Full Transcript: Edward Chancellor on AI, Capital Cycles, and Gold

Financial Historian on Bubbles, Anti-Bubbles, and the Price of Time

Kai: Hi, everyone. Welcome to the Intangible Economy. Our guest today is Edward Chancellor, who is a financial historian, journalist, and investment strategist. He is the bestselling author of Devil Take the Hindmost: A History of Financial Speculation, one of the classic books on investment bubbles. At least in my opinion, he’s also the world’s foremost authority on capital cycles, having written two books on the topic. In addition, he’s the author of The Price of Time: The Real Story of Interest, and most recently helped our former boss, Jeremy Grantham, write his autobiography. Ed, welcome to the show.

Edward: Nice to see you, Kai. What you didn’t mention is that you came to GMO as my analyst back in 2008. So we go back to the financial crisis together, didn’t we?

Kai: Yeah, and the other piece being that, to add to your many accomplishments, helping advise me on my economics thesis on credit cycles and bubbles at Harvard.

Edward: And if I’m allowed to blow our trumpet together, do you remember how we were working in early 2009, and I got you to do a piece of research that showed that quality stocks that GMO was heavily invested in at the time tended to deliver alpha or outperformance during — during bust periods, simply because they had a lower beta to the market. In which case the answer was that, when you wanted to get out of — when you thought the bust was coming to an end — you wanted to get out of quality as quickly as possible, which we actually delivered that research in February ‘09, just in a couple of weeks before the market turned.

It turned... It was that... And I credit you because you actually redefined quality in the various markets that we looked at historically as, as volatility or low-vol stocks, ‘cause we didn’t have data for, you know, full quality back then. But still, when I think of the sort of good pieces of research that I’ve been involved in in my life, I’d say that was definitely, you know, one of the top five pieces.

Kai: Very timely, and yeah, it was definitely a pleasure working with you at GMO. And so that’s why I’m so excited to have you on this podcast today, to go through some of your research and your work.

Edward: Okay. Let’s move ahead.

Kai: Let’s start. So the topic I wanted to start with is, I guess, the topic of the day. US big tech companies are investing trillions of dollars into AI data centers in what is set to perhaps be the largest investment infrastructure boom in history. Setting aside any view on the technology itself, what does the history of capital cycles — from the railway mania to the Japan bubble to the dot-com boom in the late ‘90s — teach us about how the current cycle may play out?

Edward: Well, it’s hard to leave out the efficacy of the technology, so we’re gonna have to get back to that in a minute. But the general principle, looking at past technology booms — and as you know, I wrote a paper with Jeremy Grantham on this in February this year, which is available on the GMO website — the general picture is a new technology arrives, people get very excited about the impact of that new technology.

Sometimes the new technology doesn’t attract too much attention in its early years. I’m thinking for instance of, you know, when the railways came to Britain in the 1820s. 1825 was the launch of the first passenger railway in the UK, which was called Stockton and Darlington. And the first few railway companies had dominant positions, no competition. They were pretty profitable, and their technology was proven. And then we had two successive waves of investment, one a decade later, sort of 1835, ‘36. That led to a boom in the stock market. A bit of overbuilding came down, but it wasn’t too much damage.

The real problem or mania came in sort of 1843 to 1845. And that is really a period in which there was a massive launch of many, many railway schemes across Britain. In terms of projected capital expenditure, I think it was running to about 10% of UK GDP. So actually much higher than AI today. And there were too many — as a result, not all the schemes went ahead. But the upshot was far too much duplicative investment. And, you know, famously, as I would say, you know, three railway lines between London and Peterborough, which is in East Anglia, three railway lines between Leeds and Manchester. And obviously, if you have three lines running between two places, they’re gonna be less profitable than if you have one line. And the upshot of that is that the railway index, I think, lost about, you know, 60% of its value.

And ironically, I just looked at this the other day — canal stocks, canals were the most obvious losers from the railway mania, and they did lose in the end. But actually, you did better investing in canal stocks between 1845 and 1850 than you did in railways, because canal stocks had already been beaten down a bit, whereas the railways still had plenty to fall. Now look, in the very long run — let’s say, you know, 20, 30 years — that investment was pretty benign for the economy.

And, you know, although the railways were never quite as profitable as they’d been in their early years, it was good for the economy, and if you got into railway stocks in 1850 once the bubble had burst, it was perfectly fine. Now then, we have other sorts of new technologies that attract much more competition. I mean, just think motorcars in the late 19th century. I think the US had something like 2,000 motorcar companies. And the upshot of that was, I mean, General Motors, which was a winner, had to be recapitalized twice. And actually, GM was really a sort of roll-up of failing car companies.

And Henry Ford — I think Ford only succeeded on his third attempt. You know, it was hugely over-invested area. Aircraft was much the same. I think — I mean, Warren Buffett points out that there were actually three aircraft companies in Nebraska, believe it or not, in Omaha, Nebraska. Far too many aircraft companies were founded. Very loss-making. And that pattern has really continued, obviously, into the dot-com era, where, again, massive CapEx. And what’s interesting about what we used to call the TMT bubbles — technology, media, and telecommunications — is that the CapEx is largely done by telecoms companies, and they did the heavy spending, but they really weren’t the winners.

And so one of the points I made in a recent column of mine was that actually markets have no trouble investing tons of money during these technology transitions. But they do have trouble spotting the winners. And then as you know, after the dot-com bubble bursts, NASDAQ loses — say, 78 to 79% of its value. Amazon goes down more than 90%. So even though Amazon emerges as an eventual winner, there is a massive loss. And it’s dangerous to say, in hindsight, that you can spot Amazon as the winner, because there could have been other businesses around — Webvan, this and that — that might have actually taken the prime place.

So anyhow, that’s the general picture. The general picture is a new technology, everyone gets very excited about it, a huge amount of investment. And investors tend to anticipate the profits flowing more quickly than they actually turn out to be the case. And there has always been, I think there’s always been a shakeout.

Possibly the one exception is the arrival of the telephone — another sort of revolutionary technology where the Bell telephone system sort of began to insert itself and got a pretty early monopoly. Telegraph also moved to monopoly very quickly. So you could say that if you’ve got a revolutionary new technology, and for reasons of inherent monopolistic...

Kai: What do you wanna call it? Sort of...

Edward: ...environment, there is a case that the new technology can arrive without a huge overinvestment. But if the barriers to entry are relatively low, and therefore you can have more than one dominant player, then historically you have had overinvestment. And that’s all we know. We can’t say that the future will resemble the past, but it should sort of guide our judgment, at least. And then one can assess current circumstances and say, to what extent is this time different or not?

Kai: Right. So if I’m hearing you correctly, you are saying that over the course of history there have been many successive waves of capital cycles which generally start due to the advent of a new technology, which of course attracts investment capital. In a few cases, such as the telephone, the industry has consolidated into monopoly pretty early and has been relatively stable. But the vast majority of instances end up with an influx of capital and a fragmented market structure whereby profits are competed away to zero, and that leaves investors in these companies with a pretty rough outcome.

Edward: Well — I’ll add something. You remember one of the sort of key precepts of capital cycle theory is that it draws on the prisoner’s dilemma game-theoretical problem. In the prisoner’s dilemma you’ve got two prisoners, and the question is, do they keep quiet and both serve a moderate sentence? Or do they both rat on each other and get very long sentences? In other words, it’s suboptimal to rat. But the way the prisoner’s dilemma game is structured, there’s an inducement for both to rat. You don’t benefit by not ratting.

And I think it’s the same, you know, during a capital cycle — during a new investment boom — is you see other people going into this new area. And if you don’t go in, there’s a possibility that they might come out the dominant monopoly and crush you. And so you go in. But if you go in, the end result may be suboptimal for, let’s say, the investment world as a whole. You may actually hang on in there. So it’s not necessarily irrational for the individual — it may be a sort of agency problem, but it’s not necessarily irrational.

And I think with AI, you know, there’s one narrative. I don’t know how robust it is, but one narrative is that when OpenAI came out with ChatGPT in November ‘22, Microsoft saw this as a way to team up with OpenAI to break open the Google search monopoly. And then Google felt it had to respond. And then of course everyone else was standing by the wayside saying, “This is a great technology. We want a bit of the action too.” So of course we’re gonna get a huge CapEx boom.

Kai: Yeah. I mean, it’s interesting to think about how the different big tech companies have responded differently to the game theory, right? Because we have Apple, of course, as a famous example of a company that has largely abstained from the arms race, and it’s kind of waiting to see how things shake out. Whereas amongst the hyperscalers, you see ever-increasing investment and quotes from CEOs being like, “I’d rather go bankrupt than lose this race.” So there’s clearly a dynamic happening amongst at least some subset. And maybe your point is just that you only need a few players to be bought into the idea in order to drive the entire cycle.

Edward: Yeah. And you may have been following this case more closely than I have, but it’s not just the largest-cap US tech companies doing it — there’s a whole load of other competition. You’ve got DeepSeek in China. They’ve listed a few Chinese AI companies recently, and then you’ve got a vast amount of VC capital going into the same space. So the big tech CapEx into AI gets most of the coverage, and I suppose in absolute terms it’s the largest section, but there’s a lot of other investment going on at the same time.

Kai: Right. So I guess the key question then is: we know that there’s a lot of supply coming online, a lot of investment coming into the AI sector. The question is, will there be enough demand? Will there be enough demand to meet supply? How should we think about this side of the equation?

Edward: So it’s often the case, or it has been the case, in these tech booms that people overestimate demand. I mean, going back to the railways in 1845, the CapEx spending at that time would’ve required — someone’s got to crunch the numbers — passenger and rail traffic to increase by threefold over the next five years. And given that there were a fair number of railways already in the UK by that time, it wasn’t gonna happen.

During the dot-com bubble there was this urban legend going around that data traffic was doubling every two months, when in fact it was only doubling every six months. And that little factoid originated with some company that was later taken over by WorldCom, which later went bust. And it was found absolutely everywhere — all the brokers picked it up. Even the US government picked it up. So everyone believed it.

But in fact we actually had data. There’s this guy I know called Andrew Odlyzko, who was at the time at Bell Labs, and in 2000, just around the time that the tech bubble was peaking, he put out a paper giving the true demand growth. And no one paid attention to it. The actual, accurate data was available in real time, and no one paid attention to it. And the upshot was WorldCom went bust, and a host of those other so-called alternative telecoms carriers — altnets — went bust, and there was massive overcapacity in fiber optic cable.

All the telecoms equipment suppliers like Nortel and Ericsson and Lucent took big hits. And actually there was a massive decline in profitability. So one of the features of the CapEx booms is that they actually produce profits, because if someone invests and the buyer doesn’t actually immediately depreciate what they’ve acquired, then aggregate profits rise. And so what you see in the late 1990s going into 2000 is a massive surge in reported profitability. And then because that capital turns out to be misallocated, new CapEx is immediately curtailed, and you have to depreciate past CapEx. And so you have a collapse in profitability.

And something very similar is happening today, in that the depreciation schedules for these AI chips — the GPUs — has been extended. I think from roughly an average of three to three and a half years to six to six and a half years. And I understand that if you buy a GPU and keep it in a warehouse because you haven’t actually built your data center yet, you don’t actually start depreciating the GPU until it’s actually in the warehouse. But there is a sort of technological depreciation going on even before you actually start using the chip.

So yeah, we’ll see. But the market is being driven, as far as I can see, by a strong economy on the back of a lot of CapEx and very strong earnings growth. But those are contingent on the investment turning out to be profitable and the demand being there.

Now, can I just say — going back to the demand question — there is this view put out that AI is... we’re on the cusp of artificial general intelligence, even singularity, and that this AI, whatever that might be, will be able to do almost every conceivable human function apart from merely physical work, like plumbing. And I’m very, very skeptical of that view, because — when one’s talking about AI as the large language models — large language models work through inference, through probabilities of what is the most likely next token to pull out. I can’t see, A, how that can lead to really genuinely original thinking or activity. And then B, as you know, sometimes it’s gonna make the wrong selection, and therefore you’re gonna get the hallucinations. And I think the hallucinations — because these machines aren’t genuine reasoning models — are inherent to the large language model technology.

And if that’s the case, then I really don’t think that the demand that is mooted currently is gonna be there. And I don’t know if you picked this up — did you see, last week or so, a story about some company that provided software for car rental businesses? They used Claude for vibe coding. They used Claude to update their software, and it wiped out their entire customer database. And it did some other quirky things too. So the whole business crashed, and the companies that were using the software — the car rental companies — were suddenly frozen.

Kai: Hmm.

Edward: And a couple of months ago, the winner of the Royal Society’s Faraday Prize — the top scientific prize — a guy called Michael Wooldridge, gave the Faraday Lecture, which I think is worth listening to. He’s making these sorts of points that I’ve made — in other words, I’ve just taken my quotes from him. And he then thinks there might be what he calls a Hindenburg moment. The Hindenburg moment was — I don’t know when it was, 1937 — when one of these zeppelins, everyone was excited by these zeppelins, the Hindenburg zeppelin blew up somewhere over some US airfield, and everyone had second thoughts about the technology.

So I may have this wrong, but as far as I can understand, the CapEx into large language models is not simply about improving search and helping with research, but actually is posited on a great breakthrough into agentic AI. And as I say, we’ll see. It’s not uncommon for investors — or, I mean, investors both in companies and in markets — to imagine that technology is more advanced than it actually is. I mean, if you think about it, a lot of those businesses from the dot-com bubble would’ve been viable had we had faster internet connection. But at the time, people were largely working on incredibly slow dial-up connections. The businesses that would’ve been viable 10 years later were not viable then.

So yeah, I don’t know what’s gonna happen to AI development. Maybe someone merges the large language model technology with the reinforcement learning technology, and maybe they can harmonize them and get something much better. But at the moment, they don’t seem to be quite there.

Kai: So Ed, you believe that these models can be error-prone, and that will limit the total addressable market. I guess where I would... I think an interesting analogy would be that of, like, self-driving cars, right? So we know, for example, that Waymo is not perfect, and its error rate is not 0%, but so too are human drivers imperfect. And one of the interesting phenomena we’ve seen is that for whatever reason, politicians and drivers have a bias against AI, whereby the tolerance threshold is so much lower. If there’s just one big headline about a bad Waymo accident, suddenly the interest in the technology wanes significantly.

And so with your Hindenburg moment — is that kind of what you’re saying here? That we all know that AI is not perfect, we all know it’s a nascent technology. The AI bulls will say it’s gonna be getting better over time, this is the worst it’ll ever be, and it’s kind of hard to argue with that. How much better it will become, we don’t know. But you’re saying that there is a tail risk for the AI thesis around if just one really bad thing happens, that could potentially put a pause on AI development.

Edward: No. I’m being a bit more adamant. Namely that, given the problems of hallucinations — and I looked at one report from, I think, October of last year, and found the best performer was 2% hallucinations, and I think the near-20th performer was, say, around 20. You’re not going to... There are certain activities where you cannot tolerate that degree of error. You won’t — you can’t.

Kai: Yeah. So I think what that means is that certain things you would not want to give AI. Like, you don’t wanna make AI in charge of the nuclear weapons, right? But in the case of, say, doing R&D and doing research into a completely greenfield technology where it’s like the alternative is to do nothing, then I think the bar is much lower, right? So then the question becomes: what percentage of economic activity falls into category one versus category two? Because that will constrain the total addressable market of a technology like this.

And what I guess is being implied is that when the demand forecasts are being written by the AI companies, they’re to an extent baking in both categories. And you’re saying maybe only one category is actually available, in which case the potential demand curve is a little bit less enticing than is currently being priced by the market.

Edward: Yeah. Or, put it simply, the ruling principle of Menlo Park is fake it till you make it. And the amount of hype — by definition, all these technology manias have large doses of hype. And in many cases, that hype over the long run is, you know, validated. But there’s really never, ever been as much hype as we see around AI. And frankly, if you take the amount of hype relative to the proven efficacy of the technology, that ratio is just more extreme than anything before.

I think Thomas Edison hyped the incandescent bulb technology before he’d really fixed the problem, and it worked out in the end. But I think what’s happened is that in these speculative markets we’ve lived through in recent years, there’s the reality distortion field that Steve Jobs popularized. And then Elon Musk taking it into the public markets with Tesla, always suggesting you’re gonna do something or about to achieve something when you haven’t achieved it. And that in itself can become, to some extent, a self-fulfilling prophecy in Musk’s case, because it gives you the capital to then make the development.

But if everyone is doing it and there is a huge amount of great competition, then I think that just increases the prospect of it coming undone.

Kai: So I want to kind of take the flip side of this. So it sounds like on the AI investment boom — you can feel free to interject if you disagree — but it sounds like you are, you know, negative. That you kind of see a repeat of history, these historical booms and busts, in the making. And, you know, that would generally make you bearish on — I would assume — all the companies in the investment supply value chain. Or are there particular pockets that you find to be areas where the prospects may be less dim?

Edward: Well, first of all, we all have our own sort of specialties. Perhaps some people who look at markets, whether investors or strategists, can cover many different bases. So I’m not saying I’m solely a capital cycle investor, but if you are a capital cycle investor, you’re sort of principle-bound to step back whenever you see a surge of investment across many companies. And a capital cycle investor can rationalize investing in a company that is itself investing a huge amount, providing the industry or sector they’re operating in is not investing massively too. But when a whole sector is plunging in, then the capital cycle investor prefers to step back.

And at the moment you’ve got a lot of the picks-and-shovels makers — all the DRAM companies, Corning, a lot of the same companies that were involved in the TMT boom have now come back. And I own some copper stock, and copper is also a beneficiary, as you may know. The reason I invested in the copper stock was not because I anticipated the AI thing, but because I thought there’d been under-investment in copper. So everyone — this is one of those very big investment booms where everyone really gets caught up in it whether they like it or not.

Kai: Right. So you actually wrote a piece recently for Reuters where you said, “Markets have poor scorecards for spotting AI losers.” This is the piece on the anti-bubbles, right? And so maybe you could walk me through this, because I think you talked in this piece about some cases where obviously the market got ahead of itself saying, “Hey, they’re gonna be the long-term winners,” and they were not. But there are also cases of the market — pockets of the market perhaps today even — that are trading at depressed multiples on this idea that maybe they’ll be disrupted by AI. So maybe walk me through this argument, which I think is pretty interesting.

Edward: Okay. Well, first of all, as you know, we used to have discussions about this at GMO — that the problems of shorting bubbles are that even if you’re accurate, you can have massive drawdowns, and so it’s probably not worth shorting a bubble even if you have perfect foresight.

And then I think at a sort of top-down level, I think bubbles do have a sort of crowding-out effect. So capital gets drawn into one sector, and it gets pulled out of other parts of the markets. And we saw that during the TMT bubble with what we’ll call the old economy stocks. Companies that were deemed not to partake in the new internet economy — the old economy that was soon about to be replaced. Those old economy stocks sold off, and that offered, you know, very good opportunities. Very good companies were selling at cheap multiples even while the market itself was at an all-time high. And then, as you know, value stocks as a factor were very depressed, small cap was very depressed, emerging was very depressed. So there were a whole load of anti-bubbles by 2000. And those were areas where they offered really great investments. So you could make money in the anti-bubbles while the Nasdaq was losing 80% of its value.

And then I drew attention, in that same piece, to the excitement over the energy transition. And if you remember, the SPAC boom of 2020–’21 was largely around stuff relating to electric vehicles — all those lidar companies and EVs and so forth. In 2020–’21, the traditional energy stocks were very beaten up. I think the energy sector went down to about 2% of the S&P 500 from its average level of around 8 to 10% weighting. And Tesla was worth more than the entire listed North American energy universe.

And that, to me — it didn’t take a huge amount of analysis to see that oil companies weren’t gonna disappear overnight, that their investments weren’t gonna be these so-called stranded assets, and that EV growth was largely driven by subsidies. And so that was a really nice anti-bubble. You didn’t have to short the SPACs, but it had created a wonderful investment opportunity. A friend of mine, Jonathan Tepper, who runs a company called Praevus Capital, was pointing out that there was this company called Garrett Motion that provides the turbos that go into internal combustion engines, that had a sort of 40% market share, 60% of incremental market share, selling on seven times P/E. And the stock’s up 150% over the last year or so. That’s an anti-bubble.

And where are the anti-bubbles today? Tepper shares the same skepticism about agentic AI, and he owns, you know, stocks like a car auction business or, you know, online travel. He doesn’t believe that these online businesses — whether in the real estate market or travel or whatever — are gonna be disrupted. And he also mentioned, I think, the London Stock Exchange Group, which not only provides the market for stocks but also a lot of other financial data. He doesn’t believe that their moats are fatally compromised. And the market — I don’t think, but definitely two or three months ago — was, I think, sort of overblown. There was a massive overblown sell-off.

The only company I saw that really has been genuinely impacted was one of these California educational technology companies that provided essay notes for lazy students. AI is gonna replace that business — stock’s down ninety-five percent. I’ve got no argument with that. The AI may make one or two errors in a student’s essay, but we all make mistakes. It’s not gonna be the end of the world if it does. But you don’t want AI to be booking you a ticket to London and then finding that it’s flying you to Timbuktu. That is a real problem, and unraveling that problem is gonna be very costly for a travel company that went down that route.

Kai: Got it. So you and Tepper believe that a lot of what sound like software companies or other companies that, you know, are selling down fifty-plus percent on this idea that they will be disrupted by AI — those are interesting places to at least look, to the extent that you don’t wanna short a bubble because, as we know, that can be quite challenging. So where are areas where investment has been crowded out? And you’re saying potentially the perceived losers of the AI disruption — which, as you write in your article, historically the market has a pretty bad scorecard when it comes to identifying who will actually, in the long term, be the losers, not always the folks who initially sell off.

Edward: Yeah. And it requires a bit of analysis, because people who hold that view also point out that software-as-a-service companies were a tremendous bubble three or four years ago. And part of their selling off is really just to do with getting back to what was fair value. And the other issue is that they love stock options, so they pay out most of their earnings to their employees. Those are businesses that, everything else being equal, you wanna be pretty cautious of. And when I last looked, some of those companies may have sold off a lot, but they weren’t optically cheap. And I think once you take into account stock-based compensation, they were probably worthy of avoiding. So look for the anti-bubble, but do a tiny bit of analysis.

Kai: Right. Of course, because there will always, in these cases, also be companies that are truly being disrupted by technology, and those will be zeros, right? You don’t wanna be buying Blockbuster into its downfall. Okay, so let’s switch gears now from technology to other capital cycles. When we were working together at GMO — I know you were spending a lot of time on this, in the early 2010s — spending a lot of time on China.

So I think this is a really interesting example of a case where, you know, I think the capital cycle did a good job explaining what ultimately happened with China, in terms of the fact that shareholders have not received a good return on their investment in Chinese stocks, despite pretty robust economic growth. So maybe walk me through that episode, and then also bring me to the present and where we are now in terms of the capital cycle with China and other emerging markets, if you can.

Edward: I wouldn’t say investors ended up poorly despite robust economic growth. I’d say because of the growth. Because the growth was faster than — the growth was higher than the returns on capital — the companies, to grow, had to raise more capital. So if your returns are low and the environment is growing quickly, it’s actually particularly negative.

I think my lesson from China — or the lesson from China as far as I’m concerned — is this: when you’re looking at markets, you don’t want to look at valuation alone. You want to look at valuation and returns on capital and the capital cycle. And China had — went through the greatest investment boom ever. They had relatively strong economic growth — tailed off a bit over the last few years, but still very strong growth — and absolutely miserable returns. And the shareholders were diluted, partly because the Chinese have a tendency to add companies to the index at very high valuations. Then the people who bought in at the high valuations — the index gets diluted by these new issuances and therefore gives a poor return.

So I think capital cycle analysis has been very well vindicated there. As you remember, we did a lot of work on the real estate. And, well, yes and no. I saw the other day a chart showing real Chinese house prices below where they were in 2010. A lot of those big — you remember we used to be short some of those big Chinese real estate companies like Evergrande. They’ve all gone. Many of them have gone bust. It took longer to play out than I expected. But as you know, however long you’ve been in this business, you might make a fundamentally sound observation and then be surprised by how long it’s taken to play out.

So yeah, I’m not following China so closely today. I know some people think that — I do some work with Marathon Asset Management, and their emerging markets people think that some of the Chinese real estate developers now, particularly in the tier-one and tier-two cities, are in a relatively good state. I’m not following the Chinese story closely enough to tell you where we are now. Just from 15 years ago, I’d say our positions have largely been vindicated.

Kai: Got it. So here’s another fun one. This is from one of your Reuters columns entitled “Big Booze Can Sweat Off Its Multi-Year Hangover.” Basically in the COVID boom, people were stuck at home drinking booze, these stocks did well, and then following the 2022 unwind, they have since collapsed in price and their valuations have fallen as well. But you, in your article from July last year, argue that these companies are Lindy. So what does that mean, and how are you thinking about these stocks today?

Edward: So the Lindy effect is a sort of joke. It’s interesting. It was a notion — the urban legend has derived from a Broadway cafe — where someone at the bar is saying, “How long do you think this show is gonna run for?” And the person says, “The show is gonna run for as long as the show has run.” So in other words, if the show’s been going for one night, it will go for one more night, then it’ll close. If it’s been going for five years, it’ll run another five years. And it turns out, actually, that as a sort of rule of thumb, that actually turned out to be reasonably accurate.

And what I would suggest is that’s probably the case for the spirits companies. Particularly spirits — the companies that were beaten up are companies like Pernod Ricard, Campari, Diageo, Remy Cointreau. Brandy sales really took off during the pandemic because people were drinking, and they were speculating. So they went hand in hand, and they were stuck. And apparently they loaded up their drinks cabinets, so by the end of the pandemic there was no space in the drinks cabinet, and they’d been slowly drinking off the spirits in a sense.

So you could say that was a stock problem. And then the other issue is these GLP-1s came along — these sort of weight loss drugs — and they apparently put people off alcohol. And then there might be an issue that the younger generation is taking ketamine rather than drinking. I don’t know. But my guess is that these brandy companies and spirits companies — they’ve been around for a couple of hundred years, and you’ve got countries like India getting richer. I don’t really like to play the emerging consumer demand story because it’s often, you know, what the brokers do.

But yeah, I just think it’s obviously not a hard and fast rule, and one can find exceptions. But I think as a good principle it works. And where they are now — I’m again not following them particularly. I saw that Diageo recently, the stock picked up because it had better-than-expected results, whereas Campari is still down. I particularly liked the story of Remy Cointreau, where the company was valued at less than the market value of the brandy in its vaults. And that reminded me of a story during the German hyperinflation when the Daimler-Benz company was worth less than the cars in the factory lots.

So I thought that was, you know, a good story. It wouldn’t surprise me if the GLP-1 fad diminishes somewhat over a few years, and eventually people will drink down their alcohol at home, and they’ll go back to the bottle. That’s my — anyhow, that’s what I sort of... There’s another anti-bubble theory, yeah.

Kai: Well, I guess one thing is that the AGI crowd — they don’t like to drink, so maybe that’s two sides of the same coin. And maybe it is the anti-bubble.

Edward: Yeah. It’s quite — if AGI were really to take over, there’d be a lot of idle hands, so they might actually start drinking again. Who knows?

Kai: Right. So maybe it’s a hedge against AGI — we have nothing better to do since the machines are doing all our work. Okay. So one question I had that’s kind of more personal, based on my own curiosity. As you know, my area of focus is often on intangible investments such as R&D and software, advertising, human capital. As you know, in the US for example, in 1995, intangible and tangible investment were both about 12% of GDP. Since then, intangible investment has increased to 16% of GDP, while tangible investment has fallen to 10%, over the past couple of decades.

And so capital cycle theory, of course, having originated many years ago, tends to be more focused on investment in physical capital — you talked about the energy cycle, emerging markets, AI data centers. So my question is, is it fair to assume that the capital cycle also applies to intangible capital? And if so, where are some notable examples that we can point to, either in the past or today, where we may have seen capital cycles in mostly or purely intangible assets?

Edward: Well, capital is capital, isn’t it? So I don’t think — if we believe in the concept of intangible capital, which I think we probably should, because it’s really largely to do with accounting conventions — we tend to expense R&D, and we don’t tend to put brand values on the books unless there has been an acquisition. Is that correct? Yeah. So then the question is, can you get over-investment in intangibles as in any other type of physical over-investment? I say the answer’s pretty obviously yes.

Probably — and you might be able to give me some better examples — but I remember that back in the early 1990s, there was huge excitement when Glaxo, as you know now GlaxoSmithKline, had this great blockbuster ulcer drug called Zantac. That encouraged a huge amount of investment by big pharma in R&D. The cost of blockbuster drug development soared over that period until the blockbuster drugs that came out weren’t delivering a decent return on equity. So I think that’s one that comes to mind.

Can you think of other intangible examples? I mean, I would say — as I say, I thought anything really to do — what we’re seeing today in the AI space we’ve been talking about: you’ve got obviously huge physical CapEx in the data centers, and then a huge valuation placed on the AI scientists. Meta going out and, you know, hiring people for — am I right? — $100 million something, as if they’re like football players. And so there are these AI companies — say, Mira Murati, former chief technology officer of OpenAI. She left a couple of years ago to form her own company. And they raised money at a valuation of $12 billion, even though she had said she wasn’t gonna tell investors what the company was gonna do. So that seems like an intangible capital boom.

And then, Kai, go think of brand valuations. This is not so much CapEx, but just brands getting overvalued. Do you remember how in the late 1990s, Buffett — who normally talks a lot of sense — got a bit carried away with the likes of Coca-Cola and Gillette and started calling them the inevitables? I don’t think either Coca-Cola or Gillette attracted a huge amount of CapEx competition. But they definitely had brands, definitely overvalued at that time. And Gillette was eventually swallowed up by Procter & Gamble. And Coca-Cola had a very poor decade after 2000.

Kai: Yeah. No, I think that’s right. Obviously an asset’s an asset, capital is capital, whether it takes the form of a factory or, you know, the IP embedded in an NVIDIA GPU, which is at least half the value of the actual data centers. It’s embedded in the value added of the semiconductor supply chain.

Edward: And Kai, we’ve been talking about software as a service, and we might as well draw out that I did write a piece back in 2022 about how these software-as-a-service businesses were attracting absurd valuations, attracting capital inflows and so forth. And do you remember — one of the capital cycle metrics is when you know you’re really in favor because someone creates an index, and then they’ll backdate the index to prove how well it’s historically performed, and then you can pretty much guess that from the moment of the inception of the new index, it’s gonna do poorly. So the SaaS bubble of 2022 is another very good example of an intangible bubble.

Kai: And another interesting related topic is if you could follow human capital, just talent. Where’s the Harvard MBA indicator? Where are all the best and brightest going? It was big tech in the past cycle. Before that, if you remember, it was Goldman Sachs and finance, right? In the bubble that peaked in ‘07, ‘08, everyone wanted to go work for Wall Street until they didn’t, and then it was tech. So maybe there is an interesting element here where the capital cycle has both human capital elements as well as the actual physical accounting capital.

Edward: Yeah. Well, Harvard — not just HBS, but the undergraduate graduating body there — very, very reliable contrarian indicator.

Kai: Well, I graduated from Harvard undergrad and went to work as a value investor in 2009. So I think I invested at the bottom of the capital cycle.

Edward: Would’ve got out of value investing at the time.

Kai: Yeah. Would’ve done much better. All right. I’ve got a couple more questions for you, Ed. So here’s one I like. There’s this idea in Silicon Valley that’s popular today that bubbles may actually be productive — that they may actually be good for the economy in the long run. There’s a book that’s been making the rounds called Boom, Bubbles, and the End of Stagnation by Hobart and Huber. The idea is that even if these bubbles form and they eventually end in tears for investors, they’re ultimately productive to the extent that they accelerate the development of a genuinely transformative technology.

So in a sense, the argument would go, even if AI ends up being a bubble — not that everyone there is conceding it — but to the extent it were, it would have still been a good thing to have happened. Do you find this argument at all compelling?

Edward: Well, it depends from what perspective one’s looking at it. I tend — and I think you do too — we tend to view the world from the perspective of an investor who’s trying to maximize his gains and minimize his risk. And from the perspective of society, if you can bring forward the technology — whether it’s railways or cars or aircraft or internet telecommunications, or EVs or batteries or AI — that’s all well and good. It sometimes might be a sort of dead end, but perhaps not the end of the world.

And I think I mentioned the railroads earlier. They tended to come in surges of CapEx both in the US and UK, and that was probably a good thing in the end for both economies. But pretty disastrous for the investors who partook in them. And so you have to bear that in mind. You’re under no compulsion to make a loss-making investment for the benefit of society.

The other issue is, as I’ve already mentioned, investment booms tend to lead to a misallocation of capital and overinvestment, and that tends to slow rather than increase economic growth. And even if the dot-com bubble — because of that fiber optic cable that was laid — meant that you could get Netflix up and running and all the video conferencing, whatever — that’s all well and good. But if you remember, the downside of the dot-com bust, with the S&P down about 40% as far as I remember, required — well, it didn’t require, but as a response — the Federal Reserve slashed interest rates to 1% to fend off deflation, and then we got a housing boom, and we got a bust.

So in fact, the way I see it, the upshot of the dot-com bust was the global financial crisis. The upshot of the global financial crisis, with the low interest rates and so on and so forth, was a collapse in productivity growth. So we’ve got the new technology, and that’s all very well for the companies that end up as winners. But actually the very long tail of these busts can be quite severe. And so I think it’s sort of irresponsible to argue that bubbles are good for society.

It’s very typical — one of the things you see during bubble periods is a sort of lightheadedness. “Hey, what does it matter if there’s a downside?” But of course people don’t feel like that when they’re nursing 90% losses on yesterday’s high flyers. And what’s going on in the AI space — this is a big CapEx boom taking place with very weak consumer confidence, a so-called K-shaped economy, and problems happening elsewhere in the financial system: private credit and private equity, and really a lot of legacy VC stuff that was badly invested that you didn’t really hear too much about. So the whole private alternative investment world seems to be in a pretty bad place. And I would’ve thought that if and when this AI boom ends and turns to bust, a whole load of problems will emerge, and as the valuations come down, people just simply won’t be saying that. That type of commentary is the sort of giddy commentary that invariably accompanies the bubble.

Kai: That’s really interesting. And so one point you made that was particularly interesting is this idea that the long-term consequence of the dot-com bust was a Fed mindset that led to ultra-low interest rates for a long period of time, which — and this is in your book, The Price of Time — was the original sin behind a lot of the speculative excesses that came afterwards. So something can happen, it looks fine, we can patch it over, but that’s just sowing the seeds for the next cycle. Now, obviously rates have increased over the past few years from their COVID lows. Do we think that the regime has actually changed? Or does the fact that we now are seeing a massive CapEx boom in AI investment suggest that the money never left the system?

Edward: Well, I don’t like to hedge my bets, but I’d say a bit of both. I think that the interest rate cycle turned in 2021/22 for long-term yields. And historically, those interest rate cycles have tended to be multi-decade — often 30 to 40 years. And if you remember, back at GMO, that was another thing that we did a lot of work on: why were our bond models so wrong? Because we always kept on talking about mean reversion. And what we found is these very long bond cycles. So I always say you can’t predict anything about long-term rates, because they’re not mean-reverting like, say, equities. But they do go on these very long cycles. And I do think that we have entered into a prolonged period of an upward trend in long-term rates.

However, there was clearly a lot of liquidity left over from that COVID lockdown quantitative easing splurge — adding in what, $8 trillion of money printed by the world’s central banks. And a lot of that excess savings was still held on the Fed balance sheets. And that has been drawn down. Interest rates remain low relative to nominal GDP growth. And so the question is: what’s gonna happen? Are interest rates gonna go up, or is nominal GDP growth going to go down? Or perhaps, possibly the worst of both worlds, NGDP goes down and interest rates go up.

So I think there was a lot of liquidity left over from that period. It’s always difficult to know what’s going on in the plumbing of the financial system. But there was perhaps more liquidity than I’d suspected left over. And there are, as you know, these problems in private credit — which is really just the financing arm of private equity — that are still percolating away. I expect that will continue to be the case.

And the real estate market — I think in America it’s fairly moribund, isn’t it? Because house prices really haven’t come down enough in response to the shift in interest rates. So that seems to be stuck in — it’s not gonna sort of muddle along. I think it’s not gonna sort of middle — it’s either something good is gonna happen or something bad is gonna happen. I’m not quite sure which.

Kai: Fair enough. Yeah. Okay. So I just have one more question for you, which is our standard closing question. What’s the one thing you believe about investing that most of your peers would disagree with?

Edward: I suppose one area where I would have differed from a lot of sophisticated investors is that I’ve always been a bit of a gold bug. And people like our old boss Jeremy Grantham and the team head Ben Inker would always say, “I don’t like gold, has no dividend,” this and that.

And I’ve always — yeah, I think I’ve always got this atavistic attraction to gold, and I think that it does... I do like gold’s hedging aspect. I know it is difficult to value, but I like the way that gold is an asset without a liability. And I think that in a world where equities — particularly in the US — look overvalued, and bonds might be on a 30 to 40-year downward trend in valuations, as well as really severe debt problem dynamics, I hold the view that having a decent allocation to gold is a good portfolio position.

And I may have this completely wrong — the last 10 years, having a sort of equities-gold portfolio has been obviously a lot better than having bonds. So I might be touting something that’s near the end of its run. But I’m drawn to gold in a way that the average CFA is not. So I think that’s probably the one area where I might disagree.

Kai: Interesting. Well, thank you, Ed. I know I’ve taken up a lot of your time, so really appreciate you coming on.

Edward: Good, Kai, and I’ll see you in London soon.

Kai: All right. See you soon.