AI, markets, and power: A conversation with Paul Krugman (2025 re-run)

0:00In early 2025, I hosted the legendary economist and Nobel laureate Paul Krugman on my show. We never released it in the podcast feed, but it really is too relevant to leave in the archives, so we're sharing it now. In that conversation, Paul and I covered the most critical economic questions raised by artificial intelligence. We explored whether today's tech and AI valuations look more like a genuine productivity boom or a replay of the late 1990s. We investigated why the US is pulling away from Europe, whether AI might substitute for globalization and what that might mean for emerging economies.

0:36And we asked what happens to workers and communities as this technology transition unfolds. Listening back, it's a really sharp guide to the economic backdrop for everything we now discuss on Exponential View. Enjoy. We need to talk about the economy in 2025 and what's going on. I know you've got opinions. So why don't you start with your view of where we are in 2025, a sort of short-term horizon, headwinds or tailwinds, and then I'll respond to that. Okay. So, I mean, we are very close to a Goldilocks economy in the United States. Unemployment is low historically. Inflation is low historically.

1:21It's a little bit above So we have an official target of 2 % in patient, and if you look at probably the better measures, it's more like 2.5. And if you think that's really important, maybe you should have a drink or two and calm down. We're really in good shape. It's starting point. This is about as good a moment as we've seen for a very, very long time. And the question is, There's always going to be something. Some stuff happens, as the bumper stickers don't quite say. And there's really two kinds of stuff that can happen. One of them is politics. And, you know, I'm reading whether there's now, it's possible that my president is about to impose 25 % tariffs on Canada and Mexico tonight, or then again, maybe not.

2:17And that's a big deal. And we have a bunch of potential political shocks to the system. I've just been doing some work. And if you look at stock valuations, it really has that sort of end of 1999 feeling. And are we going to be facing a sort of financial bubble burst? It's, you know, and so they really calm, good times never last. And in this case, I think it's unusual that we have at least two obvious threats to the times we're in. But all of this would take time to unwind. So the numbers will continue to look good for the next few months, at least. Yeah, I mean, it's interesting that we head into this with such a strong American economy.

3:08I'm sitting here in North London and there's a particularly depressing series of charts which show OECD growth rate these are the club of rich countries and you have the US as part of it and the US is just motoring away right it's the biggest economy there it's growing much faster far faster than the UK where essentially the economy is roughly the same has been the same size for you know a decade or more and and you know Europe at some cases up until about a week ago, it almost, the mood felt funereal. But I think that there has been a change in the last few days. The bit that I query, wonder about is this issue of the tech stock valuations.

3:51I mean, we're definitely at a point where this magnificent seven, I love the way how the investment bank has always come up with these wonderful terms. So the mag seven are highly concentrated. In fact, This is the highest level of stock market concentration since before the Great Depression. And of course, stock analysts love to look at things like this because history rhymes in a sense. But it does feel that there is something that is different this time around. I mean, for one thing, there is revenue momentum in a lot of these companies compared to the dot-com bubble where companies had valuations without revenue.

4:31The second is the world economy is much richer and much more complex than it was in 99 or in 1929. And so when I look at that, I wonder, of course, animal spirits can upset stock market valuations as they did on Monday with DeepSeek. And if stock market valuations drop, they can spill into the real world through wealth effects or other dynamics. But it doesn't quite feel as overinflated as as 99 because we can see the the revenue flow into these businesses. OK, I would actually mildly disagree with you on that, because in 1999, we think about the conspicuously sui's. We think about pets.com and so on.

5:19But, you know, that bubble was in large part telecoms companies rather than dot coms. And the telecoms companies had real revenue and just not enough to justify the valuations. And the technological, the technology actually did deliver, maybe not on quite the ethical scale that people had hoped for, but we really did get a decade of substantially accelerated productivity growth out of the technology. So it was real stuff. What happened was that the idea that early movers were going to be able to, as we now say, build a moat around their position, that they were going to be able to earn sustained high profits, turned out to be wrong.

6:06And in many ways, we look similar now. I think we're past the point of wondering whether AI is actually going to do anything useful. We still don't know how much useful it's going to do, but it's clearly something real. And there's real revenues coming in. But on the other hand, our valuations, the valuations are far, far above anything that you would normally expect from those revenues. So the valuations depend on the belief that these are going to lead to entrenched positions for the existing companies. Yeah, well, I think that's really true. And in fact, that is probably what we saw on Monday, right?

6:44When this Chinese model DeepSeek, it wasn't released. I mean, DeepSeek R1 had been around for a while. They declared they were doing it weeks, if not months ago. I had written about DeepSeek back in December 23. And in December 2024, I said, DeepSeek is a Chinese Sputnik moment for the US. So that was before the new year. because I thought that a lot of these stock market valuations have embedded within them the idea that the US will have a sustained AI dominance. And sustained AI dominance will mean the best technology turning into the best product, turning into the fastest revenue growth, dragging forward with them the supply chain.

7:32So effectively, the demand for NVIDIA chips, ultimately AMD chips, and the manufacturing. and that that bet was so concentrated that, of course, any little wobble will have animal spirits running through that China shop. Yeah. And not just that the U.S. will continue to dominate, but that the MAG-7 will continue to dominate and that there won't be some contender, U.S. contender, European contender. You know, it's worth remembering that although European macroeconomics doesn't look too good, But there's a whole lot of, it's not as if the technological sophistication is restricted to the United States.

8:14And so, and it's really hard. And some of it is, some of it is just, I mean, I suppose we need to talk about Tesla at some point. And that is really bizarre. I mean, their actual business doesn't seem to be doing too well. But people have decided that somehow or other there's magic associated maybe with the political partnership. it. But I take the lesson of the dot-com telecom bubble of the 90s to be that even a really great technology does not necessarily justify really high valuations. Right. Okay. Okay. I hear that. I'm going to stick on with this point because here's where I think today is different to the dot-com.

9:02And I'll stop by telling you a story. So back in 97, 98, I was building internet companies. And one of the companies that I invested in was a realtor, online realtor in France. Of course, online real estate is a huge business now, right? You have Zillow and Redfin and a whole bunch of others. When we got into that market, not only did only 4 % or 5 % of French people have internet access, they were still largely on the Minitel system. But that 5 % wasn't far advanced in other parts of Western Europe. None of the realtors had PCs. And so when we went off to, or the founders went off to win business, they had to buy PCs for the realtors, train them on how to get onto the internet and how to upload their images and their inventory.

9:55I mean, it was a complete nightmare. The difference today is that every company or the large proportion of companies, particularly big one, are sitting on a digital stack, which they've built over 20 years. All those billions of fees they paid to Accenture and PwC and SAP to do digital transformations means they've got a digital infrastructure. And today it takes a single decision in Microsoft headquarters to put an AI tool into every copy of Microsoft Excel because it's all run on the cloud. And that feels really different to 99. It feels like, There's a whole set of hurdles that now don't exist that would allow a much more rapid deployment.

10:39And with rapid deployment should come much faster increase in revenues. Well, this does, yes, it does mean much faster increase in revenues. I think it's very clear that the whole AI thing is moving much faster. I mean, we thought things were moving fast in the 90s, but they're much faster. the adoption rates have just been, you know, the S curve of adoption is just far steeper. The question is going to be revenue for whom. Right. And that's, now what is true, and I think is, which I'm, I think I'm learning to some extent from you right now, is that the fact that this stuff in some ways sits on top of existing technologies and existing networks may actually be a reason to think that the incumbents can actually continue to capture a lot of this.

11:34I mean, when we say something like Microsoft is, it's all about network externalities. Everybody uses Microsoft products because everybody uses Microsoft products. And the interesting thing about AI is in a way how undisruptive it is that it goes, you know, it can be just put on top. I spend a lot of time turning these things off because I don't want it. But, you know, Copilot is right there, and, you know, you're still, after all these years, using Word and Excel, which is the network externalities and the moat, but it's not some upstart competitor bringing AI to word processing or number crunching.

12:22it's quite straightforward for Microsoft to actually go right in there and do it itself. So that might be a difference. Now, it may be that we're just not thinking big enough and that there's something just radically different will break through. But it is true that right now it's looking as if this is, in terms of market share, kind of less disruptive than some past technologies. Yeah, I feel that that's right. Like, you know, Gemini from Google and Microsoft are all doing reasonable well. I mean, we do have the new players. But what's interesting about the configuration of the market is that each of the new players, like OpenAI and Anthropic and others, are really closely embedded with the last generation's giants.

13:07And it's not as if the Ford Motor Company or AT &T was closely allied so strategically with the firms that went before. So it feels like this industrial transition has different characteristics. And perhaps it is just that bits are bits and you're building on bits. And so what else can we expect at this time? I would love to talk also about something that I think is quite close to your heart that is, I think, quite difficult to navigate right now, which is what should we think about what happens to the workforce under the speed of this kind of AI transformation? And I will start with a little sort of pitch of how I see the state of the nation, which is that companies are really, really eager to make productivity improvements.

14:04And these tools can automate large parts of roles. They can automate many, many tasks. And so the natural outcome for workers will be either suppression of wages or it will be loss of jobs on the flip side of that productivity gain. And the question is not whether new jobs will get created, because historically we've always created new jobs. And we've seen work from some of your colleagues in the field of academic economics to show that most jobs in the U.S. are in categories that didn't exist 60 or 70 years ago. But I suppose the question that I have is, that I wonder about, is what is the likelihood that the economy will create enough new jobs in those categories for there not to be some kind of schismatic dislocation as this technology rolls out?

15:00How have you thought about that? I think there's quite a lot of dislocation coming, although that's not new. I mean, that has been the case for, with every major change in every major technological revolution, that a lot of jobs are destroyed and a lot of jobs are created. And in the end, one way or another, there ends up being full employment because that's more or less, you know, mass unemployment due to automation has been, you know, people have been predicting that for a very, very long time and it never happens. but the transition, a lot of people can find themselves in the wrong job, in the wrong place, sometimes geographically.

15:45I mean, it just, it feels to me like it was just a year or two ago that it was, you know, we were telling people learn to code and now it turns out that coding is one of those things that AI does apparently. I have no personal experience of it but apparently does pretty well. It does pretty well. And, you know, the way I've been thinking about it. So one of the sort of put down remarks that people make about large language models, and I'm not quite sure if it applies fully to everything else, but that it was just souped up autocorrect, which is. But the thing is, there were, first of all, you could say that something like agricultural machinery is just souped up guy with a scythe cutting down wheat, which didn't stop it from being hugely disruptive.

16:32I mean, that we basically have, you know, we have fewer, the United States is a major agricultural exporter and we have fewer farmers than we have people playing World of Warcraft. So a technology can be sort of not all that magical and still have enormous impacts and destroy a lot of traditional jobs. And in this case, souped up autocorrect, well, even if that's all it is, but very souped up, a lot of people's jobs are basically souped up autocorrect. And those are, in many cases, jobs that we considered highly skilled. So a lot of things in middle management, a lot of, I mean, I don't know if there's a way to get this, but things that I've noticed is that translation, translation software is not perfect, but damn good.

17:25And how many people were displaced by the fact that you don't actually need somebody who who speaks Mandarin. How many, the, yeah, the, just a lot of, a lot of the jobs that we, we think, we like to imagine, and maybe we can talk about this, we like to imagine that the very highest creativity level stuff can't be automated, although that may come, that may come as a rude shock at some point, but, but a lot of stuff that's just a few rungs below that can go away. I think in some ways the safest jobs are the ones that involve manual labor and dealing with the material world. Right. Well, it's fascinating because 15 years ago that the story was AI was going to come off the routine jobs in offices.

18:16And what we have seen is this souped up autocorrect. It doesn't really matter how it works in theory. If it's working in practice and doing much more than that, it will have an impact. is enabling lawyers and software developers and creatives in all sorts of areas. And I think that that has become quite a surprise. I think Goldman Sachs did some work last year, and they were showing that the pinpoint where jobs really started to be impacted by effective LLNs was about a$100 ,000 a year salary, which is well above the average. So we end up in this quite interesting world. And I agree with you that we may feel that there are certain jobs that are very high polluted and creative, and the machines won't be able to at least support or augment them.

19:12But with every release of an AI system, those systems get better and better. And I think that that does create a real pressure and attention. I mean, later today, OpenAI will release their next model, which is called O03. And I think quite a lot about what policy responses to this ought to be, given that there is so much uncertainty and given that it's really expensive to run a safety net. We know that in Europe, our tax levels are much higher. We have a much deeper safety net. And that, to me, feels like a kind of politically quite a difficult problem to close down. Nobody really wants higher taxes, whether it's here or in Europe or in the U.S.

19:59And certainly the politics in the U.S. for the next three or four years won't support it. So that, I think, is going to be a really key point of tension, which is what happens if we do start to see really mass job displacement? Well, I mean, my point of view is that since we don't know which jobs will be lost and we don't really know how to devise policies, Honestly, we could do some things here and there, but trying to, efforts to deliberately promote the jobs of the future have gone rather spectacularly wrong in many cases. So safety nets is mostly what you have. Now, I have an American perspective where our safety net is extremely threadbare compared with Europe and our taxes are quite low.

20:43And we could certainly do more of that. Now, what it doesn't do, even if you can have a safety net which protects people from real immiseration from the technology, you can't restore, you can't, it doesn't restore communities. You know, if you have a community that is based around an industry that ceases to exist because it's either because of globalization, which is what mostly is the focus, but also just technology or even changes in tastes. I mean, I like to talk about the detachable collar and cuffed industry of Troy, New York, right? That just went away. And when it does, you can have a safety net that ensures that nobody serves or goes without health care.

21:29Maybe not in America, you can't, but anyway. But you can't restore the community. And you can't restore necessarily the dignity. And that's, I don't have an answer for that. I mean, you do what you can, but we are going to be seeing a lot. I just want to key on one point you said. The technology may not, we may not be at a point where it replaces people of, you know, exceptional creativity or whatever, but it does enhance their reach. Yeah, absolutely. And we've been seeing that for quite a while. I mean, you know, I still have some ties in academia, which is a poor model, but I think it may have something to do with.

22:13I think it's also true in things like the legal profession, which is that the very top people can actually do a lot, can basically cover a lot more bases because of technology, which actually makes things more hierarchical. I mean, I used to joke about a couple of fields that I followed that we were now at a stage where there were only three people in the top 10. And that's a, again, it's an unequalizing effect of the technology to which I have no answer. No, I think we're also going to see that there is in my other hat, which is to invest in startups and particularly deep technology and AI startups.

22:53There is a mood towards technology. a belief that you could build the billion dollar single person company, like the one man unicorn, as they call it. And the idea is that there's a whole array of ordinary and increasingly complex tasks that can be delegated to tens of thousands of AI agents, and you've swapped labor for capital, and you're just paying a rental to a cloud service for that. And I think that that is a direction of travel that we've seen. The number of people employed even by a huge company like Facebook is far smaller than was ever employed by General Motors and so on and so forth.

23:37I know that we're going to go to questions in a few minutes. One thing I would love to also for us to talk about is trade and tariffs. You know, we've got a man in the White House who loves tariffs and certainly loves the threat of tariffs. And we also have this geopolitical fragmentation. Now, as a technologist, what I see happening is a number of different things. We've seen from the political side the building up of more and more walls around the Internet. So it's not just China's great firewall. It's also, you know, Russia being able to seal off its Internet. There is now this political pressure, which is around free speech in the US and a departure from European standards.

24:22And of course, in technologies with the export controls, most importantly, in a way, forcing China to start to do its own development in really hard technologies and advanced semiconductors. So I see a world where there will be increasingly a couple of lanes of technology. and that, of course, reduces the size of markets and adds quite a lot of friction to all of it. It feels like that's a general drag on growth. Yeah, it is a general drag on growth. Now, a little bit funny, the whole tariffs on everything, that is, if you want to understand that, I think you need a model of the mind of Donald Trump.

25:05There really isn't very much of a constituency behind that, except that he wants it. I really see business kind of hates it. But economic nationalism, more sophisticated economic nationalism, is definitely on the rise. I mean, the Biden administration was far more economically nationalist than any American administration we've seen for decades. Europe is clearly moving somewhat in that direction. And the reason is mainly, I think at some level, the world is a scarier place. We now are, you know, who thought that large steel conventional warfare was going to make a comeback in the 21st century?

25:49And now it has. And now you think that, well, we need to have capabilities, either domestically or in our close allies, that mean that we are not, can't be shut out, that we have the ability. And if possible, we're going to try and deny this stuff to people that we don't consider our friends. And that is going to fragment the world. And also, by the way, worth saying, and I know we need to get questions. I'm the one who said that. One of the things that is also true is that, to an important extent, technology can actually substitute for globalization. If you need customer service, instead of getting somebody in Bangalore on the line, you instead are talking to a chatbot.

26:32And that's a real change in dimension. So we may be in it for a much more fragmented world, both because it's less necessary to have global trade and because we're much more afraid, with good reason, of being too dependent upon other countries. And what I worry about is the small countries who may be left out of the cold. And so, you know, the EU is going to be fine. I think Britain will reach some kind of accommodation with the EU. Britain will be fine. America will be fine. um bangladesh vietnam yeah not so clear well i i i love this line technology is a substitute for globalization i'm going to come back to you after we we finish this live on that in in my my first book i talk about the coming fragmentation that was going to be driven by changing technology and i identified this idea although i didn't have quite the pity communication that you have technology is a substitute for globalization, that there are technologies that will do the job that globalization used to do.

27:41And that would be things like 3D printing or vertical farming that allow localized production, particularly of food. But also, I think critically, the change in the energy system. So you buy solar panels once, they last for 25 years, you buy batteries, they last for much longer than we ever thought. And you don't need to be a vassal to a sort of a petro state. And that, I think, does create an interesting path for countries like Bangladesh or Pakistan or Vietnam, where a lot of their sense of sovereignty and ability to develop is about their access to energy, right? How much energy capture do they have as an economy?

28:23How many energy services can they actually use? And in the last two years in Pakistan has been this remarkable change driven by the falling cost of solar panels, which has meant Pakistan has become one of the largest importer of solar panels and broke the link between measured grid delivered electricity and GDP growth because businesses could not be bothered to sit behind brownouts and blackouts and went off and bought solar panels to keep their businesses running. And I think that there is this possibility where the declining cost of this technology can provide a much more affordable energy sovereignty for the long tail of the world that is not energy sovereign, especially smaller and poorer countries.

29:10And the first kilowatt hour of consumption a family has in a day from zero is the one that delivers the water pump, the refrigeration, the lights, the charging of the smartphones and could make a big deal. Now, of course, that is me, Paul, as an absolute relentless optimist about the human condition and where we can go. But it is sort of one bright spot that I look at. Oh, the whole thing, yes. Just to say that the renewable energy revolution is the most hopeful thing of today's world. And it's funny how people get all, you know, people get excited about IT, which I understand, but in terms of actually really changing people's, changing the prospects for development, changing the prospects for the planet, solar, wind, batteries are probably the most important thing.

30:02And they've really had Moore's Law type productivity improvements. So it's awesome. A quick note. If you want to support us in bringing more of these conversations to the world, please consider subscribing to the show. The 2025 Edelman Trust Barometer, so trust barometer run by a big public relations firm, shows a high level of grievance worldwide, particularly against the wealthy in governments. How can markets traditionally quite poor at addressing inequality tackle this grievance given the potential negative consequences? Well, that's a doozy. Over to you, Paul. Oh, but this is not, markets are not going to solve this.

30:40This is, let me just say, you know, we had, I grew up, I'm somewhat older here. I grew up in what was really a middle-class society. At least it felt like a middle-class society. There was poverty. There were some people, wealthy people, though not the way there are now, but it was mostly middle class. And I had assumed that that was a society that sort of evolved naturally. Actually, we kind of thought that was an end state of economic development. But it turns out it wasn't. Obviously, inequality soared again, back to Gilded Age levels. But it also turns out to be that that society, both in the U.S.