← Excess Returns 
Loading summary
Transcript118 lines
- [00:00]
A
What I wanted to focus on more was this trillion dollar question, which is the one that Saty Nadella was asking, which is will we actually, are we actually seeing a sustainable increase in demand that you know, companies and users are truly earning tangible returns on investment from their investments in AI. Companies that not only talk about it but also can link it to economic gains, achieve even better outperformance of 4.8% and then the companies that can mention specific ROI do even better at 5.2% annualized outperformance. What you also see is that there's basically no relationship between how much companies are investing in AI. This is on the non infrastructure side and what valuations the market gives them. Laggards and early adopters are basically trading at parity with each other despite one group being kind of way ahead in terms of using AI and the other group being not even paying attention to this stuff at all.
- [00:56]
B
Kai, welcome back to Excess Returns. It's good to see you.
- [00:59]
A
Good to see you too.
- [01:00]
B
Our audience is probably very familiar with you at this point because I think this is maybe your sixth or seventh time on. You've done some guest hosting for us, but today we wanted to have you on to talk about the latest investor update and what I would call like quasi research paper you put out. And this is going to be a very interesting discussion because I think a lot of investors are interested in this topic. But for our audience that may not be for or may or may not be familiar with Kai, your founder of Sparkline Capital, you run a number of investment strategies through ETFs focusing on intangible value. And I think you've carved out a space in the sort of in the market for looking at and analyzing companies through the lens of intangible value through machine learning and natural language processing. And a lot of your research is sort of oriented and focused on that specifically. And the kind of the question, the article that you put out, which was titled AI adopters beneficiaries of the Boom, sort of is like asking this question that is in a lot of investors minds, which is who are or where are the beneficiaries coming from, from AI. And so you tried to take, you know, a systematic approach to looking at this and kind of thinking about in the context of market history and also just you know, how to go about identifying where these sort of opportunities are. And before we kind of get into it, this piece is available at etf.sparklinecapital.com where you can go download this report. This conversation is going to have Some charts and images from what Kai put out. So if you're not listening or watching on YouTube, you might want to pop over there because it's. Make it more from it. So anyways, Kai, hopefully that's okay. Opening. Sorry to be so long winded, but we're obviously big fans of everything that you're doing over there. I wanted to start with where you started the paper. We're just going to kind of follow this chronologically through. But you had a quote from Satya Nadella, CEO of Microsoft, and let me just read it and I want you to explain why this is important. And he said for this not to be a bubble by definition, it requires that the benefits of AI are much more evenly spread, not just the economic growth driven by capital expenses. So can you kind of explain why you started with that?
- [03:20]
A
Yeah. So this is a quote from Satya Nadella, of course is the CEO of Microsoft when he was at Davos a few weeks ago. And what's interesting is that he, he addresses the question of is AI a bubble head on, you know, and he says, look, I can give you a framework for telling you whether it is or it is not a bubble. Basically, if it is a technology that only helps tech companies and say in 10 years we're just talking about how, you know, Nvidia backed OpenAI, which backed Microsoft and so on and so forth. And then when it just kind of flows within that tech ecosystem, that will happen a bubble, right? That if the growth is driven by capex and a very narrow set of stocks, that's a bubble. Whereas conversely for it to not be available by definition, what we need is we need for the technology to diffuse more broadly through the economy into non tech stocks. He gives an example of a one use case. He says if we're talking about a miracle drug that came out in 10 years that was brought to market due to AI accelerating clinical trials, that's a win. Right? And he says, and he goes on to say his view on whether we are or not in a bubble, and his view is that we are not in a bubble because he does think that quote it's happening, that the. Let me get his full quote. Actually he says it's happening. I'm much more confident that this is a technology that will in fact build on the rails of cloud and mobile, diffuse faster and bend the productivity curve and bring local surplus and economic growth around the world, not just economic growth driven by capital expenditures. That's a narrow point in time. Right. So he does, obviously, given, you know, he could Be biased, but he does see a future in which technology does diffuse. Now you asked why this is important. I think this kind of frames the entire piece, which is this piece actually builds on something I wrote in October last year called Surviving the AI Capex Boom. And that paper, as you mentioned, Justin, draws on history and talks about the historical implications of the infrastructure buildouts that we've seen and adopted a more kind of bearish tone on a variety of factors. This paper is kind of the sequel to that, focused on more where the opportunities lie. And what I wanted to focus on more was this trillion dollar question, which is the one that Saty Nadella was asking, which is, will we actually, are we actually seeing a sustainable increase in demand, that companies and users are truly earning tangible returns on investment from their investment in AI? Because if so, then they'll continue to do so, to continue to invest in AI, driving revenues to the infrastructure builders, and if not, they'll abandon their pilots and we'll have been in an overbuilding cycle and it will have been a bubble. Right. So the whole framing of this paper is trying to tackle that question head on and say, what evidence do we have in terms of where we are in the cycle and if adoption is real or just kind of a hype?
- [06:08]
B
So one of the charts you have in here is this classical technological diffusion curve. So based on, you know, where we are today, where do you think we are on that curve and trajectory now?
- [06:21]
A
Yeah, so this is kind of that classic S curve adoption diffusion graph from Everett Rogers. I adapted it a little bit for the current situation. So basically I say, look, the, the technology diffuses in five stages. The first is what I call the infra infrastructure phase. It's like the first thing you need to do when you're building any technology, whether it's the Internet or AI, is build the Rails. And in this case it's building out data centers full of GPUs to actually run the technology. Once that's done, then the next question is, will people actually buy it? Right. And maybe this should be more in parallel. But for whatever reason, it tends to be the case that these occur in series. And then from there, after the early adoption phase, there's like the lagarit come into play. That's the early majority, the late majority, and then the true laggards start to adopt. And that's when you kind of know that the technology has reached saturation. And so these S curves can be traced through history, from the electricity, the railroad, the cell phone, et cetera. And you kind of see these things evolve over multi year periods. There's some numbers attached with this too. So in general, when Rogers put together this framework, it was 2 1/2 percent, kind of the 2 Sigma tail were considered the innovators or infrastructure builders, the Nvidias, Googles, Microsofts of the world. And it was around the, you know, next 13 and a half percent to get you to 15% was the early adopters and then the other whatever 85% were the laggards if you look at the actual adoption Data. So the US Census maintains some data on this. Around 10% of businesses are currently using AI in production in the US so it kind of puts you around that, that point. But more to the point though, I mean if you look at just, just the way that the vibes have shifted and how Nadella has been framing the issue even just two weeks ago is, you know, for most of last year the question was can we build it right? Can we get, can we, can we build enough GPUs? Will the next generation of Nvidia's GPUs actually work and can they produce them on scale? Can we network them? Can we get enough capital together to build out these massive data center sites? Who's going to pay for all this stuff, right? Those were the big questions that were asked and they're, you know, more or less solved, right? So we're now kind of on the path to spending several trillion dollars in building data centers over the next five years. So then the next question becomes, okay, great, so we can build it now the next question is will they buy it, right? Will there actually be sustainable end user demand for all the stuff we're building? And if the answer is yes, that's great, maybe it'll have been worth it. And if the answer is no, then we're obviously in trouble. And so that's what I think everyone's asking you can look at as the best barometer of where we are in the cycle. Look at Oracle stock, right? Like it's fluctuated a lot. Where last year, you know, in the summer when they announced that big deal with OpenAI went up 30% in overnight, you know, investors were at the time very euphoric around any investment into building out more AI capacity since then, the stock's gone down 50%, right? So I think the vibes have shifted a bit and investors are starting to say, hm, maybe that's not the best play and we should be more thoughtful about where this money goes and whether it will actually generate ROI or not. And if you look at the history of the, these booms, you do find that the kind of stock performance of each sector kind of mirrors, you know, the, where you are in the cycle. So in, in the telecom boom, for example, you know, global crossing, even AT&T, they, those stocks outperform not just the stock market, but even the, the Nasdaq for the first few years of the boom until they didn't. Right. So in the very beginning, infrastructure firms, you know, make they, they do well because, you know, people are spending money on building stuff and then later that the leadership rotates into early adopters and then eventually to the rest of the market.
- [10:00]
B
So the big question I think everyone is asking here is whether that spend, you know, is going to generate a positive roi. And there's, you know, people on both sides of the fence. Right. So how did you kind of attack tackle this, you know, looking at this with this AI driven ROI taxonomy framework?
- [10:19]
A
Yeah, look. So I think the challenge that I saw is that most of the data on this trillion dollar question that underpins the entire boom comes from small surveys. If you remember, there was a survey under the guise of MIT where they found, I think 95% was the headline of AI pilots fail. And that was a kind of spooky headline. But if you actually dig into what that was, survey of a handful of enterprises, so not necessarily statistically significant and certainly not representative of the broader universe and not just to pick on them. There are examples on both sides of the fence that are wildly, optimistically, wildly pessimistic, but all just based on surveys of a select few CEOs or CIOs or whatever. And so that's no way to kind of go about answering what is again, this super important question. The approach I chose here, and it's not perfect, but I think is at least gets around some of the sample selection issues, was to try to look at the information in publicly available earnings calls. Right. So this is information we all have access to each year. Thousands of companies have their CEO CFO on these calls, doing prepared statements and then answering analyst questions around various topics. Of course, AI is one of the major topics that these folks are forced to address today. And what I did was I said, let's parse these documents with an eye on one fact, which is we know that these CEOs are incentivized to talk a big game. They're incentivized to overhype how they're investing in AI because it will help the stock price. Everyone knows that. So how do we combat that and what I decided to do was to try to look not just at mentions of AI, which we do look at, but in addition look at more narrow, restrictive subsets of calls in which the CEO or CFO reports quantified improvements in revenue costs, productivity margins, risk, capital efficiency due to AI. And so an example could be we achieved a 20% reduction in headcount required to service a fixed amount of customer service requests. Right? So that's a number, a numerical 23%. 20%. You need to see those numbers, which in general gets around the issue of vague kind of puffery. And then we have the most narrow category of the three is AI driven return on investment or roi, in which case we don't only have the amount of increase in revenue, say we also have that against how much the company spent on building the AI systems in the first place. Right? So just because you can generate $20 million of AI driven revenue is great, but if it costs you $100 million to build a system, that may not be good ROI. Right. And so we want to see companies saying, hey, look, we did this project and you know, we actually had good roi. So that implies they're likely going to be scaling it up for air, right? That's the best possible outcome. So again, there's three different tiers to the taxonomy. The most broad being, do companies even talk about AI usage at all? And two, the more medium one is, are they also talking about how they are achieving productivity gains or cost savings on these calls? And then third, are they getting that? In addition, are they able to frame that as a positive ROI relative to a cost base?
- [13:40]
B
LifeLock. How can I help the IRS that.
- [13:43]
A
I filed my return, but I haven't?
- [13:45]
B
One in four taxpaying Americans has paid the price of identity fraud.
- [13:48]
A
What do I do?
- [13:49]
B
My refund.
- [13:50]
A
I'm freaking out.
- [13:51]
B
Don't worry, I can fix this. LifeLock fixes identity theft guaranteed and gets your money back with up to $3 million in coverage.
- [13:57]
A
I'm so relieved.
- [13:58]
B
No problem. I'll be with you every step of the way. One in four was a fraud paying American. Not anymore.
- [14:05]
A
Save up to 40% your first year.
- [14:06]
B
Visit lifelock.com podcast terms apply this episode.
- [14:11]
A
Is brought to you by Indeed. Stop waiting around for the perfect candidate. Instead, use Indeed sponsored jobs to find the right people with the right skills fast. It's a simple way to make sure your listing is the first candidate. C. According to Indeed data, sponsored jobs have four times more applicants than non sponsored jobs. So go build your dream team today with Indeed get a $75 sponsored job credit@ Indeed.com podcast. Terms and conditions apply.
- [14:40]
C
And I'll put up exhibit four here because this is really cool because you hear so many people talking about, like, the benefits of AI, but you don't really see these exact examples from companies. Like the one everybody always uses is the meta one that they're doing better job of targeting advertising, which you've got in here. But it's just interesting to look. This is, this is across the entire economy. It's completely different type of company. So these things are real and they are out there, right?
- [15:01]
A
Yeah, look, I mean, this, this Exhibit 4 here has a small sampling of the many. You know, you can see there are hundreds of mentions of AI throughout the calls, even over the past year. And so I just chose a few things that looked interesting to me. And obviously this seems representative, but, you know, you could, you could kind of go through it yourself. Examples would be like, Estee Lauder sees a 31% increase in ROI from the North American media campaigns. You have here. United Healthcare in an auto plant that makes auto parts. You have Phillips in customer support. Our AI agents have reduced support costs by 80% on target. They're using autonomous inventory robots to cut time required to pick an item by over 45%. Public Storage. Right. They're like a REIT or they're like a storage. A real estate company. They have. Their AI staffing systems have reduced labor hours by more than 30% while also reducing employee turnover. You have RTX using it to build missiles. Bank of America, they have a coding use case as well. C.H. robinson, which is a logistics company, is using AI to respond to quotes more quickly and win more business. Pfizer is talking now about using AI to help save on the R and D side. They think $500 million is a good target for them. So you can see across the economy, from finance to industrials to healthcare, use cases far beyond what we see just kind of in technology and encoding.
- [16:25]
C
Yeah, this is really cool. For me, as someone who can. Somewhat of a skeptic. Not a skeptic of AI, but somewhat of a skeptic of, like, is all the spending worth it? And we don't know that. But like, something like public storage, like, that is not what you would expect in terms of it's not a tech company. I mean, they're running storage facilities and they're seeing that kind of benefit. So that, that does just show how this is going to diffuse, I think, throughout the economy.
- [16:44]
A
Yeah, I mean, this definitely supports Nadella's thesis.
- [16:47]
C
So in terms of where these gains.
- [16:48]
B
Are coming from, you have a good.
- [16:49]
C
Pie chart we're going to bring up here. But you can think of you mentioned before, they can come in a lot of different ways. They can come top line in terms of revenue, they can come in terms of costs. So can you talk about this pie chart we're seeing and just what this is showing us?
- [16:59]
A
Yeah, so when I designed this, this model to kind of go through the, this LLM, to go through the earnings fall transcripts, I kind of defined about six different buckets of things that I thought could be areas of uplift. So on the top line, as you mentioned, revenue growth, and then there's cost savings, productivity gains, margin expansion, risk reduction, and capital efficiency. It turns out that about three categories explain the vast majority of mentions. And those three categories are revenue growth, which is increased revenue demand, pricing, power conversions or retention, productivity gains, improved cycle time output for employee operational efficiency, and then cost savings, headcount reduction, automation savings and lower IT assurance and cost. So at least at this current point in the cycle, those three things comprise most of the hundreds of mentions that we're seeing so far. Yeah.
- [17:51]
C
And in this next exhibit we're seeing, it's also just across companies. These mentions are rising. Right. In terms of companies that are mentioning better ROI from AI.
- [17:58]
A
Yeah. So you're seeing two things. So first, you're seeing just an increase in the number of companies that are talking about AI. Right. So like the universe is roughly fixed in terms of, you know, there's a few thousand companies that are doing these calls, and then the number within that has been increasing a lot of, especially over the past few years since ChatGPT. But then what's even more interesting is that within that subset of companies that are talking about AI, the share that have actually not just been talking about it, but also been reporting tangible gains has been increasing. And so if you look at the middle category, so companies mentioning AI driven economic gains that has increased from about 10 to 32% of the of that of the companies mentioning AI at all. So in other words, one third of companies who are talking about AI are also able to link it to a specific gain in terms of a numerical advantage they're achieving via AI. And then the most strict subset is the companies mentioning ROI as a share of total AI mentions that's increased from basically nothing to 7%. So that's a much smaller share by definition. But we're also seeing kind of that trajectory increase pretty steadily through time. And this is by the way from 2017 until 2026. So roughly over the past decade.
- [19:13]
C
And I assume the investor relations departments are probably saying, like, make sure you mention AI a little bit more. I assume, like, especially some of these companies that are not necessarily directly associated with it.
- [19:23]
A
Yeah, I think it's, you know, the share of companies mentioning AI all has increased, but I think also what we're finding is that the share of those companies who are talking about specific, you know, 52% increase in cycle times, those things are also increasing. And it could also be, you know, the SEC and various others are starting to kind of wise up to the idea of like, what they call like AI washing. Right. Window addressing. Right. And so companies, I think, are also being told, hey, look, if you have a specific example, like actually talk about that, don't just say, oh, yeah, we're doing a digital transformation, which anyone can say, of course, which is cool about.
- [19:56]
C
What you're doing, because you're actually separating those firms out. Like, you're not going to be talking about words like cycle times and stuff. If you're just making this stuff up, you know, to put it on your earnings call. Like, you can kind of separate the people that have legitimacy behind what they're doing.
- [20:07]
A
Yeah. And look, and the nice thing about this too is obviously companies will put the spin on they want, but, like, these are public statements. Like, these are. There's like a fraud component to this. If you say something, if you lie on these earnings calls, like, you're gonna go to jail. So there's an element of like some, you know, reputational enforcement at least, or legal reinforcement behind what's being said on these calls. Like, they're not gonna outright lie, at least.
- [20:27]
C
So as we move to Exhibit 8, this is kind of cool.
- [20:29]
A
You are.
- [20:29]
C
You'll take a look here at the companies by category, and then are they generating excess returns? So can you talk about that?
- [20:36]
A
Yeah. So one of the things I've talked about for years actually at this point has been that companies that talk more about AI in their earnings calls, in their 10Ks or really in any way have done better than the markets. And that is shown here in a 3.2% annualized excess performance. What's really interesting is that companies that not only talk about it, but also can link it to economic gains achieve even better outperformance, 4.8%. And then the companies that can mention specific ROI do even better at 5.2% annualized outperformance. And by the way, this is for the universe in general. If we look at just like sector neutral or exclude tech companies, this pattern still exists. So this also exists amongst the kind of pure industrial old economy names that we went through earlier.
- [21:22]
C
And it seems to me like as we move to this Exhibit 9, which is by industry, it seems to me like this is something that could benefit almost any industry. Going back to your examples before of know, public storage, I mean this is something that we would expect to benefit almost any industry. Right?
- [21:35]
A
Yeah. So I think there's two dimensions to this chart here. So the first thing is the cross industry distribution. Right. So first what you see is, you see that there's some sectors that are more aggressively adopting AI and others that are not. And it kind of lines up with intuition. The most aggressively pursuing AI are software, semiconductors, hardware, technology, media, entertainment, telecoms, commercial and professional services, healthcare and financials. Right. Which, which lines up. So I think it was maybe two years ago I wrote a paper on AI and what I did at the time was I attempted to define for each company and then at the sector level, the company and sector's exposure to LLM uplift. So in other words, based on all the different day to day tasks that you do at company X, what percentage can be automated by AI? Right. So for a call center, that's like 100% and then for like a, I don't know, stonemason, that's like zero, at least at the current point in time. And what's interesting is that we are then able to wrap that up to the industry level. So I have this exact same chart from two years ago looking at the potential increase from AI efficiencies. And it lines up really well with this. Right. So it is the case that if you are a CEO of a company, you're like, yeah, I'm a software company and I could be disrupted by AI as these guys are kind of learning, figuring out I better invest in this stuff on average. Whereas if I'm a food, beverage and tobacco company, that doesn't really have too much to lose from AI, maybe I'll just kind of wait and see. So that kind of lines up. Although I did notice a couple outliers. So if you do like a scatter plot of those two things I just mentioned, you find a few outliers, notably regional banks, while they are really exposed to AI, aren't really doing anything on average. And same with pharma. Pharma was disappointing. Right. So despite Nadella's use case about the AI using R& D efficiencies, I'm not really seeing as much investment from the pharmaceutical R and D side as I would expect to see, given the potential for a lot of transformation there, probably due to the kind of regulatory red tape would be my guess. But they seem a little bit behind where they could otherwise be. So anyways, that's the first dimension which is AI Diffusion is uneven across the economy. One dimension is a core sector, the other, even more interesting, at least to me, is the within sector component which is like, yeah, the average software company has more proclivity towards AI, but even there there's a distribution and then you go down to even the most like old economy industrial sectors. And maybe most companies aren't doing anything, but there's going to be a few early adopters that are really leading the charge. And the distribution here is pretty right tailed, I guess, where, let's take a number, 85% of companies are basically doing nothing within a given sector, but then 15% are really going after the opportunity. So it's a handful of early adopters and then a bunch of laggards, which really mirrors the distribution that we saw in the S curve. That diffusion model talks about how you go from the innovators to the early adopters to the laggards, that kind of power law distribution. We are kind of seeing that in the data, which is pretty interesting.
- [24:39]
C
I'm just curious as an aside, because you mentioned software like that's been in.
- [24:42]
B
The news right now.
- [24:42]
C
Those, those stocks are just getting absolutely destroyed. Like I think Ritholtz's latest podcast is called like the software Slaughterhouse or something. They had, they had a tech analyst on there. I'm just wondering, do you think that's overdone? I mean, do you think there is a risk like that? The sales forces of the world and these big companies, do you think people are just going to come with like AI and build, build these things like with Claude code or something like that? Or do you think that's overdone or do you think that is a huge real risk to these big SaaS companies?
- [25:06]
A
Yeah, I think it's a nuanced question. I would say that, yeah, all SQL, these companies do face significant headwinds from AI and so it does represent a real disruption threat to the sector. And so the fact that the sector has sold off is actually directionally correct. I'd say that there are a couple things going on, like they had maybe elevated valuations going in, but that's, you know, that's, that's fair. Now I think where there's an opportunity is that there's a lot of dispersion, right. So like everything's kind of been indiscriminately sold off. You know, you mentioned Salesforce, they're kind of more system of record. And then you have like Adobe in there, you have, you know, a bunch of other stocks that do totally different things. If you have like SaaS or software in your name, you go down 50%. That's how it's gone. But obviously these companies are going to be, you know, you know, are going to end up, you know, weathering this episode very differently. And so I think it's, you know, a couple dimensions matter. Of course it is case by case. But like, you know, the firms that are more kind of the system of record, as I mentioned, are probably more insulated from disruption than those who are effectively dashboards or wrappers on the underlying data. You know, firms that are more kind of just like silos where you just kind of focus on one vertical are probably more at risk than those that are more kind of generalist in use. There are probably a few variables like that that matter. But again, I think it comes down to case by case. So like coming out of this, this sell off, you know, I think there are certainly going to be a handful of high quality tech names that will have been oversold and it will be, you know, great opportunities. Looking back. Oh yeah, that was a generational buying opportunity to buy company X. But yeah, I mean as a, as a basket, you know, I think the fact that these stocks are down is not like directionally wrong. I mean that, that does feel like. Yeah, what's supposed to happen when you have, when, when coding becomes basically in again kind of infinite abundance.
- [26:55]
C
So back to the paper. You bucketed companies into three groups. You had infrastructure, early adopters and laggards. So could you first just talk about how you did that, how you figured out what was in what group?
- [27:04]
A
Yeah, so it was a two stage process. So the first thing I did was I carved out what are infrastructure companies versus not. So infrastructure companies are anything that's any company that's integral to kind of the scaling and deployment of AI. So that could be like Nvidia on chip side, it could be memory or storage, it could be networking, the data center, REITs, the model companies themselves. So like the software, the companies like OpenAI or Anthropic that train the alms, that would count as being an infrastructure play too, you know, Oracle and the AI cloud. So that, that is one category. And then what I said was, who is not in that category? That's most companies. And then within that not infrastructure category, I split into two halves or two categories. Let me call it on those that are early adopters in AI. So think about like the bank who has actually, you know, investing aggressively in trying to, you know, incorporate AI into their business, and then everyone else who are the AI laggards. Right. So there's kind of these three categories.
- [27:58]
C
Now and how is the percentage breakdown in terms of total firms between the three categories?
- [28:03]
A
Yeah, so actually I should probably mention in terms of the data I use to do that. So it's the. When I define what is an. What is kind of AI investment, like which companies are investing in AI versus not. What that is, is. It's a combination of a lot of different data sources, like looking at patent, patents and trademarks, employee profiles, you know, narratives in various documents, like we discussed kind of putting it all together and compiling all this information to kind of get in a sense of which companies are truly investing in AI versus not. And that's the dimension on which continuous dimension on which I created that sector chart, but also the one in which I define the cutoff for early adopters versus laggards. But yeah, to your question, around the percentage, again, it kind of falls in that. It's almost like it's very kind of like textbook. It falls in that same distribution that we saw earlier in the diffusion model where about. I think it was about 80% of companies are considered laggards. So in other words, they have made no meaningful strides towards incorporating AI into their business. Of the remaining 20%, about 10%. So half of it is infrastructure companies and then the other half is early adopters.
- [29:09]
C
So let's look at this exhibit 11, because this is kind of cool. So you actually took the infrastructure, the early adopters and the laggards, and then you looked at them on a sector basis. And I guess, like infrastructure is sort of what you think it would be. It's mostly technology companies. But I thought it was very interesting, like behind the scenes in terms of what you saw here.
- [29:23]
A
Yeah, so infrastructure is 85% tech companies. Laggards is kind of the opposite. So it's only 15% tech companies. The other 85% are industrials, financials, material energy, et cetera. So that's also kind of as you'd expect. What was interesting to me was the middle category, the early adopters. So the early adopters category, they have about 33% in the same tech sector, as I mentioned, technology, communications, discretionary. So they're tilted towards tech, but they're not overwhelmingly tech. In fact, the majority of early adopters are actually these AI leaders in old economy sectors, the examples, many of which I gave earlier. And so it's kind of an interesting mix of companies where it's, it's, you know, not it's tech, but it's not really tech. Right. Or at least not in a traditional sense.
- [30:07]
C
So the early adopters also are the ones reporting the most roi, which I thought was interesting.
- [30:12]
A
Yeah. So this is exhibit 12. What I did was I said now that I had this, this basket of stocks that are defined as early adopters based on my kind of ex ante measure of adoption. Right. So going into the year, we know who we can define which group. And then now we can look over the course of that year. So going into 2025, now over 2025, what actually are these companies saying about their ROI in AI driven projects? And what we find is a pretty strong correlation. So we find in early adopters, about 81% are talking about AI, 51% are reporting AI gains and then 17% AI ROI. On the opposite side of the spectrum, the laggards, you only have 38% talking about AI, 9% both the gains and 1% ROI. And then infrastructure is kind of in the middle where they're reporting 26% report gains and 9% report ROI, which is interesting.
- [31:09]
C
Right?
- [31:09]
A
Because obviously these are the picks and shovels. So they're the guys who make money when they sell stuff to the folks using AI, of course. But it turns out that they may not actually internally in their own operations, be using AI as aggressively as the users. I guess that's partially definitionally, but it's a kind of interesting observation. Right. To the extent that AI is truly transformative and allows us, as Nadella predicts, to bend the productivity curve and increase our margins, who will actually benefit? Obviously the infrastructure companies will benefit in as much as they're selling stuff to make this happen. But really it's really between the laggards who comprise 80% of the stocks, and then the early adopters who likely will take market share from the laggards to the extent that they are utilizing AI to, to drive these gains on the.
- [31:57]
C
Infrastructure plays, is a lot of this because they're building it out for themselves? Like I'm always interested to think about, like when they actually use it for themselves versus building it out for others, like, what's the difference? So like for instance, the meta example before, you know, I think meta seemed pretty positive. ROI when they spend it on advertising, but then Maybe the ROI is not going to be as good. Like if you're just building out infrastructure for other people. Like, is that an important delineation?
- [32:17]
A
That's a really interesting question. I mean, you have, yeah, you have met as a good example of a company that is, you know, among the top four spenders on AI infrastructure and they're basically spending it all for themselves. Right. Microsoft's in a tough position because they're kind of trying to do both and that's creating some conflicts. I think in the last earnings call we saw that come to bear. And then they're, you know, I guess the pure play, like core weave. Right. They're purely a, a kind of arms dealer. Right. And, and so, you know, and again, it comes down where the value accrues will depend on a few factors, I think competition being the big one. Right, right. You know, if it is the case that the ability to sell AI models or AI chips is a commodity, then yeah, there's not going to be that much profit pool to go around. You know, Meta is a unique company because they, I don't want to use a monopoly word, but you know, when it comes to social media, like in their given apps, whether it's like WhatsApp or Instagram or, you know, the Facebook blue, I mean they have a pretty significant share and so, you know, they're able to, we maybe get ROIs that other companies that were more kind of in commoditized businesses would maybe not be able to achieve. So they're kind of a unique case. We all have that dream trip we've been wishing we could go on. But too often life or usually price gets in the way. That's why Priceline is here to help you turn your dream trip into reality. With up to 60% off hotels and up to 50% off flights, you can book everything you need for your next adventure. Don't just dream about that next trip, book it with Priceline. Download the Priceline app or visit priceline.com and book your next trip today. Go to your happy price. Priceline. TaxAct is here anytime you want to easily file your taxes.
- [33:57]
B
TaxAct is here for the early birds who like to knock them out as.
- [34:00]
A
Soon as the season opens.
- [34:02]
B
And for the procrastinators who like to.
- [34:04]
A
Wait until the very last minute. TaxAct is also here for the midlers.
- [34:08]
B
Who file right in the middle of tax season.
- [34:11]
A
No one ever talks about the middlers, but taxact sees you and taxact respects you.
- [34:19]
B
Taxact.
- [34:19]
A
Let's get them over with.
- [34:23]
C
And one of the interesting things here is. So the infrastructure companies maybe haven't been getting as much ROI as the early adopters, but the market's been rewarding the infrastructure companies the most. Right, in terms of valuations, is that right?
- [34:32]
A
Yeah, I mean, and look, I mean they have been getting roi, just not from the, not from like AI efficiencies. It's not like they're like, oh yeah, you know, I was able to cut my head count more. It's, they're, they're getting ROI because they're, you know, you know, building out cloud infrastructure and then selling it for a profit. So they are getting roi, but just not, you know, purely from the use of internal AI tools. Just to be clear.
- [34:56]
C
So, so what are we seeing in exhibit 13? This is kind of a scatter plot. What is this telling us about adoption versus valuation?
- [35:02]
A
Yeah, I think this is a pretty important chart. So what I show here, this is a scatter plot. On the x axis is an AI adoption score from low to high and on the Y axis is the valuation. So things like price to book, price to sales from low to high, and what you basically see is a cloud. So first of all, on the X axis you see a lot more companies in laggards than early adopters, which again is that 80:10 distribution. What you also see is that there's basically no relationship between how much companies are investing in AI, this is on the non infrastructure side, and what valuations the market gives them. So in other words, companies that are investing more in AI are not being rewarded with higher multiples than those who are just, you know, punting and waiting for things to play out in 10 years. Which is surprising. Right, because that's kind of the opposite of what we saw on the infrastructure side.
- [35:49]
C
Yeah. What's interesting about this is, so it's not being. What we saw earlier is they are getting the roi, the early adopters, but we're not seeing in valuation. So within that might be some sort of opportunity, right?
- [36:00]
A
Yeah, I think there's a couple of reasons why, I mean, look, part of it is just that the, it's taking, it takes time for things to trickle through to the bottom line. You know, likely what happened was that in 2022, 23 companies, these, these are large companies, right? They start doing these pilots and you know, testing things out, first of all with like vague KPIs and then over time say, hey, this is starting to work. Let's do it on a bigger scale and bigger scale. Right. So it may take some time for this to really push the bottom line. And we are seeing examples, some of which I mentioned showed earlier, of companies that actually have had huge quarters, huge quarterly earnings, beats due to, and they contributed directly to AI. But for the most part, companies are still in that early phase. Now, keep in mind it takes a long time to build the expertise to do this. So even though they have yet to, you know, they're still kind of early in that J curve, many of these early adopters, they're still three, four years ahead of the laggards. So even if it turns out that these companies are getting huge wins, say starting in a few years for the laggards to now catch up, they're now five years behind. Right. So I think really what I'm interested in here is not so much I care about the absolute, but I also care about the relative thinking a little bit about how far are the early adopters ahead of their competitors, which are these laggard companies.
- [37:17]
C
And just in terms of the valuation of the infrastructure place, I also think about, like, how much of this is algorithms or people like to invest in simple things. Like, you know, if you want to invest in AI, because you hear that all the time, like among retail traders, like, I want to invest in AI. Like, these simple things are not going to think at the level you're thinking in terms of, like public storage could be a beneficiary. They're going to think about the obvious names. And maybe early in the adoption curve, that's what happens. Everybody puts the money in the obvious names. Is that right?
- [37:40]
A
Yeah, I think it's two things. I think first of all, it is what you're saying that people, the first order of plays are more easier to think through than the second order plays. So first order beneficiaries do better. They have done better. So when you see a Stock that's up 100%, you're like, all right, that's because of AI. So maybe I'll buy some and it'll go up another 100% if AI continues. And so there's kind of the obvious names that are first order beneficiaries of AI, and then there's the kind of less obvious, more under the radar second order ones that, you know, first of all are less obvious by definition, but also like, you know, it goes back to the cycle ship thing. Whereas in the beginning it does make sense to invest in AI, invest in infrastructure. Right? You did want to buy telecoms in the beginning of the telecom boom, and then you wanted to stop doing that and buy other companies. Afterwards. Right. So, you know, even, you know, I wouldn't want to be so cynical as to say people are just dumb. It's also that people know how to play a cycle and they, they go into things that will think, they think other people will invest in and then, you know, we will see if this plays out. But historically what we would see at some point is a rotation of leadership away from infrastructure to the broader AI beneficiaries.
- [38:46]
C
And we can see this in Exhibit 14, you have the valuations by category here and you can just see how much more expensive the infrastructure plays are. They're trading at a significant premium to the market and early adopters and laggards are trading at a discount.
- [38:58]
A
Yeah, so this goes back to 2015, so about 10 years. And at that time AI infrastructure stocks were trading at 20 to 25% premium to the market. So like a little bit, but not that much. Today they're above 75%, so almost twice that of the market. So it's been, you know, a nice time to be an AI infrastructure stock. These companies have done well, right? Nvidia's earnings are up indeed a lot. But on top of that they have a double whammy with multiple also expands because people expect it to continue. And then in contrast, you see AI early adopters and laggards went from being kind of at parity with the market and then are now at a slight discount, largely just because by definition these things need to sum to one. So infrastructure stocks becoming more expensive, these guys become relatively cheap. But again, the interesting thing to me is not so much that, it's more like the cross sectional relative value comparison here, that laggards and early adopters are basically trading at parity with each other, despite one group being kind of way ahead in terms of using AI and the other group being not even paying attention to this stuff at all. And you have to assume that at some point these things will separate. And at some point the AI early adopters, if AI becomes as transformative as we expect, two, that they will actually drive meaningful gains and separate from their laggard peers.
- [40:11]
C
What I think is cool about what you do is you're able to, the way you analyze data, you're able to find benefits from these early adopters before they're necessarily showing up in financial statements.
- [40:19]
A
Right.
- [40:19]
C
Like you're probably not seeing it in EPS figures yet, but you're seeing that they are getting those benefit in that roi.
- [40:24]
A
Right. I mean that goes back to what I was saying before. Like, you know, we looked at which Companies are making investments in AI patents, for example, or AI talent. And then it turns out that the next year we also start to see them talk about, oh yeah, you know, we actually did this thing and it worked out, we had this like 25% ROI, right? And then, you know, the next logical step would for that be that, that to translate into higher EPS and then into higher multiples as people start to wake up to this, you know, new narrative.
- [40:51]
C
So you drew a parallel in the paper. This gets back to your capex boom paper as well. You drew a parallel to railroads and fiber optic cable in terms of the builders subsidizing the users through over capacity. So can you talk about how that works?
- [41:01]
A
Yeah, so if you go back to like the railroad boom of the 1860s, you know, hundreds of railroads were set up to capitalize on this opportunity in connecting the United States transcontinentally. They do that and they all go bankrupt. Right. And ultimately what you find is that the railroads did create tremendous value for the US into the gdp, et cetera. All that value was not captured by the builders or the rail, but actually by the users and customers. Right? So like if you're, you know, a retailer or something shipping goods using the rail and then same in the dot com boom, you saw, you know, tons of money and tons of fiber optic cable being laid out by these telecoms, the global crossings of the world, the AT&TS. And again the telecom index fell 95% and hasn't recovered in the dot com bust. These companies basically, you know, were not able to capture any of the value and instead it was, you know, the, the, the early adopters meta, Google, Netflix, that captured all the value. Right. So basically what we're in, in both cases we found that infrastructure was kind of a utility. It's an important thing to have. We need the Internet, we need to fabric the cables. But ultimately that's not where the profits accrue. The profits accrue at the application layer, right? The companies like the metas, who are actually building on that. Right. And, and part of the dynamic is not is, is due to this subsidy component, right? Where it's like, you know, it's the capital cycle. You have all these companies, these telecoms, let's say, you know, building, building, laying fiber up at the cable. And each one is rationally saying I should build more of this because there's more demand. And that's not incorrect. But if everyone does it, then there's a collective oversupply. And we know basic economics dictates that when Supply is too high for demand. You then have prices fall. And prices indeed fell. You know, I think it was 85% of fiber optic cable was unused and the prices fell 90%.
- [42:42]
B
Right.
- [42:42]
A
Which is really bad for telecoms and really good for the Netflixes of the world. And so this has happened over and over again. I think that as we entered the AI cycle, we should be highly attuned to how this could play out. Right. You could end up in a situation like we have in every example I could find historically, where the technology is indeed transformative, but investors in the companies actually building that transformative technology ironically make no money doing so.
- [43:06]
C
And you're probably going to wonder why I'm asking you the same question I asked you when you came on last call last month, because I do ask questions over and over again, but I've learned people don't watch the same YouTube video videos. But I want to ask you what the other side of this argument, which is this idea that this is different than railroads and it's different than telecom because there's intelligence embedded in this infrastructure. Do you think there's a case to be made for that?
- [43:24]
A
Yeah, I think certainly. I think it's quite possible that in 10, 20 years, we look back and say, yeah, AI was like the last invention, nothing will be the same. But in that state of the world, which I don't think is the base case, by the way, but if that is the state of the world we end up in, then whether or not this stock act went up or not kind of isn't the point we'll be discussing. How can we beg our AI overlords for more ubi? That's kind of the question and that's the other world. But I think your base case has to be more. All right, this will be like any other technology which is built on each other. Electricity must have been so transformative when it first came out, as was the Internet. And these technologies all build on each other. Every single time you do see displacement of human labor, people lose their jobs. The steam engine cost people. John Henry lost his job or whatever, but you then end up with new jobs that are created in their stead. So, you know, like any of these productivity enhancements, whether it's a calculator or Excel or any of these things, they will change how the world works. And you know, AI, you could. So I guess the question you're asking, Jack is, is AI kind of one of many in that long line of technologies, or is it like somehow, you know, structurally different and. Yeah, so my answer would be, you know, I think we got to play for the base case being that it is, you know, a very powerful but still, you know, merely mortal technology that enhanced productivity, like many other things, less so that it'll like completely just, you know, create unlimited abundance in society. And if that were the case, then again, we'd have other problems to worry about.
- [44:55]
C
Last one for me before I hand it back to Justin. I was to ask you about this idea of overcapacity and where we are on that line right now. Because if you listen to like the CEO of Nvidia or something, they're telling you like, we've got demand for many, many, many years out into the future. Like, do you think we've got a long way to go? Or do you think maybe just the people inside of this don't necessarily recognize, like where we are on our way to overcapacity?
- [45:15]
A
Yeah, well, I think the two things. So first of all, if you're Jensen, of course you're going to say that, right? Like PB stupidness is anything different. But, but second, I think a lot of it is time horizon, right? Like if you think about, imagine we were sitting here in the.com boom, right, like at the time these guys were saying the exact same thing as they should. They see a huge backlog of customers lining up to buy AI chips. They see, or fiber in that case. They see the constraint is we can't produce enough supply to meet demand, which is of course what you see in the early stages of the build out. But the thing is that the construction is now underway to build out, I don't know, another $5 trillion of data centers or whatever the number is over the next few years. That's kind of already done. Conclusion, these projects are already kind of in the ground being built, right? So that's happening. And so again, the question becomes not, hey, is there near term demand? Because of course there is. But if there's no ROI for any of these projects, the companies will say, you know what, I'm going to cancel my plan. You know, I'm canceling my order. I wanted these ships, but actually it turns out that these things are useless. So yeah, I don't want them anymore. Right? And that's the situation that these companies could end up in where they, you know, build for a 5, 10 year, whatever horizon. And you know, the demand, you know, that you, that that seemed to be there is ephemeral and that, you know, if there is no ROI in these, in these, in these projects, then yeah, you know, I think the other thing to Note is like ChatGPT people always point to the exponential growth of the product, which is impressive, but only 5% of those users are paid, right? So only 5% of people who actually use ChatGPT are saying, I'm going to spend my hard earned dollars to use this product. 95% are saying, yeah, this is cool, but like, I don't know if I want to pay for this. Right. And so that's the other piece too, which is, it's not that it's not just usage, it's also, well, can they monetize that usage?
- [46:59]
B
Well, it's that. And like you pointed out in the paper, like, what is the. Is there an actual moat around that business? And you have a chart that shows how much jockeying and how much competitiveness there is among these LLMs and sort of the adoption curve and how ChatGPT was, well, OpenAI was way ahead and now it's kind of neck and neck with some of these other ones.
- [47:17]
A
Yeah, I mean, this was an interesting thing to look at. I just putting the data together. So the chart starts in November 2022 when ChatGPT came out. And at the time, ChatGPT was based on a model called GPT 3.5. So that was their flagship model. And at that time, this is all on a log scale. The Y axis is. They were just miles ahead of their competitors. And people were like, wow, this is incredible. This is so much better than whatever Google or Meta have, whatever. Like, this is going to be a monopoly, like every other tech thing. It's going to be like, you know, Facebook all over again and we have a monopoly and they're going to be able to charge whatever they want. And this company is going to be worth hundreds of billions of dollars. Right? And that was true for, you know, a couple of generations. But then you start to see competition. This is the capital cycle, right? If there's a profit margin, that's an opportunity for everyone else, right? The people see high profits and they say, cool, I want that too. And they competed away. And so what we saw was, you know, Google got their act together. They were struggling. They actually invented the transformer model that underpinned GPT, but they sat on it for years for kind of bureaucratic reasons. And then their founders came back and said, they went founder mode. And they said, all right, well, we're going to turn this chip around. And the latest Gemini is on the frontiers, as good as the latest OpenAI model. And then you saw anthropic which was I think a spin out some guys who left OpenAI and they had backing from big tech companies like, like Amazon. They caught up like Claude is as good, if not better. I actually probably spend more of my, you know, LLM tokens at this point now on Claude than I do on, on OpenAI, which is kind of interesting because I was for a long time a very prolific consumer. Oh, OpenAI tokens. And so yeah, you have those three models on the frontier and then you even have like some, you know, so Deep Seq would be a good example of a company that's like a, came out of nowhere. It was like a Chinese hedge fund, quant hedge fund, made this model and open sourced it. And it's not as good as the closed source competitors, but it's almost as good and it's a lot cheaper and a lot more efficient. Right. So you're starting to see a lot of, as you said, jockeying for position. What seemed to be a kind of locked market where there is like one clear dominant player has now been fragmented into a lot of smaller ones. Right. And that gives of course these small companies much less leverage when it comes to, you know, how they inter, how they deal with their customers.
- [49:31]
B
New Year, new me. Cute, but how about New Year, new money? With Experian, you can actually take control of your finances. Check your FICO score, find ways to save and get matched with credit card offers giving you time to power through those New Year's goals. You know you're going to crush start the year off right. Download the Experian app based on FICO scoring model offers an approval not guaranteed. Eligibility requirements and terms apply subject to.
- [49:55]
A
Credit check which may impact your credit scores.
- [49:56]
B
Offers not available in all states. See experian.com for details. Experian.
- [50:01]
A
We heard you. Nine years of bring back the snack.
- [50:04]
B
Wrap and you've won.
- [50:05]
A
But maybe you should have asked for more. Say hello to the hot honey snack wrap. Now you've really won. Go to McDonald's and get it while you can.
- [50:18]
B
So now let's sort of talk about how this, these findings, I guess, you know, how they're manifesting themselves at the index level. And specifically, you know, I think you asked a question like for the s and P500, you know, we've established that, you know, infrastructure stocks in this category that you've identified, you know, they are trading at a premium over the early adopters and the laggards. But you kind of looked at the exposure of the laggards or early adopters and infrastructure across various market index and then across the portfolios that you also run. So talk about what that looks like both at the index level. And then we can get into, I think some of how your portfolios sort of are reflecting this and the way that you select stocks.
- [51:10]
A
Yeah, so I think this is a really important chart, especially the s and P500. Right. Because it's one thing if it's like, okay, so there's a bunch of like tech insiders who are like doing this crazy thing in AI, like, whatever. But the problem is that most average investors, they're just like, all right, I'm going to buy the s and P500. That's just the index market cap weighted, passive, whatever. Well, it turns out that that is an implicitly huge bet on the same thing we just discussed. So we all know that the S and P is 33% MAG7 stocks, right? But it turns out that if you add to that the rest of the infrastructure names, it gets to 46%. So almost half of the S&P 500 is in AI infrastructure stocks. Effectively, it's a massive bet on the sustainability of this AI capex spending. So if this works out, then yeah, that's amazing. But like, it's a pretty big bet to take half your money, right. As an average retail investor, retiree, whatever, in the S and P and put it into, into this high stakes game, you know, Game of Thrones thing that like Sam Altman and, and Elon Musk and all these guys are playing. Right. It's, it's pretty wild to see that. And so then your question would be, okay, maybe that's a little bit crazy. You know, if I'm a financial advisor, I'm like, all right, maybe I should diversify my client's money if I'm an individual to say the same thing, I should diversify a bit outside of these infrastructure stocks. So what do you do? Right. There's three standard things people do. These are what I call the alternative indices on this chart. They maybe go to Russell 1000 value or value stocks. Maybe they go to equal weight or small cap stocks. Maybe they go outside of the US where AI is less prevalent, into MSAI or emerging markets. Well, it turns out the problem is that that's kind of just going to the opposite extreme. So now you have no more infrastructure risk, but you have laggard risk. 44% of value stocks to 66% of IFA and 52% of S& P. Equal weight are in laggards. So now you're saying all right, well, I'm just going to completely go the opposite way and effectively be short AI basically and buy the guys who are just doing nothing in AI and hope it is a bubble and it just kind of goes away. Right. In which case that will have worked out. But the thing is that what happens if AI is the real deal is transformative? Well, in that case, you've just missed out on the entire thing. Right. It's like not buying any tech stocks in the dot com boom. Yeah, I guess you miss the bus, but you also miss the next 10, 20 years of productivity enhancements. And so I think that's a pretty difficult thing to do. And that's kind of why we honed in on early adopters being kind of this third category saying, look, it's kind of a false choice. You're not forced to choose between S and P or value stocks. There's a middle ground. There's a third option here that can potentially give you the best of both worlds, and that is exposure to AI economic gains if these technologies do diffuse, as we hope, but without the valuation risk and without the CapEx risk that the infrastructure games have. And so this is specifically focused on our funds. It's an investor letter, but it shows the positions of our funds being kind of heavily concentrated.56 to 59% in these AI early adopters, which we believe are currently positioned, you know, as a way of, you know, kind of in the best, most balanced way towards the AI opportunity, which, you know, also comes with lots of risks too.
- [54:15]
B
Yeah, so, so what's interesting is like itan, you know, you have a chart in here that shows, and you know, while you're talking to this, maybe because we don't want to gloss over it, I mean, you're selecting these stocks based on, you know, your intangible value factor. So that's an important way that you're getting at these. But what's also interesting is you're showing sort of this shift from, in the portfolio, you know, from where you had 40% infrastructure in June of 2021 to today, you know, 22% infrastructure. And then you show the, the AI early adopters going from 37 to 59. Yeah, so 59. So you're, you're sort of have seen this shift adding more early adopters in the strategy over time and. Right.
- [55:00]
A
Yeah. I mean, I think this speaks to the valuation framework. Right. I think it's important to note that this positioning today is not like a thematic override. It's not me saying top down. I think that this is going to work out better. It's just the result of a organic bottoms up process using intangible value. So traditional value will naturally rotate away from things that go up in price holding constant fundamentals and vice versa. But the problem with traditional value is that they don't give, they don't work well. Those metrics like price to book ratio don't work well in intangible intensive sectors or companies. Which is kind of the entire point of what we're talking about with AI. And you know, would have been the same with the Internet as well, right. So you kind of need a more holistic definition of value, which is intangible value, to apply this framework now into these markets. So what's interesting is you go back to the 2021 when the fund, this is itan, was incepted it was actually overweight infrastructure names, right? So you know, we were labeled like by the, by the Wall Street Journal, like the, the, we were called the, the value investor buying big tech stocks. It was kind of late, a non consensus play. Play for a, a non consensus play for a value investor to be overweight. Not just big tech but also infrastructure. Right. So we had 40% in AI infrastructure, the S&P had 33%. And, and that was, you know, companies like Nvidia and such. And then what happened is that over time as the AI narrative took off and these stocks went up in, in value, right. Just naturally our model said, hey, we still like it, Nvidia, it's a great stock. But like look, it's prices now, whatever times it was before, let's start selling it down, right? So the models naturally rotate out of the stocks that, that did really well. And the valuations increased where we saw the chart when it went from 20% to whatever 75% premium. We sell out of those companies and then rotate into other stocks. But we still want to lean long AI, right. So like the idea is that, you know, these models are always looking for stocks that are cheap relative to intangible assets, which include not just you know, brand and IP in general, but IP linked to innovative technologies like AI robotics, you know, genetic editing, things like that. Of which AI is of course the most important one today. And so we're all. So, so the models are still leaning along AI. It just happens to be a case that they're, they've found themselves leaving infrastructure and finding more value in these early adopters. Which makes complete sense given that, you know, as I showed you earlier, they're trading at the same discount to the market as the laggards.
- [57:17]
B
Yeah. And talk about this idea of what your definition of AI yield is and why that concept is sort of important when looking at these stocks through this type of framework.
- [57:30]
A
Yeah, look, so, so this goes exact back to exactly what I was saying. AI yield, it's like a dividend yield. Right. If you're a dividend investor, you go around looking for stocks with the highest dividends relative to dollar invested, not the highest dividend in absolute. Right. You're just getting big companies then. Same here. Like I want to find companies or the models want to find companies that are investing heavily in AI and that can, as you can see in this table, take the form of AI employees, AI patents, AI trademarks. But if you just went only for that number in absolute, you'd just be buying Google. Right. So what we want to do is we want to find companies that are producing a lot of those assets relative to the dollars invested. And what's interesting is you look at these three columns, look at the S and P, the first column, you see that the numbers aren't that high. And the reason why is not that they don't have. Not that the max 7 and the biggest companies don't have a lot of AI capital. The problem is that they're just really expensive and so it offsets that. Conversely, the alternative indices. Think about value stocks as an example. Value stocks have really cheap valuations by definition, but they don't have much AI. And so you get a low ratio as well. And then in the middle is kind of the point I was trying to make around the best of both worlds. The early adopters are more specifically the stocks that we're focused on with the ETFs. Just by definition, we're trying to find companies with high AI yields, among other things. And so the numbers are much higher. Right. We're looking for companies that are truly investing in AI but aren't already priced in by the markets.
- [58:53]
B
Yeah, so. So ITAN is the US focused domestic strategy and DTAN is the international strategy.
- [59:01]
A
What, you try me backwards.
- [59:04]
B
Yeah, yeah, right. But yeah, I had that correct.
- [59:10]
A
Right.
- [59:10]
B
Yeah, yeah. So, but, but what did you, I guess with the international markets, like was there, did some, did anything jump out as being a lot different? I mean, I think the trend is also the same, if I'm correct. Like it's been migrating more. I'm not sure if that was in the paper or not. But what, what did you sort of what jumped out at you with sort of international stocks that have finally got off the mat here and the recent performance is good, but was there some differences between that and the US that you found?
- [59:39]
A
Yeah. So first of all, you know, most people kind of automatically know that AI is kind of a US thing and maybe Chinese too, but aside from China, Most innovation in AI happens in the U.S. right? We all, we all kind of know that. But what's interesting is that you know, and if you only focus on infrastructure companies, that's true, right? All the best, all the best model providers are in the US of course, and maybe China. But if you are willing to do what I did and expand the definition of AI beneficiaries to also include second order beneficiaries, companies that are positioned to use AI to drive better profits. You may have noticed that some of the companies in that, in the examples of earnings calls reporting AI ROI were non US companies. Right. You know, AI doesn't discriminate with regards to where you are in terms of your user of the technology. Right. So if you're able to do early adopters, there's actually plenty of good companies out there in the non US space that are have this exposure. So 56% is about the same as what we saw in the us. So yes, we do see fewer AI infrastructure companies in the non US space. Like Stevens Energy takes lines in there. You know, a lot of the companies that we all know about like asml are kind of expensive, like it's too obvious. And there are probably more laggards, right? European companies, they may not, may be less innovative, but they have other things going for them, right? They have maybe stronger brands, like luxury products, things like that. And so you see more consumer discretionary. So you know, you kind of find other ways to find edges outside of AI. But yeah, the big thing is that is the early adopter space. And then if you look within that blue category in terms of how the sector distribution looks, in the US it's more tech companies and in non US it's more healthcare consumer discretionary industrials. And that again reflects largely just the composition of the underlying universe.
- [61:20]
B
So what do you think as we kind of move forward here? What are the biggest things that you're going to be paying attention to in terms of assessing and trying to, you know, keep your sort of finger on the pulse of whether or not we're getting the benefits we're accruing the benefits of REL from AI relative to like all these costs. Like if you were to sort of try to get at that or monitor that over time, like one idea might be like and just thinking about this work that you've done, like tracking like the number of companies that are moving from laggards to early adopters, like that might be one thing you look at just the migration of companies. But what would you, how would you respond to that?
- [62:00]
A
Yeah, I think that's right, Justin. And that's one thing to look at. Just continuation of the trend that we saw in the earnings calls. CEOs continuing to report more and more and bigger and bigger, more scaled ROI from AI. If we see that asymptote, that would be a bad thing if we see it or plateau if we see it continuing to increase and the S curve eventually asymptote around 100%. That's good. Right. So we want to see more and more uptake there. The thing I'm doing as well is, you know, I'm not representative of the average enterprise. I'm a small company and I am technical myself. But I, you know, as we discussed, Justin, I spent my entire weekend playing around with cloud code, like trying to do kind of cutting edge things and getting a sense for, you know, how good are the newest tools. Because one thing to remember is that like, you know, this technology moves so fast that, you know, Claude code, for example, which is, you know, what people have been so scared about this past week in terms of display facing software companies was not, you know, was not even that good three months ago and with the latest release became amazingly good. Right. So there was a huge step change in terms of how good the models were and we can assume that continues too. And so it's important for us, you know, as well just to kind of be using the products ourselves and understanding kind of how much progress does the technology have, even if we don't expect, you know, a large massive enterprise company to be using cloud code immediately tomorrow. But like, you know, we can kind of extrapolate out and you know, assume that with some sort of lag that will filter in down the line. Right. So I think it's a combination of looking at what people, folks are saying they're achieving also going back to the point we made earlier around, wanting to see that show up in the actual financial statements. Right. So it's one thing to get a, have a AI pilot, deliver an roi, but it's still like, you know, a rounding error on their overall income when it starts to actually move the numbers, as it has in some cases, but not, not more broadly. That will be interesting. In addition to be kind of on the ground using these tools ourselves as technologists getting a better sense of where we can kind of expect the curve to bend moving forward.
- [64:04]
B
Excellent stuff, Kai, as always. I really value your work and I really enjoy seeing sort of how all this kind of comes down to the investment strategy and portfolio level and is wrapped up inside an etf. So great stuff. And I want to say I believe Let me just pull it up here because I think I am correct. Yes, Congratulations on the five star Morningstar rating for ITAN. I'm not sure about DTAN yet, but hey, we'll take one five star at minimum, right?
- [64:34]
A
That's right. Yeah. DTAN's only been around for one year, so we haven't have the time frame yet.
- [64:39]
B
Yeah, but good stuff, guy. All right, man, Appreciate it. Thank you very much.
- [64:42]
A
Thanks Justin.
- [64:43]
B
Thank you for tuning into this episode. If you found this discussion interesting and valuable, please subscribe on your favorite audio platform or on YouTube. You can also follow all the podcasts in the Excess Returns network@excessreturnspod.com if you have any feedback or questions, you can contact us@excessreturnspodmail.com no information on this podcast.
- [65:03]
A
Should be construed as investment advice. Securities discussed in the podcast may be.
- [65:08]
B
Holdings of the firms of the host Tires Matter. They're the only part of your vehicle.
- [65:12]
A
That touches the road.
- [65:13]
B
Tread confidently with new tires from Tire Rack Whether you're looking for expert recommendations or know exactly what you want, Tire Rack makes it easy, fast, free shipping, free road hazard protection, and convenient installation options. Go to tirerack.com to see Tire test results, tire ratings and consumer reviews, and be sure to check out all the special offers. Tirerack.com the way Tire buying should be for their clients.