Full Transcript: Kai Wu on Software Stocks, AI Disruption, and Moats

Value Traps vs. Complementary Assets in the Software Sell-Off

Justin: Hey, Kai. Welcome back.

Kai: It’s good to be back, guys.

Justin: Our audience is familiar with you. You’ve been on the podcast a few times. We always like having you back because in addition to running Sparkline Capital and managing the ETFs that you run and that you’ve built on sort of this intangible value framework, you’re also consistently putting out very interesting, deep pieces of research on where the markets may be misunderstanding disruption, innovation, and the way that you look at sort of intangible value.

And your recent piece that you put out in May titled “AI Disruption: Moats and Value Traps” is looking at the recent sell-off in software and this possible opportunity that it has created. And there’s this idea right now in the market that AI is gonna be this existential threat to these software names and not necessarily an opportunity.

But I think as we work through this great piece that you did, we’ll kinda get into what the setup might be in some of these software names and sort of how you’re using your unique aspects of natural language processing and research to sort of uncover these possible opportunities.

And so this is one of the episodes where we’re gonna be pulling in a lot of charts. Kai is gonna be working through these with us. He shared these charts with us and our audience so we can get, like, really down to the nitty-gritty detail on the research that he’s done. So I just thought we’d start, Kai, with your exhibit two, which kinda shows where the software premium or lack of premium is today, and how unique that is in terms of software stocks after this sell-off.

Kai: So I guess the first thing just to set the context is that historically, software stocks have commanded a premium valuation, right? Historically, investors have liked software stocks more than, say, the average industrial in the S&P 500 because they’re asset light, because they have predictable SaaS-style revenues, and for a variety of other reasons.

They fast-growing and such. So over the past roughly twenty years, since this data began, their forward P/E ratio of software relative to the S&P has been at a thirty-two percent premium, right? So that’s been the historical average. And there’s been some fluctuations. So it kind of dipped a little bit in ‘09 and then went on kind of a secular bull run, peaking in 2021.

If you remember, that was kind of the COVID bubble where people were working from home and interest rates were at all-time lows and stimulus was coming to the market. So software stocks were kind of at their all-time high valuations. And then 2022, things started to reverse. Valuations started to fall.

They went through their historical average around ‘23. And then the past two years they’ve been continually falling, kind of reverting back to first parity with the market and then more recently over the course of this year have actually fallen to a discount to the market. So software stocks, at least on this basis, are trading currently at a ten percent discount to the market, which has never happened before over the course of this sample.

There’s also a chart floating around from, I think it was an Oakmark letter. They’re another value manager. But they cited Empirical Research Partners. They take the same data back to about 1980, and what they show is the same, that over the past five decades, first we are at all-time lows with regards to the spread between P/E ratios of software versus the market. And by the way, that we’re at a discount in absolute terms to say the median stock or the average stock in the S&P, which is something that we haven’t really seen before.

Justin: And so that’s one of the core ideas in this paper: trying to determine now that this sell-off has happened and these are trading at these types of historically low valuations, whether or not these are possible value traps.

And so talk to... I think it’d be helpful. I mean, most people I think know what a value trap is, but just talk to what a value trap is and then why, I guess, value traps are problematic in the sense that sometimes when these securities get down to such low valuations they look like they’re no-brainer buys, but they’re actually a value trap and they don’t actually add value or they don’t ever appreciate from that point going forward because their model is effectively being disrupted.

So talk to that.

Kai: Yeah. So all a value trap is is a stock or a company that is basically on its way to oblivion, but that for a variety of reasons appears cheap on traditional or on standard metrics. And so for example, a stock that has a low P/E ratio but only has a low P/E ratio because everyone knows that the E is going to zero would be a classic example.

What I show here in the paper is the example of four iconic companies: Blockbuster, Borders, RadioShack, and McClatchy, which owns a bunch of newspapers, each of which were disrupted by Amazon, Netflix, Google over the past couple decades. And these were at one time large multi-billion dollar companies.

But you know, over time they kind of became cautionary tales. And what I show here in this exhibit is actually interesting because I compare the stock price to the fundamentals, in this case the revenue per share in the red. And what you can see is that when the disruption first happened, investors quickly panicked and started to sell down the stocks, right?

So as the internet became more and more pervasive, stocks like RadioShack and Blockbuster and Borders started to-- the price started to fall. But importantly, the actual fundamentals only fell with a long lag. So in the case of Blockbuster, it took many, many years for the sales per share to fall.

In the case of Borders and RadioShack, they actually increased their sales per share for a period of time before it all kind of fell, the wheels fell off the wagon. Now profits also were maybe deteriorating as well over this period. In the case of McClatchy, there was an acquisition and a lot of debt taken on.

But the overall picture I think is pretty clear, which is that to the extent that prices in the stock market are forward-looking, and they kind of price in to some extent disruption, you’re gonna almost always end up in a situation where prices fall faster than fundamentals, which are lagged, can keep up.

And so you’re always gonna end up with a window of time when, say, the price to sales ratio of these companies is looking really attractive to a traditional value investor, yet again, that’s just kind of a trap. It’s sucking you in to bringing you on board a ship just as it’s about to collapse and sink into the sea.

So that’s what a value trap is, and I think these examples pretty cleanly illustrate what investors, as they approach the software boom, should be concerned about.

Justin: You know, a lot of listeners or viewers might not realize, but I remember when Netflix first sort of came out with its mail model, you could get the three DVDs in the mail, and it was like, how is this ever gonna disrupt like a Blockbuster? You couldn’t see it. But with all these examples it’s always hard to see early on when this disruption is possibly happening in front of your eyes. And so that was another thing that I thought was very interesting in the paper, your sort of methodology for a way to measure this disruption exposure and kind of what that tells us about maybe the current environment that we’re in.

Kai: Yeah. So I think the key here was obviously those four examples are cherry-picked. They’re helpful anecdotes. But the question I really wanted to answer was that if you were more systematic about doing this, would it turn out to be the case that these four examples of the Blockbusters and Borders are actually representative of a systemic problem that traditional value investors might face?

So in order to build on that and kinda set that up, obviously I needed a way to, in real time, quantify when there’s disruption and which companies are exposed to that disruption. So the way I did this was in a two-step process. I built on a paper I wrote in 2022 called “Investing in Innovation.”

And so what I did was I looked at this dataset of all the patents ever filed with the US Patent and Trademark Office. Really cool dataset. It goes back to 1790. The first patent was signed by George Washington. And basically you can use it to see over the course of the past two centuries the rise and falls of new technologies, like from the automobile, electricity to the internet.

And so what I do is I say, look, at each point in time there are hundreds, thousands, tens of thousands of patents. What you care about is trying to cluster them into groupings of similar technologies, and then from there to see whether there is an increase in the technology. Because oftentimes you find that there might be false starts where, say, a technology starts to gain prominence, but then eventually fades, right? Electric vehicles were famously a competitor to the internal combustion engine, but then ultimately lost out about a hundred years ago or so.

And so you wanna find trending technologies. You also wanna find technologies that aren’t only just trending, but also are pervasive. And what I mean by that is that they’re not just increasing a lot in one specific sub-domain, say like in healthcare, but they also are pervasive across industries, right? AI is kind of the best example. They call it like a general purpose technology, meaning that it applies of course to software use cases, but should in theory be able to automate and do a lot of the human labor that is obviously a factor of production for most economic activity, right?

So I want pervasive things. I look specifically for increases in patent volume that are pervasive across industries. And so then that allows me to define the technologies themselves. And then the second step is to figure out what exposure firms and industries have to that disruption.

And what we do is we look at a bunch of different documents ranging from earnings call transcripts, to patents themselves, to company filings, analyst commentary, to figure out which companies are exposed to each technology. So for example, if e-commerce becomes a thing, which companies are potentially exposed to that disruption.

And then what we do is we actually roll it up to the industry level, because any single company can be noisy. The data themselves can have noise. So you aggregate to the industry level so that you can say within retailers as a whole, even though some may not be exposed and some might be exposed, on average, they have this level of exposure, right?

So we’re creating an industry-level exposure. And then in this chart, I basically highlight over the past few decades, I think seven different major disruptive waves ranging from the advent of internet infrastructure to e-commerce, social media, and then AI, right?

And as you roll through, I kind of look at what periods were each of these disruptions most prevalent, when the pressure was the highest on disruptees, which are some examples of patent clusters that could collectively form this theme, and then which sectors at each point in time were most exposed.

I think one takeaway here is it’s not always just tech. Retailers, newspapers, entertainment being good examples of sectors that were exposed to, say, digital media or e-commerce, even though they may not have been the progenitors of that technology.

Justin: Just a process question here, and then I’ll let Jack go. When these clusters are being built or formed, are you telling it what to look for, or will it automatically find it through the natural language processing? Does the system find these clusters automatically, or do you have to instruct it as to what to look for?

Kai: I mean, there are some hyperparameters, like how many clusters, like how sensitive the thresholds. But putting that aside, it’s fully automated. So it’ll kind of go through and say, “Hey, I’m gonna look through all the patents,” and then figure out the clusters, and then figure out which clusters are trending and which are not, and then separately in step two, identify companies and then industries that are exposed to each sector.

So for example, I have my code set up where we could dust it off in 20 years from now, and it’ll have a totally new set of technologies and companies.

Jack: This is just another aside, but does this at all allow you to, like, rank these disruptive waves against each other? So like everybody’s talking about how AI is the biggest disruptive wave we’ve ever seen in history. Like, does this tell you anything about that at all?

Kai: I mean, I think it’s a fair point. If you measure it by pervasiveness, as I mentioned, artificial intelligence is meant to be a general purpose technology. Not that these other things aren’t necessarily, but it can in theory affect all facets, especially once robotics and such are in play, of the economy. I think one thing I would add though is that they’re all like dependent on each other, right? This is the idea of the stacking, the stacking of innovation over time, right?

Like we wouldn’t have AI if we didn’t have electricity, right? We wouldn’t have electricity if we didn’t have, I don’t know, fire, right? So all the technology over civilization’s history has kind of compounded over time by building on each other. And so I think it’s important to remember that AI is obviously a really cool technology.

It requires really advanced computing. It requires big data, which is obtained in many cases through the advent of the internet being able to digitize information and put it into a format that we can all see. So these things all tend to build on each other.

And so maybe we are at the apex currently in terms of where innovation and disruption is. But a lot of that just depends on previous innovations that, had they not occurred, we wouldn’t be able to be where we are now.

Jack: The stacking gets to this next chart we want to look at, and poor retail is all I can think about. It’s been like getting the crap beat out of it by like every innovation for decades here. But can you just talk about the idea of what we’re

Justin: seeing here?

Kai: Yeah. So this is exhibit seven. I mean, it could have picked on a different industry. I guess it was mean. But the idea was to show through time the amount of disruption it’s been absorbing from the seven or so different themes that I mentioned through time.

So kind of the key insight here is that they stack. So in other words, first e-commerce comes around, the internet comes around, and that’s obviously Amazon, and if you’re Blockbuster you don’t even make it past that. But let’s imagine that you do make it past that.

Well, then next you have to deal with digital media, and then you have to deal with social media, and then you have to deal with AI, right? Like agentic shopping and such moving forward, right? And these things stack. In other words, it’s not that companies today don’t also have to deal with e-commerce as a threat, but they also just have to deal with that, plus they have to deal with digital media plus AI.

And so if you sum up the exposures to all the different themes over time, what you find is that yes, they come in waves, right? The peak of e-commerce disruption happened and then it kind of settled, subsided before social media really came into play.

But the trend is kind of this secular increase over time. As innovation is accelerating, as technology compounds on itself, we end up in a situation where these companies are now being exposed on all fronts. They’re waging like a multi-front war, so to speak, against innovations and technology coming from all different angles now.

Jack: It also explains a lot because as a value investor, like if you’ve watched your value screens over the past however many decades, like these retailers have like perpetually been in them. Like they don’t ever leave them. Like these retailers in the mall and stuff like that, like those are always sitting in these screens.

And so like just innovation getting stacked. I knew your long Abercrombie.

Justin: Yeah. It just explains why it’s happening. I knew your long Abercrombie & Fitch over there.

Jack: Yeah. I mean, you’d be surprised, like Justin, we’ve been running quant models forever. Like you and I have seen these various names

Justin: Williams-Sonoma, Abercrombie & Fitch. Yeah. You got Claire’s used to be in there. You got they’re

Jack: all-- Oh, yeah. They’re all in there. Hot Topic. Yeah. And like I don’t even know that even exists anymore. Like all these things.

But anyway, back to the paper. This next chart gets to this idea of the death of value investing. So before we get into kind of what’s going on now, it’s just good to maybe take a cumulative look at this and how the value factor has performed and why it hasn’t been working for a long time.

So can you talk to this chart?

Kai: Yeah. So I mean, I’m sure your listeners are somewhat familiar at least with this idea. Value investing, buying beaten down retailers -- the general idea has been a long-held tradition amongst many investors ever since Ben Graham in the 1930s. Warren Buffett, of course, being a famous proponent of the school. The thing is that value investing, however we define it, has really had a tough time of it. And a lot of it has to do with disruption, which we’ll get into more, over the past couple of decades. Obviously, there are many different ways of quantifying it.

What I’ve done here for this exhibit is to do something pretty simple where I create like a long-short factor. So in other words, you go long the cheap stocks, short the expensive stocks on a valuation metric. In this case, it’s a blend of, I think, four different things: price to earnings ratio, price to book ratio, price to sales ratio, and price to free cash flows.

And basically, the reason why you kinda diversify across different metrics. And the point is this, which is this is a factor that, if you extended the back test all the way back to the initial work a hundred years ago, you would have seen consistent outperformance for decades.

And then around 2010, you would have started to see a drawdown. It starts to turn over and really has never recovered, even as of today, right? And so this is what has led many people to declare this the death of value investing, that perhaps we should be doing meme stocks or the principle of value doesn’t apply anymore. Value investors have lost the plot. They’re all too old school, and they’re just buying a bunch of Abercrombie stocks or whatever, right?

But to me, it’s always been: that can’t be right. I mean, value investing makes sense by definition. It’s just maybe the way we measure it that could be problematic. And so this is kind of a really interesting study that we did here. And kind of the conclusion is that value investing is not dead. It’s maybe just being disrupted.

So what I do here in exhibit nine -- and yeah, this is really the key exhibit of the paper -- is I say let’s not apply the value factor to the entire stock market, but let’s instead apply it separately to two different parts of the market.

First would be what we call exposed industries. Those are the industries for which their technological exposure score, which we showed in the case of retailers, exceeds a fixed threshold. And then the second would be insulated industries. So those are industries that are not exposed, right? And those two things collectively by definition comprise the market. So divide the market into exposed versus insulated halves.

What you find is that when you apply the value factor in the insulated sectors, actually the performance has been great. Just fine. You almost see no difference between 2010 on and the beginning period.

In other words, value has worked just fine as long as it’s not in industries that are exposed to technological disruption. However, if you look at exposed industries, industries like retail starting in the mid-2010s or around then or a little earlier, you find that the performance has been quite bad.

And starting in 2010, you have this big drawdown. And by the way, the drawdown is so big it overwhelms the positive returns from applying the factor in the insulated industries, such that the net return for the factor is negative. So said differently, you can explain the demise of value investing through the lens of: if you want to apply these traditional metrics to real estate companies or asset-heavy businesses, fine, go ahead and do that. It’s no different. But if you want to try to apply it into sectors that are now exposed to technological innovation, not just software but sectors like retail that maybe initially were not exposed but now are heavily exposed, you’re gonna have some issues and that’s not gonna work.

And then to the extent that the market is more and more, as we’ll see, in exposed industries, that’s gonna overwhelm the positive returns you get from this kind of vanishingly small part of the market that is insulated.

Jack: It’s such an interesting point because if you go back to my point about retailers or the mall, like if I had known in advance that they were undergoing a disruption, I could have not applied value investing in that industry. And that’s how it proved out to be the truth. Like you did not want to use value investing in retail like basically at any period in the past however many decades plus, you know.

Kai: Right. Yeah. And I think it requires two things. One is a recognition that this disruption is coming, and second, a recognition -- kind of like a lack of hubris -- to be like, “Yeah, and by the way, I’m not gonna try to apply my metrics in this thing because I just don’t think it’s gonna work, right?”

Warren Buffett talks about this circle of competency. For the longest time he avoided tech stocks. He said, “This is just not my cup of tea. I don’t know how to-- I just don’t do this.” Which works when tech isn’t like the entire market. And then when it is, you’re mainly in cash.

Jack: So this next one, you did some robustness checks to check deeper to make sure there’s nothing else going on that would explain why value’s not working here, right?

Kai: That’s right. Yeah. So one fun aside is that this paper is kind of the first one that I did where I relied heavily on, in this case, Claude Code, to do a lot of the experiments.

And so the fun thing was that I basically created the baseline script. And once I had it, I was like, “Hey, you know what, I wanna test a bunch of different things. Not just the US. I wanna test it in global stocks, international stocks, emerging market stocks. I wanna look at sectors, sub-industries, industries, sector groups, so on and so forth.” And so kind of the workflow ended up becoming like, “Hey Claude, can you take this and apply it to these other things? Show me the results. Let me look at the table with you. I have a couple follow-up questions,” so on and so forth. So it’s kind of a fun way to scale the analysis by delegating a lot of the robustness exercises to Claude.

But yeah, so that allowed me to cover a lot of ground. I mean, what I’m showing here are just five of the major robustness checks.

What this shows, by the way, is the spread between exposed and insulated returns. So this is a negative number because when you apply the traditional value factor in exposed industries, it does worse than insulated industries. And so the baseline shows a spread of negative seven percentage points per year. That’s very bad.

And then you can say, what if you look at just global stocks, not just US stocks? Kinda not so good as well. What if instead of looking at this blend of valuation metrics, we focus just on the canonical Fama-French price-to-book ratio? Okay, that doesn’t work. What if we sector neutralize? It’s a little bit less bad because you’re explaining some of the variation. But even within sector, you’re seeing that there is kind of an effect here.

What about if you do these double sorts? Where what investors will do is they’ll say, “I recognize that price to book or price to earnings could have value trap risk, so therefore I wanna intersect it with, say, ROE or some profitability metric, or momentum, in order to ideally weed out the value traps.” So you want companies that are both cheap and profitable, or cheap and have not bad momentum. Well, it turns out that that actually helps in absolute terms, but on a spread basis it’s the same, right? So you’re not -- maybe the lines are not going from positive to negative, but they’re going from positive to positive but less positive.

But the point just being that the gap remains, in this case six point three percentage points, so a meaningful, meaningful gap. So the point just being that this finding seems to survive the exact specification of the value signal, the universe applied to, and so on and so forth. So it’s a pretty robust finding.

Jack: Yeah. We just did an episode with Cliff Asness, and this reminds me of exactly what he’s done in a lot of his papers. Like his international paper, “Value Investing Is Dead,” he’ll ask like every question as to what could possibly explain this, and then once he’s eliminated all those, he’ll say, “All right. My conclusion’s okay.”

Kai: Well, I mean, I guess it’s impossible to prove anything, right? So we have to just kind of narrow it down and try to throw out as many competing hypotheses as possible, right?

Jack: So exhibit 12 kinda gets back to what I asked at the beginning, but this idea of how big the disruption is. So what are we seeing here? I mean, obviously this is a very large number of companies that are exposed.

Kai: Yeah. So what we see is the percentage of market cap in, I think this is in the US market, exposed to innovation however defined, increasing from about forty percent to mid-seventies, seventy-five percent, let’s call it, over the past twenty years.

Right? And that’s the result of two things. So one is the fact that technology is affecting more and more industries, right? If you went back to the 1980s, tech was just like IBM, right? And now all companies have tech, to give you one kind of simplified example. And second is just that the tech industry, or whatever you would call these technologically exposed industries to be more precise, are just a bigger part of market cap, right?

But the point is that even if you look at things on an equal-weight basis, on a names basis, not just market cap, you get a similar result. I think it’s seventy-two versus seventy-eight percent. And by the way, if you look outside the US, the numbers are a little bit less extreme. So within like developed markets or emerging markets, the numbers aren’t quite seventy-eight percent, but they are still above fifty percent, and they still have the same feature of an increasing trend.

So even within emerging markets, which have been the kind of least technologically advanced of the major economies, you do find a trend where the idea that “Hey, I’m a value investor. I wanna do the thing where I just like hide in non-exposed sectors” -- that has been an increasingly challenging thing to do as those non-exposed sectors kind of go away.

Jack: Yeah. This reminds me, we talked to Andy Constan in the last episode. He was talking about this idea of what’s different from ‘99 to now, and one of the differences was like tech’s just a way bigger part of the economy and the market than it was then. Which I think kind of is a corollary to what you’re talking about here.

Kai: Right. And not only is tech as a GICS or MSCI definition a bigger part, but tech cross-cuts across all industries. Even an industrial is more reliant on technology today than it was twenty-five years ago. But yeah, I would agree with Andy’s point on that.

Jack: So I guess the big question here is how do we differentiate the companies that are gonna survive this, that are gonna thrive in this, versus the companies that are gonna be disrupted? And I think as we get further in the paper, that’s what you’re trying to address, right?

Kai: Yeah. So I think this is where we switch gears. So I think the first part of the paper was more around what not to do, right? And now it’s like, okay, so can we actually study history and can it be actually illustrative as to what you should actually do, right? And so here is where I bring the examples of Walmart and the New York Times into the discussion. Obviously Walmart is a retailer, New York Times is a newspaper. Two of the most beaten down industries from disruption. What you find is that these companies survived and then ultimately thrived, despite being in the crosshairs of disruption.

How did they do it? Two things. First, they -- maybe not initially, but eventually -- leaned into the technology that was disrupting them, right? Walmart has one of the biggest e-commerce businesses today. And second, they leaned into their unique intangible assets that, outside of technology, let’s say, allowed them to be who they were, right? Their brand, their human capital, and network effects.

And so this is where I bring in a paper that I thought was really interesting by this guy named David Teece. So this paper was about who profits from technological innovation. It was written in like 1986. But the principles, while dated, are timeless, I think.

And the key insight was this, which is that the long-term winner of an innovation isn’t always gonna be the initial winner or the innovator itself, right? Oftentimes the firm who ultimately accrues the value or captures the value of an innovative cycle, a disruptive cycle, is not the core innovator, but in fact the firm that possesses the complementary assets.

This is his terminology -- complementary assets -- that surround the innovation. So I have here an exhibit 14 that shows some examples from directly from his paper of things that are considered complementary assets. That’s manufacturing, distribution, customer service, and then complementary technologies. So outside of the focal IP, other IP that surrounds and kind of cements a moat around that, right?

And he gives all these really cool examples. He gives the example of a company called EMI, which is a UK-based company. They actually invented the CAT scanner, which is a machine they sell to hospitals. But the problem was that selling stuff to hospitals is really hard. And it turns out that you need to like do a whole enterprise sales cycle, and then you need to effectively deploy engineers, like train their people how to use the machine and then service it once it breaks down. Like, and that’s a really hard thing for them to do.

What ended up happening was GE, General Electric, came in there, and they had those other complementary assets in place. They didn’t have the technology itself, but over time they figured it out and they developed it, and then they won the market.

They have another example he talks about called RC Cola, which I guess was like a small cola company. They actually invented diet and canned cola, so that was an innovation at the time. But they didn’t have like the shelf space or the distribution or the brand that Coca-Cola and Pepsi had, and they obviously won that market.

And then another example was kind of the opposite case: IBM. Teece talks about how IBM at the time was late to the PC market, but they managed to capture it in the ‘80s at least. And he says it’s not through the strength of their technology, but rather through the ecosystem of software and peripherals that they kind of built around the IBM framework, right? What we would call network effects today.

And so you have these examples which are quite illustrative, right? So who wins? IBM, Coca-Cola, and GE, right? They win on the back of these intangible assets, even though they weren’t the ones who actually innovated the underlying technology. They didn’t invent Diet Coke, they didn’t invent the CAT scanner, but they had the necessary assets to win the market, right?

So I think that’s a really important lesson to think about as we approach the software sell-off. These software companies -- if their core moat is code, then yeah, maybe they are in trouble. But if it’s the complementary assets around that, then per Teece’s framework, they actually might be fine, right? And conversely, yeah, Anthropic and OpenAI are the early winners. They are the innovators of this new technology, the LLM, but that doesn’t necessarily mean that they will capture all the profits, because there’s so many other things that matter when it comes to the way that these competitions unfold.

Jack: Well, first of all, RC was really good. I don’t know if you guys ever had it, but it was actually, I thought, I thought it was better than Coke and Pepsi. Really? Did you ever have it, Justin?

Kai: Okay.

Justin: Jack, no, but you are a soda expert, so I would imagine your ranking would be important here.

Jack: I mean, you don’t run the five thirty miles Justin runs if you’re drinking things like RC Cola. It’s people like me that are doing more of the consumption of RC Cola. But the other thing that was interesting, Kai, is we were talking about the idea of leaning into the disruption, and one of the things that crossed my mind is it’s interesting because like in the past you’ve had the Walmarts who had to lean into technology disruption. Now, this time you’ve actually got tech firms that have to lean into technology disrupting their technology, which makes it a little bit unique what we’re going through right now.

Kai: Yeah. I mean, in a way they are positioned a little better than some of the firms of yesteryear when the new technology comes around. ‘Cause they’re already kinda tech-facing. Is it a different type of technology? Yes, right? Like being a good AI coder isn’t necessarily the same thing as being a good traditional engineer, but it’s definitely a lot closer than being somebody at Blockbuster, let’s say, when the streaming comes around.

And we’ll see the data on this too that software companies are amongst the most aggressive in terms of their adoption and investments in AI. So they see it coming, they know it’s a threat, and they know they’re vulnerable, and they’re doing, in many cases, what they can in order to offset that exposure.

Jack: So one of the things you talked about in the paper, which we’ve talked to you about in previous interviews, is this idea of intangible moats and their ability to protect themselves using these things.

So before we get into that, I thought maybe it would be good just to revisit that quickly. I’ll put up this exhibit 16 here quick -- intangible value -- if you can just talk about the different intangible moats that you measured.

Kai: Right. So I touched on each of these. So the first of the four is intellectual property, right? This is not just patents, but any kind of proprietary knowledge, data, software, technology. Second is brand equity. That’s customer relationships, brand loyalty, things like that. Third is human capital. That’s not just the possession of a talented workforce, but also one that’s culturally aligned around a common goal.

And then finally, network effects, which is this ecosystem of external producers and consumers. We saw this with IBM. Examples today might include like Uber or the New York Stock Exchange, ICE. And so when you have these four types of intangibles, they can be really important.

In fact, I would argue that most value today, most value captured today in the economy, is due to these four intangible pillars as opposed to traditional tangible capital, which I think is just having a lot of book value. Is that really a moat, right? Does that actually give you the ability to kind of earn excess ROIC? I would argue probably not.

But yeah, to your question, how do we build the metric? We have a bunch of different underlying proxies for these pieces. So for example, like we looked at for the traditional value metric: price to earnings combined with price to book, price to sales. In this case, for IP, we might say let’s look at the patents to price. We might look at R&D expenditures to price, so on and so forth, smush them together into a metric. Do the same for brand with like trademarks and social media, human capital with maybe job postings and employee profiles.

And you create these scores, similar to what we did with the traditional value -- these yield-based metrics looking at price relative to X, X being a measure of intangible capital for these four different pillars -- and then we combine it into one final composite score so that every single company in your universe, whatever, five thousand global companies, have a score that you can then look at least at any point in time to see who’s high, who’s low, and you can kinda build factors around it the same way we did with traditional value.

Jack: And as we get into exhibit 17 here, this looks at that idea that traditional value investing is not working in these exposed industries, but when you adjust and use intangible value investing, we get a different story, right?

Kai: Yeah, that’s exactly what we see here, right? So if you remember the exhibit from before, we saw that traditional value applied in the whole universe worked okay, and then it stopped working around 2010 and was in a drawdown. And when you split it into two pieces, it did totally well in insulated industries, but struggled in exposed industries.

What we’re seeing now for intangible value, so once you look at not just the traditional metrics but you also add in these intangible moats, what Teece would call these complementary assets, right? What you’re finding is that the factor works in insulated industries as it did before, but most importantly, it now goes from not working to actually working quite well in exposed industries, right? Because exposed industries -- put aside the jargon -- these are just industries where you’re facing disruption from a technology, whether it’s e-commerce, whether it’s cloud computing or social media or AI.

And what allows you to survive, what allows you to be The New York Times or Walmart, are these complementary assets. And in addition to, of course, embracing the technology itself, right? All of which are in theory captured by this four-pillar framework.

I think this is actually an important point, right? The Teece framework and the intangible value framework are actually really related. You think about it this way, which is like Teece says there’s focal innovation, right? The focal innovation is a subset of the IP pillar, right?

So of course, when we go out and we say, “What is the intangible value looking for?” It’s looking for companies that are doing AI, of course. But we’re not just looking for companies doing AI. We’re looking for companies that also do other types of innovation, other types of IP. As Teece would call them, complementary intellectual property innovations, like in robotics, like in genomics. And we go even one step further and say we’re also looking for companies that have strong brand moats, human capital, network effects, the true complementary assets, right? So we kinda want all these different things.

And so going back to the exhibit here, what you find is that once you kinda look more holistically outside of just backward-looking earnings and book value and look at what intangible moats companies have, now you’re starting to put together a framework that works not just in insulated, but also in exposed industries. Even when you’re facing technological disruption, you’re able to separate the kinda Walmarts from the Blockbusters.

Jack: What’s interesting too is in exhibit 18, like if I was trying to put together a more ideal value strategy, what I would want it to do is work regardless of the disruptive period. I wouldn’t wanna have to figure out if I’m in the disruptive period. I want it to just work regardless. And I think that’s what you’re getting at here with intangible value versus traditional value, that even in this disruptive period, non-disruptive period, the performance has been pretty similar of intangible value.

Kai: Right. The key is consistency. I guess what you’d call all-weather, right? It appears to work... So first of all, what this exhibit does is it cuts it into two dimensions. One is exposed versus insulated, and the other dimension is by time. So we’re looking at the first half of the sample when things were kinda better, and then the second half when things have been more challenging for traditional value, right?

And so what you find is that intangible value, regardless of the time period or whether you’re looking at exposed or non-exposed industries, has tended to be pretty consistently around the same outperformance. Whereas if you look at the traditional, it’s highly dependent. If you’re looking at like the first half of the sample and the insulated industries, you do great. But as soon as you start to go more recent or you start to go to more exposed industries, traditional value kinda falls down, right?

So that’s the challenge, which is like when it becomes so contextual, then if you do factor timing, it can work. But then you need to have a good model and a good understanding of when to apply it and when not to, versus it being more all-weather.

Jack: This next one’s really interesting ‘cause you actually looked back to 2007, and you looked at the companies that were out there, and you looked at traditional value, and you looked at intangible value, and you looked at what they agree on, what they disagree on, and then what ended up performing well.

So what is the lesson from this?

Kai: Yeah. So what this exhibit shows, it’s like a matrix, a two-dimensional thing. So on the X-axis, it shows like the traditional value score from expensive to cheap. On the Y-axis, it shows the same but for intangible value, right? So we have like the four quadrants where the diagonals are where they agree, and then the off-diagonals are where they disagree.

So the upper right is where companies where both metrics agree. The lower left is where they both disagree. The lower right is where a stock might look cheap on traditional but not intangible. And then in the upper left, it’s the opposite.

The other thing I did here is I color-coded each dot into three colors. So blue means it’s a company that over the next ten years was a winner, right? Apple, Kroger. Gray means it did okay. And then red means it was a loser, like Las Vegas Sands, GameStop -- did not do well from ‘07 to 2017, right?

And what’s immediately visible once you look at the colors is, first, that intangible value worked pretty well because most of the blue dots, the winners, were in the top half of the exhibit. In other words, intangible value, regardless of where it scores on traditional value, cheap intangible value stocks had done well the next ten years.

The other thing you see is that, in the off-diagonals, traditional value had some challenges. Like stocks that looked cheap on traditional value but expensive on intangible value, like Macy’s or Wells Fargo, tended to be losers. And stocks that looked expensive on tangible value but cheap on intangible value, like Amazon or Apple, tended to be winners, right?

So this kind of explains, I think, more intuitively what we just saw. Like, why was it that traditional value struggled in exposed industries? Well, it was because they sold the Amazons, and they bought the Macy’s. Whereas intangible value, because you’re now taking into account the intangible moats that an Apple might have, the network effects, the brand, the human capital, the IP, suddenly Apple no longer seems expensive, it seems cheap, right?

So it helps you kind of more discriminate between companies that might seem expensive optically but are actually truly disruptive, and also companies that might seem cheap optically but are actually truly being disrupted.

Jack: What struck me the most about this is no blue dots in the bottom. So there were no, like, extremely expensive companies according to intangible value that ended up being the biggest winners.

Kai: Right. Not in like the bottom, like, yeah.

Jack: Yeah, the bottom of the whole chart. Which means that it was measuring value right, is what I think it means. Because there were no -- there was -- maybe there were some that were slightly expensive according to intangible value, but there were none that were like extremely expensive according to intangible value that then ended up being like an Amazon type company.

Kai: Right. I mean, to be clear, this is just the top 100 stocks, the 100 largest stocks at the period in time. So there could’ve been like some other names that would’ve been there. It just would’ve been too many dots, so I didn’t wanna show 1,000 dots.

Jack: Yeah. But still, it still is pretty interesting.

So this next exhibit gets at the idea of looking at the same four quadrants, but now we’re looking at return by quadrant, right?

Kai: Yeah. So all I wanted to do here was just make sure that the results generalized. The previous chart showed just a 10-year period from ‘07 to ‘17. Now I wanna look at the full sample.

But the setup’s the same. And so what you see is the stocks that both metrics agreed were cheap did the best, 4.2% annualized returns. Stocks that they both thought were expensive did the worst, -5.1. When there was disagreement, intangible value won. So in other words, the quote unquote expensive disruptors, the Apples and Amazons, did well, 2.8. And then the value traps, the stocks like the Macy’s, right, that looked cheap on traditional but not on intangible value, did -1.6%.

Now, a couple interesting findings. So first of all is the fact that the intangible moats do appear to matter, so that’s good. The second thing we find though is that when there’s agreement, it’s actually more powerful than when just intangible value thinks something, right? And so that kinda goes back to this idea that I think we’ve discussed on the podcast in the past, that potentially there’s a role for these two metrics to be complementary with each other, right?

That the real red flag is not when something’s just expensive on intangible value. When it’s also expensive on tangible value, when it’s expensive on both metrics, that’s like pretty concerning, right? So I think that’s another interesting takeaway from this exhibit to me.

Jack: So this next exhibit, we’re actually taking this now and we’re applying it to software. So we’re looking at an intangible value score, and you’re putting some of the names here that a lot of people would recognize and looking at whether they’re cheap or expensive on intangible value.

Kai: Right. So yeah, what we found up to here -- we spent most of the paper talking about the historical disruptions, like going through the past waves, thinking about what metrics do and do not work, going through Teece’s framework of complementary assets to understand how to think about moats.

So now what we do is we kinda say, “Let’s bring it all to the present. Let’s put it all together in a way that would be applicable to today.” We’re going through the current disruption with software stocks having sold off significantly due to AI disruption fears. What this chart shows, exhibit 21, is software stocks that are down thirty percent or more in the past one year.

So these are not like your software stocks that are like... This is not all software stocks, but it’s the ones that are considered losers because of AI generally over the past year. And what I did was I showed the distribution histogram of the intangible value scores for these names at this point in time.

And so the first thing you see is that the average is positive, right? It looks to be about like point three or something, suggesting that these stocks, which are in a large drawdown -- they sold off like thirty percent with the market up thirty percent over the past seven or eight months, so a sixty percentage point spread -- that these stocks may on average have been oversold. Shoot first, ask questions later.

But the second thing you see is pretty decent dispersion and, more importantly, dispersion on the left side, right? So you look at the left tail on the red. This is actually really important because this is not usual, right? You don’t usually see this much dispersion on the left side of companies that are basically value traps, companies that are down eighty percent or something but are still expensive on these metrics, right? That’s generally an unusual thing to see.

And again, don’t read too much into these logos, they’re chosen for illustrative purposes. But like you do see that look at like HubSpot versus Salesforce. Both of these are kinda CRM type companies. Salesforce is looking at least on these metrics on the cheaper side, whereas HubSpot is looking more expensive. So you do see some dispersion even within comparable names, which I think is worth noting.

Jack: It is interesting, by the way, too, because to your point, like GoDaddy, registering domain names, building websites. Like I saw Wix, I think, is laying off a bunch of people. Like it makes sense where these are based on what you would think in terms of what their moats are. Like it would seem like a GoDaddy would not have a very strong moat.

Kai: Right. And so there’s another exhibit I have in this paper where I actually use this framework of the four intangible pillars and say like, what are the moats that a company might have, right?

So like accumulated business logic, like embeddedness in customer workflows, customer relationships, regulatory compliance burdens, right? And so one of the insights here is that, and this is pretty intuitive, I think most people know this, is that more enterprise-facing software companies that face the largest enterprises will tend to actually have wider moats because just the switching costs are a lot higher.

These things are a lot more embedded. There’s systems of record. The compliance requirements are so much more onerous than say consumer-facing things like GoDaddy, for example. Or I’m thinking of Duolingo here too, where it’s a little easier for a random person to just switch off an app, right.

And so I think these things do correlate, and you can look through each of the four intangible pillars and actually have this in an exhibit and look at kind of, you can sort bullet point by bullet point to say, “Hey, for a given company X, where does it score on these four intangible pillars and then on each of the, say, 20 or so sub-points within those pillars?”

Jack: So there’s basically two ways, if I’m a software company, there are two ways I can succeed here, and I think you get at this in the paper. One is I can have a moat, which we’ve talked about. The other is I can really embrace AI. So as we get into the rest of the paper, those are kind of the two things you’re looking at, right? In terms of the way to differentiate these ones that might succeed from the ones that won’t.

Kai: Yeah. So if you remember, the last paper we did together was on -- it was called like “AI Adopters: Beneficiaries of the Boom.” And the idea there was to find companies that are positioned for AI adoption, right? ‘Cause presumably they, over time, if AI becomes a thing, would separate from the laggards. And that was like a one-dimensional thing.

What I’m saying now is let’s take David Teece’s framework and say, “Hey, look, that’s obviously important, but it’s not the only thing that matters,” right? The fact that Walmart figured out e-commerce was important, but they also had a lot of other things going for them, right? That allowed them to survive, relative to any other legacy company that was trying to become an e-commerce company. And that is the complementary moats, right?

So I’m adding to the AI adoption lens this additional lens, which is really the remaining parts of the intangible value four-pillar framework. So that together you have these two things that sum up to the intangible value framework, but I decompose it in an interesting way where I have AI adoption and then everything else. And you can kind of look at those things as almost distinct lenses.

Justin: One of the points that you brought up in the paper was some of the firms that actually survive this disruption -- AI might actually help improve the margins and the profitability of those companies. Can you just explain the logic in your thinking there?

Kai: Yeah. Look, I mean, the idea is that obviously there’s a ton of dispersion, right? So in the software sector, there are some companies that are aggressively adopting AI, others that are doing not much, some that have defensible moats, others that do not.

And so there’s gonna be some winners, there’s gonna be some losers. But when the whole shakeout happens and all is said and done, the companies that do survive are actually in an interesting position. Because you think about what is the biggest cost center for these companies, right? It is the production of code. That’s like the main factor of production for these companies, at least from a cost standpoint. And bringing to this the additional complexity around stock-based comp -- so stock-based compensation has become this big, big flashpoint amongst the investor community because these companies have been really kind of liberal users of SBC for a long time.

But now that their stocks are down, investors are kind of like, “Wait a second, what’s all this? Why are we doing this?” Right? Because software engineering talent is expensive. And so to the extent that AI has the potential to reduce the labor intensivity of software code, that’s actually potentially gonna alleviate this bottleneck, allow these companies to do what they’re currently doing but at a fraction of the cost. Or said differently, for a fixed number of employees, be a lot more productive, right.

And so you could conceive of an argument that, actually contingent on surviving, which of course is a big if, AI is actually a boon to these companies.

Justin: Talk about this next chart, the sparkline AI adoption score versus AI exposure, and sort of what you’re seeing. You mentioned the dispersion, but, like, software companies are way up to the right, so they’re obviously embracing AI. But, yeah, like, how should we be kind of thinking about this, would you say?

Kai: Yeah. So this chart, this is exhibit 26, is comparing two different analyses I did over different points in time. So on the x-axis, it shows exposure of a given sector to the technology of AI, right? So in other words, to what extent can large language models in theory impact the day-to-day tasks of a company?

Exposed sectors are of course software, banking, hardware, pharma. Non-exposed sectors are like, I don’t know, food and staples retailing or whatever, right?

And this is on a production side. And then on the y-axis, we see the adoption score, right?

And so what this chart is showing is the extent to which these companies are leaning into AI, whether it’s they’re hiring AI employees, getting AI patents, repositioning their businesses for AI. And then what I did here was I showed all the different industries in a scatter plot, and I draw a red line, which is basically the line of best fit, the average.

And so any company, any sector that’s above the red line, in theory, is adopting AI more aggressively than they are exposed. So they’re kind of your early adopters. And anyone below the line is actually kind of lagging. Relative to how exposed they are, they’re really not doing enough.

And so yeah, software is actually the outlier here in terms of having the highest exposure, but they have by far the highest adoption, right? So they are, as I said earlier, truly recognizing the extent of the threat and on average, at least -- not everyone’s doing it, but doing the best they can to respond to it, right?

If you -- I have another chart in my paper showing AI job postings, and software and software services and IT services are like by far the highest sector when it comes to the hiring of AI talent.

Justin: So this next one is sort of like the sweet spot where we’re coming back into software, and now we’re looking at the software companies that are high or low based on AI adoption and high or low based on intangible value.

Kai: Right. So all I’m doing here is putting these two dimensions together. So remember, one dimension was how much AI adoption a company has, and then the other dimension was the everything else section, which is your intangible value score minus AI adoption, so we’re not double counting. And what I do here is I show, in this case, this is for the software sell-off.

So all the software stocks that have fallen thirty percent or more over the past year. So these are kind of your software losers, or perceived to be losers, based on the AI disruption. And what you can see is the upper right is where you wanna be.

Upper right is the sweet spot. These are companies in the upper right that in theory have a strongly defensible business due to the strong brand, human capital, network effects, and complementary IP, yet are also leaning into AI, so they kinda have the full package. And then in the lower left are the opposite, so companies that have very limited intangible moats, and they’re not doing enough in AI, and then there’s the kind of middle category too.

And so you do see that there are a handful of companies in the red, and then you see some companies in the middle, and then the vast majority of names are in the kind of not-so-good section, right? So the point just being that there’s a ton of dispersion, right? That there are plenty of companies out there that have good pre-existing businesses, plenty of companies out there that have good AI, a fewer number that have both, and many that have neither.

Justin: And then you have the next chart, the high dispersion of disruption scare stocks. So there’s a lot going on in this one, but explain to us what we’re sort of looking at here.

Kai: Right. So I already observed earlier that software stocks have huge dispersion, right? So if you go back to the very beginning of what I mentioned, software stocks are down thirty percent as an index, right? The IGV is down about thirty percent peak to trough. But there is huge dispersion, right? Like GoDaddy, Salesforce down fifty to eighty percent. Adobe, some of these other names are down big. And so you see that, and then you also see the thing I showed a few slides ago, which was that the intangible value scores have a wide dispersion as well, with this big left tail of companies that are potentially value traps.

So the third element of dispersion I wanted to bring into the mix was this idea of historically, when you have these events happen, what happens over the next year to returns, right? Because this is another way of measuring dispersion. Now, obviously today it’s software stocks, but if you go back through time, it would have been newspapers, it would have been retailers, right? Those would have been the exposed sectors.

So in order to build a metric of who are the folks in the crosshairs of disruption, what I did was I said this. I said, “Let’s look for historically companies that were in both exposed sectors.” So remember the definition from before: technologically exposed sectors that are also over a trailing twelve months in a thirty percent loss, right? So this is guys in retail who are also -- the market thinks are gonna be losers because they’ve punished them.

So the question becomes, all right, so when the market thinks you’re gonna be disrupted, are you actually disrupted, or do you tend to bounce back, right?

And what’s interesting is, first of all, the medians. So what I show on this chart is the distribution of next one-year returns for this group of stocks -- that’s in the red -- relative to, in the blue, all stocks. And you can see that the medians are basically the same, six or seven percent.

And you go to average, it’s about the same too, depending if you’re doing arithmetic or geometric, whatever. The point being that the fact that you’re down on price alone says very little with regards to where you’ll be the next year, right? So just because software stocks are down today doesn’t mean we should all panic and say, “Oh, they must be zeros.” It has very little informational content with regards to the median expected return over the next year.

But if you look at the distribution, this is where things get interesting. They’re very different, obviously. So the blue line looks more normal, right? All stocks tend to have a more normal distribution, whereas disruption-scare stocks have a really fat-tail distribution, super wide.

So in fact, it looks like 10% of these stocks go on to double over the next year versus 3% for the full market. 16% go on to lose more than half versus 7% for the full market. So in other words, the dispersion of winners and losers is so much wider for these guys, these beaten-down, disrupted stocks, both to the upside but also to the downside.

When technology comes around, it reshuffles the deck, and the ball’s back in the air and everything’s in play, right? And so I think that’s a really important point to add to this idea of dispersion, right? So there’s dispersion in terms of historical returns, future returns, and then current valuations.

And all these things have just kind of blown out due to the indiscriminate selling and panic around AI.

Justin: Well, I think this kind of really ties back to the value traps versus the moats. Like, clearly in this case, you wanna avoid the 16% or so that lose more than half, and try to be on the right side of the chart with the ones that survive, right?

Kai: Right. And discernment, like the ability to discern winners from losers

Justin: Right...

Kai: matters more in a time like today than it did historically or in an insulated sector.

Justin: So what about -- what happens when we apply the intangible value factor to those high-scare dispersion stocks? Exhibit twenty-nine.

Kai: Yeah. So just to be clear, I’m using intangible value as a way of illustrating this point, because I already built on it, but the point’s more general. The point’s more broad.

Jack: Mm-hmm.

Kai: But I’ll explain the exhibit first and then we’ll get to the point.

So what we see here is the returns, which we already saw, of the intangible value factor applied to the full universe in the blue, and then the exposed sectors in the red. And then what we do in addition is to do disruption scare stocks. And so remember, the full universe -- think of like a bullseye, like a dartboard. The full universe is the widest circle. Exposed stocks are a subset of that, and insulated being the other part of the subset. And then within exposed stocks are disruption scare stocks. Stocks that are both exposed and down thirty percent, right? And also, at least at the time, investors perceiving them to be the losers.

And what you find is that the return for this factor as applied to that final segment of disruption scare stocks is much higher than in the other wider circles, right? So in other words, when you apply the intangible value factor to disruption scare stocks such as like software stocks today -- but it could have been newspaper stocks in the past -- the ex-post returns have been higher, right?

And what is this saying, right? This is kind of your Grinold and Kahn if you go back to your finance textbooks, which is that ultimately, dispersion is something that allows you to amplify your edge. So for a given edge, if you have a lot of dispersion in the market, that means that your winners may do better and your losers, your shorts, will do better too, right?

And this is also something people talk about in the context of venture capital or private equity. Like one reason why people love VC and private equity historically is because they’ve had high dispersion, right? And so therefore, a given edge can be amplified over higher absolute return, right?

And again, this principle generalizes from intangible value to any edge. So anyone who has an edge in picking software stocks in disruptions, right? Which is again a big if. But if you think you have a framework for picking that, that works not all the time, but more importantly also specifically in times of disruption such as today with software stocks, then this is actually a great time to be doing stock picking because high dispersion is one of the things -- we don’t know -- we think the mean will be about the same, but the dispersion will almost certainly be higher.

Will likely increase the returns to being able to separate winners from losers.

Justin: I love that idea of the dispersion and the edge kinda coming together, and that the intersection -- that if you have high dispersion, if you have any edge, that’s when it can become possibly amplified.

I think that’s a very great way to think about it, and just conceptually I’ve never heard anybody explain it that way, so that’s pretty, pretty cool stuff. It’s

Jack: interesting too, by the way, just on the human side of things, like thinking about the great software stock picker -- they’ve probably got their best opportunity set they’ll ever see in their career right now.

Kai: Yeah. Yeah, ‘cause not only do they have an edge presumably in software, but there’s just a crazy amount of dispersion and there will be many companies -- many shorts will go to zero, and many of the longs will be multi-baggers, right? Companies that were sold down 60% that may go on to be the next Walmart of their sector, right?

So very interesting time to be a stock picker in software these days.

Justin: Kai, your research is always super impressive, and we’re very honestly privileged and appreciative of you coming on with us and our audience and kinda working through all this stuff.

If there is kind of a main takeaway from all this research, what would you say it is?

Kai: Look, I think for many of these companies, say software stocks, I think the takeaway is that the code is not the moat, right?

For many of these companies, code is one of the many things they do. But we as investors need to look beyond that to ask the question of what other intangible assets or just moats in general do these possess? Because if you go look historically based on all the work through prior disruptions, it turns out that these other complementary assets are potentially the most important indicator of which companies will survive and ultimately thrive through a disruption, right?

Now, of course, AI adoption is important too. But I think that doing this research over the past month or so has given me a kind of a deeper appreciation of the extent to which customer loyalty, brand equity, human capital, network effects -- these other moats for software in particular -- are more important than maybe we initially thought when it comes to being able to survive a paradigm shift in the way technology works, right?

And so simply saying, “We’re gonna buy stocks ‘cause they’re cheap” -- I don’t think that’s sufficient. A cheap stock once you price the earnings, I don’t think that’s sufficient. Saying, “I wanna buy these stocks because they have the most AI adoption” -- now, obviously I’ve talked about that in the past, and I do think that’s important, but I think it’s insufficient.

I think really what’s come together in my mind more, having done this research and especially bringing in the work of David Teece, has been the extent to which complementary assets -- brand, human capital, IP -- are really quite important as we think about which companies will ultimately be winners and losers long term from the current sell-off.

Justin: Good stuff. Thank you, Kai.

Kai: Thank you.