. Yeah, NVIDIA is making the chips. That’s like a real actual thing. It’s like a pick and a shovel, but a little bit more sophisticated and complicated to make. And if they were selling them all for basically cash, you’d say they’re pocketing that money. But in some sense they’re now not just selling them for cash. It’s essentially almost as if they’re lending them to OpenAI, and they’ll get paid back and they’ll get paid back with multiples if A.I., OpenAI succeeds, but if it doesn’t, then they won’t get any money or won’t get as much money as they would have gotten for selling those picks and shovels. So it’s so in the gold rush economy, even if there’s less gold at Sutter’s Mill or in the Klondike or wherever else than people thought, at least if you’re an investor, the pick and shovel money is going to prop you up. Whereas here, if there isn’t, if there isn’t enough gold out there, your investment in the pick and shovel company is also in deep trouble. Yeah and this is something that’s changed. Six months ago, I’d say NVIDIA was the pick and shovel company that was guaranteed to lock something in. And now that’s changed. And they’re not getting all the money up front for selling those picks and shovels to people. And then the second part of this is just the opacity of it. We have in our economy different ways. For a company to get money. One is you sell sell a bond and bondholders buy it. That’s a way of lending you money. A second is you go to a bank, and the bank is super careful about who they’ll lend money to because they’re incredibly highly regulated. And the third is you go to what are sometimes called shadow banks. These are companies like Apollo, and they lend you money, often with fewer questions asked. They themselves face less regulation. And a lot of the lending that’s happening in this sector is happening with companies like Apollo that are shadow banks that are less regulated. Now, to date, these are enormously profitable, enormously successful companies. They’re incredibly sophisticated. I would for the most part, bet on them knowing what they’re doing. But one has to be just a little bit more nervous about them. O.K, now argue the other side. Tell me why this is not at all like the railroad bubbles or boom. Why should we not be alarmed about the Shiller index being almost as high as it’s ever been. The biggest reason why I have, frankly, full disclosure, kept all of my money in broadly diversified index funds and haven’t reduced my exposure. So in some sense, we were going to come around to the personal investment question. So that’s good to know. We can come around to that. So in some sense, that’s how I’ve answered this question for myself, is, I think, a lot about a speech that Alan Greenspan made in December 1996, where he said there was irrational exuberance in the market. There was a lot of reason to think that the market was pretty frothy and pretty bubbly. And what happened after he gave that speech. The stock market ended up doubling over the next three plus years, and then the bubble burst. But if you had bought stocks when Alan Greenspan made that remark, and then you live through the bursting of the bubble and sold at the very bottom of the broad market, you still would have made money. And that type of pattern has repeated it over and over again throughout history that people thought something was a bubble, went up a whole lot before going down. And it turns out if you call a bubble but you’re early, that’s not very impressive. That actually means that you are wrong. That’s very different from almost anything else. Anything else. You predict it and you’re the first one to predict it. You should get lots of credit if you’re the first one to predict a bubble. You probably were wrong because it went up a whole lot before it went down. And so getting the timing of these is just much, much harder than knowing that eventually there probably will be one. And so in that example you’ve mentioned already, pets.com, which stands as the paradigmatic example of the overvalued internet company, though honestly, the one I remember most from my college days is kozmo.com, which I think promised that was the delivery one. It was a delivery one. And that one was awesome while it lasted. They were just giving you almost free stuff. Well, I mean, what’s striking about those companies is that, of course, we now have very profitable delivery companies, and we now have very profitable, pet supply companies online, right. So in the long arc of history, there were good ideas there. But yeah, kozmo.com it was like, well, we’re going to deliver pizza. Oh, you don’t want pizza. Well, we’ll deliver ice cream, we’ll deliver. I remember getting one of the Harry Potter books from them at midnight the day it was released. Oh, well, so there you go. So you were the benefit and beneficiary of that kind of overexcitement. But to me, I feel like we don’t yet have the pets.com and the Cosmo coms of this moment. Again, if I’m trying to talk myself into the idea that it’s too early to call a bubble, you have these big, well-established, extremely profitable companies that are investing heavily in this, but which are not dependent right now on A.I. for all their profits. And then you have a set of companies that are not yet profitable and maybe never will be profitable, but their revenue is going up. The A.I. companies are on a positive trajectory overall, would you say. Oh Yeah, they’re on a positive trajectory and they’re real companies with real business models. Now, there are hundreds of others. Many of them have not gone public yet. So-called unicorns that are worth over $1 billion, that don’t necessarily have business models that people are in the venture industry, are still throwing money at them, and I’m sure dozens of those, if not hundreds of them, will end up imploding. But you can have lots and lots of failures along the way, and it not be a bubble. For it to be a bubble, you need the successes to make up for the failures. And it wasn’t just that pets.com failed, it was that the successes, at least in the short run, did not make up for the failures in 2000. So you were in the Clinton White House then, right. Yeah, I was What were you what were you doing. I was doing two things simultaneously, one of which was private. And I can be really proud of one of which was public and is totally embarrassing. So why don’t I talk about them in reverse order. We were planning a new economy conference. That’s what it was called back then, to celebrate just how awesome and transformative everything was in this new economy, and we ended up in late February announcing that we were going to hold this conference in April. I just went back and reread the fact sheet for this announcement, and I think I probably wrote much of that fact sheet, and it was just going on about what a high fraction of economic growth was coming from this one sector. And isn’t it amazing. Just look at the disproportionate investment growth, the disproportionate GDP growth. And it’s all coming from the information technology sector. Isn’t that super cool. So we announced the conference in late February. The bubble started to burst about two weeks after we announced it. And by the time we held it in April, I think the market was down something like NASDAQ at least, maybe one sixth from its high. And in one respect, that was embarrassing in retrospect that we were celebrating a new economy, even as on the financial side, it was imploding in another sense, though, we were right in that productivity growth over the next five years was even higher than it was in the five years leading up to that. And so in some sense, it was a new economy, just not nearly as exciting as what financial markets thought. So that was the public thing. In private, the Clinton administration was running a project that my memory is that it was called Project nirvana in an optimist in an ironic type of way, to try to understand what would happen if all of this fell apart and if the stock market fell was one scenario. An investment collapse was another scenario. And to see what we needed to do to be prepared. My memory is that we didn’t actually ever get to a real answer, but we largely reassured ourselves that even if the bubble burst, the macro consequences would not be very large. And so we didn’t actually prepare anything, to my knowledge, other than just to reassure ourselves that even the downside was not so bad. And I think that was a good exercise that we went through. And largely was vindicated by what ended up being a pretty shallow recession when the bubble burst. What do you think will happen if this is a bubble and it does burst or leak or diminish. So we’ve had two bubbles that burst in the last 25 years. One was the tech bubble and the consequences were relatively mild. The housing bubble was massively devastating to the economy because it wasn’t just that house prices went down, which mattered for families and consumers. It was also that all the mortgage debt had been thought to be completely, 100 percent safe and was used as collateral in other lending throughout the financial system. And then all of a sudden you found out it wasn’t completely safe, and a lot of other parts of the financial system experienced problems and runs. So which one is this. One most of me thinks it’s more like where if this bubble bursts, it’s just the stock market goes down. People spend less, some businesses invest less. You have a recession, but it’s not a particularly terrible one. The one piece of it that makes me nervous, though, is these companies like Apollo that we were talking about before. I’d like to say that if one of these big shadow banks failed, we would just let it fail in a way that we’re willing to let most hedge funds fail. But maybe we can’t. Maybe they’re too systemic. Maybe they spread their effects throughout the economy. And if all of a sudden you were in a position where you were facing systemic risk because of a largely unregulated financial institution that was doing hundreds of billions of dollars of lending. If this ends up being a problem, that’s where it is. But partially, how we come out of it then depends on just the underlying health of the economy. And there’s a lot of different indicators in the air, right. And I feel like people give you different assessments of the underlying state of the economy based on jobs numbers one day and consumer sentiment the next day and manufacturing sentiment the next day. And there’s a lot of disagreement with indicators pointing in a lot of different directions. What’s your assessment of the underlying economic situation. Yeah, so there have been two broad disconnects in the economic data this year. One was for a while the so-called hard indicators where you looked at what a business actually did were much better than the soft indicators where you looked at consumer confidence or business uncertainty or what businesses said their plans were. That was true earlier this year, this disconnect between the hard data and the soft data. And so far, it looks like the hard data was actually vindicated, that all of the complaints about uncertainty, all of the lack of confidence, was not something that seems to have translated into decisions that businesses and consumers made. It was more like mood music around politics. So it’s like a pollster comes to you. What do you feel about the world. Oh, it’s all going terrible. I hate it, but then you go to the store and spend just as much as what you were spending before. So I’ve devalued, frankly, how much emphasis I place on the importance of uncertainty, which never was very central to the way I thought about economic policy. But as it may be even a little bit less now, the current confusion in the data and I should say we haven’t really gotten any government data since the beginning of October, since because of the government shutdown. But the current confusion in the data is that the labor market is slowing quite rapidly, but GDP is growing quite quickly, and you see that in the unemployment rate, which keeps ticking up. You see that in the hiring rate, where businesses are hiring at a rate that you would normally associate with recessions. You see that in the number of jobs added per month, where we’re averaging less than 50,000 jobs a month, and it used to be 200,000 jobs a month. So a pretty big deterioration in the labor market. At the same time, the economy in the third quarter, the latest GDP tracking estimate is for growth above 3 percent and consumer spending is growing. Businesses are continuing to invest. So everything that goes into GDP looks quite strong. Does so I want. I want to talk about Trump administration policy and how it might affect those numbers. But just on that point, do you think immigration policy has any effect on any of these numbers. Certainly on employment numbers. I think you’ve had Trump administration officials saying, well, you wouldn’t expect rapid job growth in an environment where we’re deporting people and lots of people are self-deporting relative to the Biden era. What do you make of those kind of theories. Immigration is having a large effect on the economy. A couple of years ago, we were getting a few million people into the country a year. Now we don’t have a really good estimate of net immigration, but maybe it’s 0, maybe it’s 500,000. And that means you’re going to have slower job growth. You’re going to have slower economic growth, everything else being equal. And in some sense, I wish there were more people in favor of immigration restrictions. That just made the honest argument that our goal isn’t to have the highest job growth, it isn’t to have the highest economic growth. To work better for people here in the United States. And, we’re going to sacrifice on those other metrics in order to do that. Just to be clear, by the way, I’m not sure I even agree with that argument, but that is A.I. think that’s internally coherent one. Yeah I think you’ve heard that argument from some people in the Trump administration or people arguing for their policies. I think obviously that’s an argument they’re more likely to make in an environment where the job numbers are not great. If the job numbers were great, they would just they would just take credit for it. What about tariffs. You wrote an essay for the times for us about arguing basically that economists to some degree got the Trump tariffs wrong. Not in the sense that they’ve proven that they were a stroke of genius. But in the sense that there was an expectation that they would be totally disastrous. And they haven’t been. Where do you think that argument stands now. It stands roughly where it stood when I wrote that essay in July. I have no doubt that tariffs have been a negative for GDP growth, a negative for employment growth and a positive for inflation, a positive for inflation also being a bad thing, something you’d rather not have. But all of those were maybe plausibly been on the order of something like half a point rather than something catastrophic. And just to gauge things, the economy normally grows at about 2 percent a year. And so if you do this, then it grows at 1.5 percent a year. No, I think that’s a pretty bad unforced error. Works out to be about $1,000 for every household. And that’s not a cost that as a policymaker, I would want to impose on every household. But it’s also not like the history books are going to remember. The year growth was 1 and 1/2 instead of two. What about the extent to which the Trump administration seems to have tried to basically exempt or shield artificial intelligence from the impact of tariffs because that seems to be part of the story here as well, that there is this special zone for A.I. where the protectionist agenda doesn’t apply in the same way. Do you think that’s part of the story. Yeah, that’s certainly part of the story is Yeah tariff rates are lower than they were originally announced in April, and they’re much lower on things like microchips. There was going to be an expectation of under a national security procedure called Section 232, that there would basically be across the board tariffs on microchips. They were expected to be at least 25 percent and that hasn’t happened. So they’re betting on this sector. And it’s interesting to me, if you look at the original defense of tariffs, a lot of it centered around restoring American manufacturing and American manufacturing jobs. It’s completely failed, at least so far, and it’s obviously still very early. Manufacturing jobs are down now. You’re seeing a lot of defense of the tariffs. Some of it is revenue raising. Which part of me warms my heart when I hear Republicans getting excited about tax increases, but wasn’t exactly where it started. Surcharges we’re calling them surcharges, Jason. Surcharges I apologize, not taxes. Just edges. In the Clinton administration, we had a plan to raise tobacco taxes. And we also called that a tobacco surcharge or something like that. So I’m not going to get too moralistic about Orwellian language when it comes to the tax code. But the second argument that you’re hearing is that the tariffs are leading to all this inbound investment into the United States. All these companies and countries are committing to invest more in the United States. It’s still too early to figure out whether those promises are real or fake. But what’s notable is if they come about it might even mean that the trade deficit widens. So what started out as a plan to lower American trade deficit and revive manufacturing is turning into a plan to increase investment into the United States, raising the trade deficit. And, by the way, taking the main manufactured thing and importing more of it rather than building it here in the United States, that being microchips. So what will the Trump administration do in the event that the stock market starts going down. Well, the first thing that will happen is everything that they criticized previous administrations for doing, both Fed and treasuries. You’ll see them doing them because in a crisis, you do all sorts of things that you’d rather not do in terms of buying assets or lending to certain parts of the economy and that thing, just to keep stuff afloat. And so I doubt some of the purity that you’ve heard. For example, from Secretary Scott Bessent, who’s talked about and criticized the big expansion of the Fed’s role in the past. Well, a lot of that expansion happened in crises, and I expect that he would be a pragmatist, not an ideologue in a crisis. And I’m sure Donald Trump would be. In fact, we’ve seen him be a pragmatist, not an ideologue, in dealing with the financial aspects of COVID in 2020. Second, though, I worry about the government is directly inserting itself into the sector in advance. They’re buying equity stakes in these companies, or getting things that are almost like the equivalent of an equity stake, where they get a fraction of all of NVIDIA’s export revenue. And that, in some sense says to me they are more willing to do things for individual companies than you’ve seen in the past. Now Bush and Obama basically bailed out the auto industry. The theory at the time was in the middle of a severe economic crisis. What was basically a liquidity crisis that was temporary could turn into a solvency crisis and then spread throughout the economy. And be very costly. I think ex-post, they were largely justified in bailing that industry out, and much of the money that went into it, but not all of it was repaid here. In 2000, we didn’t do that. Global crossing went bankrupt. No one in the government tried to do anything about the firm Global Crossing, which is a firm building a lot of the fiber optic cable and switching networks and stuff like that. My guess is that 2000 approach would be the right one here. I could see this administration wanting to be much more interventionist, especially for its favored companies, and getting involved with them directly in a way that gets in the way of what capitalism and bubbles bursting are supposed to accomplish. Well, so Yeah, and this is a theory that I’ve been kicking around for a little while. It’s based in part on what you describe, the Trump administration’s eagerness to be in partnerships and take ownership stakes and so on. But it’s also based on the extent to which the vision of the A.I. future is deeply entangled with issues of national security. When you do exercises and thought experiments about accelerated A.I. timelines, including people we’ve had on this show, they tend to very quickly turn into arguments about the new Cold War with China. And I feel like when you put those pieces together, plus the fact that the Trump administration has made such a bet on A.I., all of that makes me feel like from the point of view of the administration, maybe these companies are already too big to fail. So absolutely, I think you do need to inject national security into this conversation in a way that you didn’t really need to. In the bursting of the 2000 or bubble or the housing bubble. And there are things that are economically costly that you want to do for national security reasons. In fact, just in normal times, for example, I would subsidize chip companies to make more advanced microchips in the United States, not because I think it’s a great source of jobs in the United States, not because I think we’re good at making the microchips, just because it terrifies me that almost all of our advanced microchips come today from Taiwan. So there’s all sorts of costs, economic costs you want to pay for national security. I’d also say one more thing here, which is that you want to be very targeted. So Intel got a subsidy under the CHIPS program from the Biden administration. And that law was passed with a decent number of Republican votes as well to build microchips in the United States. To me, that made sense for national security, but was a cost economically. This administration has taken an equity stake in Intel. And the problem is Intel does lots and lots of things, and an equity stake isn’t very well targeted to that one activity that they’re not doing enough of right now, which is building microchips in the United States. So it’s not just that we want Intel, we want Intel doing certain things, and we should be willing to pay transparently for those certain things. And I think we have less of that well thought out targeting now than we had before. And by the way, it was very far from perfect before. All right. Let’s do a couple of big picture questions before I let you go. Both about both that are connected to the unknowable aspects of the A.I. phenomenon. So it seems to me that when I look at the future trajectory of growth in the United States or Europe or the entire developed world, there’s a lot of underlying reasons to be pretty pessimistic. That you have these societies that are aging, that have sometimes extremely low birth rates, that are running permanently large deficits. And in that environment, sometimes it seems like we’re betting so heavily on A.I. because there isn’t any other bet to make. And the alternative is just depressing. If this technology doesn’t achieve the kind of takeoff that people imagine, then you’re just headed for a kind of global experience like Japan from the 1990s on, of stagnation and old age. Are you optimistic about the long term future of the American economy. Otherwise? I’m reasonably optimistic. I mean, in some sense. Why are you optimistic. We have had over the last 50 years, nearly 2 percent annual productivity growth. So every year you basically figure it out how to do 2 percent more with a given amount of labor than you did a year before. Now, that’s not as high as what we had in the 1950s and 1960s when we were coming off all of the World War II innovations. But that’s pretty cool and impressive that no matter how advanced stuff gets that, we keep figuring out how to squeeze out more. I might be add on top of that might not even add on top of that. It might be how we keep getting to 2 percent each year, but that accumulates up and makes a difference over time. Second, though, is my optimism about the U.S. economy does rise and fall a lot with what we do on immigration. Immigration matters both for your labor force and how many people you have to work, which is an increasingly a challenge for the United States were it not for immigration. And not every country can overcome unfavorable demographics, but we really can with this. And then immigration also matters for that 2 percent number, that productivity growth. Because a lot of that innovation, a lot of the companies we’re talking about are, founded by, staffed by, run by first or second generation immigrants. So that to me, a lot of our future does rise and fall with immigration. But I grant you, I would love to see the world have more of a growing population than it has now. And I’d feel better about the next 200 years if I knew either that population was growing, or that the robots could do everything else, and we could just sit around and chat with each other on podcasts, on podcasts. Well, so let’s talk about that as the last as the last question that kind of scenario. This has been a conversation about what normal economic models and normal historical economic experiences can tell us about the short term likelihood that we’re in a bubble, what policymakers can do, and so on. But one of the bizarre things about A.I. is just this fundamental unknowability in the tech right now. That just doesn’t seem to me to have been the case with the transcontinental railroad or even with the internet. Derek Thompson, who is a fellow podcaster and a writer on all these subjects. He was going through these arguments the other day and he said he’s talking about pets.com, right. And he says, did you know that pets.com delivers food, drinks, and toys. That was a real line from an actual pets.com ad. So it wasn’t just Cosmo pets was delivering it too. But he says what they didn’t say was, did you know that pets.com will solve Alzheimer’s, invent nuclear fusion, and loosen the icy grip of inevitable cellular death. And it’s very easy for me to get guests on this podcast who will make predictions like that about A.I., along with the prediction that it might well kill us all. As an economist looking at this landscape, do you just have to set all that to one side and just narrowly focus on the numbers in front of you and say, we have to assume this is a normal technology until it starts to act like one that isn’t. How do you put that into models of our situation. Well, I’ll tell you how I handle it in a class that I co-teach with David Laibson. It’s called ec10. It’s the principles of economics class. And last year in our last class, we taught the students exactly the answer to this question, and we gave them exactly the right answer to this question, which is to say, I told them the future would look like the past and we could extrapolate. David Laibson told them it would be totally different from the past. And one of us was correct. Do you want to know which one was correct. I mean, Yes, I suppose I do. My argument is, we’ve had some big transitions. For example, in the 19th century when almost everyone was working, when the majority of people were working on farms. Be impossible to imagine what life would be like in a world where only a few people were able to grow enough food for everyone. And largely the reason the unemployment rate has stayed the same is several things. We create new types of jobs we couldn’t have imagined. We have more demand for old types of jobs. A lot more people eat out at restaurants now than did 100 years ago, because we’re richer. Radiologists, A.I. is helping them now, but there’s more radiologists than there were ever before. It’s not turning into a radiologist because you need to talk to the patient, coordinate with the other doctor, figure out what model to use, and that finally, and this is the scary part of it, new technologies can change relative wages. So if it partly substitutes for what people do, often they keep their jobs by their relative wages falling. And that’s part of what we’ve seen, the increase in inequality over the last many decades. So all of those forces have kept it so that at any point in time, about 96 percent of the people who want to work can work. And that’s been true most every year for a very long time now, except in a recession or something dysfunctional, largely, if you’ve seen something for 100 years. My first bet would be to assume it happens for the next 100. I a set of stories under which the future will be different from the past. A lot of them in the labor market go under the story of horses, where every argument I just said applied 100 years ago, but yet now 300 horses would have as much horsepower as your car. But I don’t think even for free, you would take 300 horses over a car. In some sense, I have some children who would make that deal, but I would advise them against it. So that’s the other side of the story, is sometimes things are discontinuous, but there are so many powerful economic forces that have worked for a long time that I’m going to still emphasize them in my teaching, even if my co-teacher is more visionary and feels otherwise. All right. Well, on that note, I think you have to go teach that class. So Jason Furman, thank you for joining me. Thanks for having me.