Episode 311: AI and Investing: Surviving the AI CapEx Boom with Kai Wu

What happens when a world-changing technology attracts trillions of dollars in investment and how do you avoid being on the wrong side of the boom?

On this episode of the Women's Money Wisdom podcast, Melissa Joy is joined by Kai Wu, founder and chief investment officer of Sparkline Capital, for a thoughtful, data-driven conversation about AI and its real investment implications.

Rather than focusing on hype or fear, this discussion centers on capital allocation, specifically what happens when massive amounts of money pour into infrastructure. Using historical parallels like the railroads and the dot-com telecom buildout, Melissa and Kai explore why transformative technologies don’t always translate into transformative returns for the companies building the underlying systems.

You’ll hear what “CapEx” actually means, why AI requires unprecedented spending on chips and data centers, and how today’s market concentration around the largest tech companies creates both opportunity and risk. 

This episode is especially helpful if you’ve benefited from recent market gains but are wondering about concentration risk, diversification, or how to think more strategically about AI exposure in your portfolio.

In this conversation, you’ll learn:

• Why a transformative technology does not automatically mean a great investment
• What capital expenditures (CapEx) are and why AI infrastructure is so expensive
• How past technology booms rewarded users more than infrastructure builders
• Why highly capital-intensive companies often underperform over time
• How S&P 500 concentration may create hidden portfolio risk
• What “AI early adopters” are and why they may offer overlooked opportunity
• How to think about diversification without abandoning innovation

Whether you’re managing significant wealth or simply trying to make smarter long-term decisions, this episode offers practical insight to help you think clearly — without hype, panic, or binary thinking.

Follow Kai and read his research: https://sparklinecapital.com

X: https://x.com/ckaiwu

LinkedIn: https://www.linkedin.com/in/ckaiwu/

The previous presentation by PEARL PLANNING was intended for general information purposes only.  No portion of the presentation serves as the receipt of, or as a substitute for, personalized investment advice from PEARL PLANNING or any other investment professional of your choosing. Different types of investments involve varying degrees of risk, and it should not be assumed that future performance of any specific investment or investment strategy, or any non-investment related or planning services, discussion or content, will be profitable, be suitable for your portfolio or individual situation, or prove successful. Neither PEARL PLANNING’s investment adviser registration status, nor any amount of prior experience or success, should be construed that a certain level of results or satisfaction will be achieved if PEARL PLANNING is engaged, or continues to be engaged, to provide investment advisory services. PEARL PLANNING is neither a law firm nor accounting firm, and no portion of its services should be construed as legal or accounting advice. No portion of the video content should be construed by a client or prospective client as a guarantee that he/she will experience a certain level of results if PEARL PLANNING is engaged, or continues to be engaged, to provide investment advisory services. A copy of PEARL PLANNING’s current written disclosure Brochure discussing our advisory services and fees is available upon request or at https...

Transcript

Welcome And Episode Setup

SPEAKER_01 0:00

Welcome to the Women's Money Wisdom Podcast. I'm Melissa Joy, a certified financial planner and the founder of Pearl Planning. My goal is to help you streamline and organize your finances, navigate big money decisions with confidence, and be strategic in order to grow your wealth. As a woman, you work hard for your money, and I'm here to help you make the most of it. Now let's get into the show. If you aren't talking about AI, um, you must be living in a cave because it seems to be coming up so often, whether it's the investment world, you know, thinking about jobs and the labor market, and in so many other use cases. And so we're going to be talking all about AI and investment implications to AI on today's episode. And I'm thrilled to have a friend and an investment analyst and portfolio manager joining us, who's going to be diving in. He's done a really great research report when it comes to AI. So Kai Wu is the founder and chief investment officer at Sparkline Capital, which is an investment management firm applying state-of-the-art machine learning to uncover alpha in large unstructured data sets. Kai, I'm going to let you define alpha during the episode. Prior to founding Sparkline, Kai co-founded Kaleidoscope Capital, which was a quantitative hedge fund in Boston and worked at GMO, where he was a member of Jeremy Grantham's asset allocation team. Kai, welcome to the podcast.

SPEAKER_02 1:36

Thanks so much for having me on.

Guest Intro: Kai Wu

SPEAKER_01 1:39

I'm so glad you're here. It's great to have a friend on. And, you know, our listeners, some of them are investment geeks, and others are just trying to get their feet under the ground when it comes to money. That's why I said, like, you know, we may be defining some of these concepts. Um, but I'm going to refer everyone who's listening to a research paper that you publish for Sparkline, which is called Surviving the AI Capex Boom. Um first of all, thank you for doing this great research and joining us on the podcast. Can you give us a little bit of background of why you think AI is important and why, and then we can talk about um, well, first of all, what is CapEx, but and uh how to survive the Capex boom?

SPEAKER_02 2:24

Yeah, of course. So look, I I've actually been working firsthand with AI for five, six years now. Um so starting in like 2019, 2020, I got really interested in the idea of, you know, do these models, which at the time they were just called large language models, um, do they have the capacity to take what's called unstructured data, right? That was in my bio. This is like text and audio and and and video, and to process it in a way that now the computer can understand what's going on. Right. And these models have obviously evolved into ChatGPT and what's now called AI. So what why is AI important? Well, look, this is a technology that you know allows us to bridge um the kind of traditional statistical machine learning tools that are primarily associated with like um Excel spreadsheets and structured data, right? Okay, can we take do some time series analysis and predict the next like uh line of this trend? What AI or or um large language models to be more precise allow us to do is to now make sense of the world of unstructured data, which involves you know uh write writing novels or video games or um you know uh movies like a Hollywood movie. So all this sort of information that is so core to the human experience is now inbounds for these models. And that is, of course, you know, obviously a game changer for how we live in society, how we do our jobs, how the economy works, and and really everything in between.

SPEAKER_01 3:49

Can you, since I know that you are a user of AI, can you talk about like a use case when it comes to investing where AI is changing even your investment process?

From Structured To Unstructured Data

SPEAKER_02 4:00

Yeah, so I started off as a traditional quant analyst. You mentioned GMO. Um and you know, they they started doing this work in the 1970s, so decades ago, and it was at the time extremely cutting edge, using computers to take um financial information from um 10Ks and 10 Q's, which are the annual statements and quarterly statements of companies, and then automating the process of finding cheap stocks. Um, you know, what I was really interested in doing is saying, can we go beyond um just that traditional, again, I'm gonna use the term structured data to instead incorporate these other data sources. So, for example, like we talked about trademarks or patents, um, employee job postings. All this information is, you know, obviously very relevant for companies. I'd love to learn for company X who they employ and what sorts of skills they have. I'd love to learn for company Y what kind of culture they have based on their Glassdoor reviews, the text of those things. I'd love to learn about company Z and what sorts of IP intellectual property they have based on their patent corpus. That's great. The problem is that to actually manually go through, for example, five million patents, that'd be a very arduous task and probably just not tractable for an individual analyst and definitely not using, not definitely not for a quant, using traditional tools. So where large language models come into play is as a way to kind of solve this problem. Or now I have these five million patents. What I can actually do is, and now this is going to be a uh kind of loosely, this is gonna be a loose description, effectively.

SPEAKER_01 5:24

Just an example, yeah.

SPEAKER_02 5:26

Just you're having Chat GPT loop through each one, giving a very specific instruction set. Hey, I want you to look for patents related to say self-driving cars or robotics. And it can go through there as if you're giving a junior analyst a task, you know, go through this patent and tell me, is this about AI or not, or about robotics or not, and then do that five million more times, right? And so it's it's a way to kind of automate and scale what is a you know, generally pretty basic task, like something you could give to an analyst or an intern, um, but do so in a rapid um, you know, really quickly um over a large scale. And um, as a quant, now we're able to take these um unstructured data sets, these patents and things, and create factors around, hey, is this company cheap relative, not just just to its trailing earnings, which would be a traditional metric, but cheap relative to the strength of its IP library, right? That's something that you couldn't do um before large language models or um their predecessors.

Real Investing Use Cases For LLMs

SPEAKER_01 6:23

So the I just think about how much we're swimming in data. Like there's new information every minute of the day. We're we're kind of inundated with information more than our parents' generation and like many, many times more than um before that. And this is a way to kind of filter and sort through all of that junk when you really can't filter through it to provide a better use for it, maybe.

SPEAKER_02 6:52

Yeah, no, I think that's a good description. Look, it's like it's like a two-way street. There's two ways that AI can, you know, do things. One is like as a filtering tool to take this, you know, ocean of information, all the noise, and you know, you go on like social media and there's like all these different posts, kind of like filter through that to find, extract the signal of things that are relevant to me and my investments. Um, and then the reverse, which is the generative use case, which is hey, you know, I have a kernel of an idea for like a new movie script. Why don't but I'm not really a professional in this space, why don't I just like give it to you know one of these AI models and have it kind of create this whole um, you know, script based on you know my description, then then I can iterate with it. Right. So there's more kind of expansion and and and then compression of information um available via these these models.

SPEAKER_01 7:36

So in essence, it's like a bigger toolkit. Um, but then with that toolkit has come serious dollars being invested um by some of the biggest companies that we're familiar with. And that is really what your research report talks about, which, like I mentioned, can be found in the show notes. Or if you go to um Sparkline Capital, I believe you um have um every few months um new research reports on hot topics. Like I know last um the beginning of the year in the spring last year, you had a tariff topic. Um so the title of your article published October 22nd last year um was surviving the AI CapEx Boom. And I think like as soon as you published that, everybody else was like, oh my gosh, AI um investment is really getting expensive, at least in terms of investment conversations. Um, but can you tell me like what some of the um well, what is the the thesis of the conversation? What are you, what problem are you trying to uncover or expose?

SPEAKER_02 8:44

Yeah, I think like if I could only say one thing, it would be that just because something's a good technology and it will be transformative doesn't mean it's a good investment. You can go back to the dot-com boom. If you had bought a basket of the most um exciting internet stocks in 2000, you would have been underwater for 20 years, despite the success of the internet, right? So let's go back to the current situation. Look, I I do believe, and I think you kind of heard some of the use cases, that AI will be transformative for the economy. Let's take that as a given. Now, does that necessarily mean that value will accrue to the folks actually building out the infrastructure? Maybe yes, maybe no. History would suggest no. So let me first define CapEx. I think that was a question from earlier. Yes, please. So um, in order to run these AI models, what's really quite interesting is it's not just a software problem. There's a huge hardware component. Um, there are these things called GPUs, um, um, which are kind of special chips that are designed um specifically for running these parallel computations, effectively linear algebra. And they're just they're essential for writing these models. And what's happened is that um these models are so much more compute intensive than the linear regressions and other traditional things I mentioned earlier. It's starting to create a bottleneck where if you're a company or an individual who wants to, if you're open AI, let's say, and you want to serve you know hundreds of millions of people, these AI models, you need to have a huge amount of servers in order to do that.

SPEAKER_01 10:05

And what's being what's happening is that you know, not just and each of those servers has the chips in them. Is that correct?

unknown 10:13

Yeah.

Filtering Versus Generating With AI

SPEAKER_02 10:13

So for the average data AI data center, it's about 60% of the value is these chips. So it's more than half the value of these things. Generally, NVIDIA, which I'm sure you've heard of. Um, that's the company that makes these chips. Um, there are a few others, but NVIDIA is like the dominant market share. And then the other 40% of the value is like the actual physical, you know, server racks, the cooling systems, the energy, the actual real estate, and the um, and there's like a memory component and net networking equipment, right? You may have heard of some of these Taiwanese companies, um like, you know, or Korean companies that make memory and storage that have all done really well these stocks as well. And that's because they're kind of um, you know, part of the part of the overall package um of that that's required to build up these data centers. Um but but the the point is that if you look, you know, if you look at how much money is being spent on this infrastructure build-out, right? Because we need this platform in order to now run these all all the anticipated demand um in terms of AI usage, the the numbers are kind of staggering. Um so just the big four hyperscalers, so that's Google, Amazon, Meta, and Microsoft are anticipated to spend like um, what is it,$400 billion a year for the next several years in doing this. If you look at like the broader estimates that include not just the big tech companies, but more generally, um, you know, folks are estimating five trillion dollars of uh capital expenditures or CapEx being spent over the next five years. So a trillion dollars a year overall in building out the data centers required to run AI. I mean, this is a massive amount of money. Um, you in in the paper, I compare it to past um similar build outs. So you can look at the dot-com build-out, you can look at the railroads in the uh 1860s when we connected um, you know, the the entire country of the US um via rail. And as a percentage of GDP, if you adjust for the depreciation and useful life with these assets, it's you know basically unprecedented levels. So we're as a society, largely led by Stan Altman and the big tech folks, and we don't, as ordinary people, always have a say in what what's happening, but you know, collectively we have decided, we being the country, that we are going to be investing, you know, an uh inordinate amount of money on building out this uh next what we what would is believed to be the next platform um for the kind of AI revolution.

Defining CapEx And The AI Stack

SPEAKER_01 12:28

Interesting. I want to break that down. First, let's start with the examples that you use. Um, railroads, I'm gonna use my um knowledge based on Gilded Age to say that you know, we had this new technology which can move people around the country and more importantly, economic goods. Um it it helped to foster the expansion of the company, uh company country. Um, but also somebody had to go and build the actual, you know, lines that the real um the rail system was built upon. Um, what were the outcomes for those companies that really kind of cornered the market, seemed like you know, the richest guy in the room. Um, what happened after that?

SPEAKER_02 13:15

So, like an anti-technological boom, right? So this had happened first in the UK and then later in the US, people saw the the potential for this, you know, game-changing technology to um, you know, create vast amounts of wealth. And as a result, a lot of people tried to do this. So it wasn't really like one guy cornered the market. There are hundreds of railroads that were um created specifically to build these out. And so a very fragmented market structure. But what happened is that um, you know, the the adoption of rail ended up taking longer than these guys expected, right? So it took decades actually for the usage of the rail to really ramp up to a point at which these were profitable. And in the interim, basically all these companies went bankrupt. Um, there were you know these different panics of two or three different waves when there are these big bankruptcies and consolidations and um and so on and so forth. But you know, in round numbers, the investors in the builders of the physical infrastructure underlying the rail all went bust. Um and instead, what's really interesting is there I have this chart here that shows that the actual contribution of the rail to US GDP was actually quite massive. But the thing is that that value did not accrue to the builders of the infrastructure. Instead, it went to their customers and the society. So if you're like a, I don't know, like a retailer trying to ship goods across the country, well, this is a way more efficiently to do that. And you can kind of build a business around um this technology if you're kind of open-minded. Um, those are the folks who did really well on the back of the rail, not ironically, somewhat, the uh guys who actually went out there to build um the infrastructure.

SPEAKER_01 14:50

And I'm aging myself in this part of the conversation, but I actually started my career in the first internet boom. I um got my first financial services job in the summer of '98. Um, and at the time we had this huge use case for the internet. Um, there was everything.com that was, you know, just what everybody wanted to invest in. Um, and in particular, the um telecoms, who had been a huge part of our economy um for the 20th century, were building um a huge amount of um, you know, ways to move information on the information highway. Is that appropriate to say? Um, so tell me about what happened um, you know, between the late 90s and into the 2000s.

The Scale Of AI Spending

SPEAKER_02 15:40

Yeah, we we basically saw a repeat of what happened a century earlier with the Rails. So going into the 90s, there's a lot of excitement around the internet. And folks said, hey, why don't we start investing a ton of money to build out the physical Falber Optic cables required to boot the information? And thus led to the telecom boom. Um and the these firms, I I want to say it was like 80 million miles, but it was a very large number of um of cables that were laid. Um, it was both kind of the more established players like AT ⁇ T as well as startups like Global Crossing and such who were founded specifically to go after this opportunity. Well, what happened is what you know, folks, what historians call the capital cycle, which is this idea that um, you know, like like the economy in general, it tends to operate in cycles with booms and busts, that there's a kind of initial phase where folks get really excited about the opportunity. Um, capital pours into this um this uh market, in this case, the build-out of telecoms. And then what happens is there's an overshoot where too much capital comes in, demand does indeed increase, but doesn't increase quite as quickly as um is forecast. And there's a point at which people realize, oh, wait, this is now a glut of supply. There's too much build-out, not enough usage. In the case of the.com boom, I think 85% of fob rocket cables were not used um in the bus. So in 2002, let's call it, um, after things went south, um, you know, 85% of cable was dark. And what that meant, of course, just based on you know econ 101, is that um, you know, the prices collapsed. And so bandwidth was basically 90% um, you know, on at a 90% discount from where it was earlier, which was obviously very bad if you're a telecom um company whose like lively relies on being able to charge your toll. Um, you're effectively a utility, right? And so many of these companies went bust. Um, those who didn't suffered. So there's an index calls called the Nasdaq Telecom Index. That thing went down over 95% in the dot-com bust and still hasn't recovered even today, which is kind of wild. Um, so yeah, maybe you wouldn't have gone gone bankrupt had you bought a basket of these guys, but it wasn't a great opportunity. Now, and then and then the the postscript is kind of what I mentioned with the railroads, which is there's a the flip side of being an infrastructure builder is being what I call an early adopter or a user or you know, someone who's actually not outlaying the capital to build, but the one who's actually um you know trying to take advantage of the new technology in order to you know drive gains.

SPEAKER_01 18:04

So let's say you're trying to increase productivity in your existing business.

Lessons From Railroads And Telco

SPEAKER_02 18:08

Exactly. Like, you know, what we're what I just discussed. You know, I could go out and hire a bunch of analysts to read patents, but you know, I'm just gonna use my AIs to do that instead. That's a uh creativity enhanced enhancing use of this technology. And there are many of them as well during the dot-com boom. And so those companies actually did really well. And like the best examples were Netflix and Google and Meta, right? Where they basically, their businesses are premised on the idea of streaming um uh content, you know, back and forth over the internet. And to the extent that's expensive, it makes it really hard to make that business work. But with bandwidth prices 90% discounted, actually it was a great time for those guys to launch. So, look, like everything's a flip side. There's almost a subsidy in effect here. The overbuild at what I call the infrastructure layer. So too much fiber led to a subsidy for the users of technology. So it's kind of like there's like two sides of each coin. Um, and again, like when you think about investing in any technology or any market, you got to think that maybe it's not a mon monolithic asset class. Like there's no such thing as an AI stock. People are like, oh, yeah, I want to buy AI stocks. What does that even mean? Right? There's like different parts of the value chain. And I think the easiest way, there's obviously you could segment the value chain a million ways. I think the easiest way to do so in the simplest, fewest um buckets is to say there's two halves. There's the um infrastructure builders, NVIDIA, you know, supplying the chips, uh, Microsoft putting it in the data centers, core we firms like that, and then the model companies like OpenAI and Google, who are actually building the investing you know, billions of dollars to train the models. And then there are the users or the early adopters of the technology that then take that model and use it to make their businesses work more efficiently and hopefully drive profits from that. And so I I think where the hype is, where the excess has been both on the capital side and on the valuation side is really mostly focused on the first part and and very little on the second part.

SPEAKER_01 19:56

Well, and I think that makes sense because when you just look for pure plays, it's it's attracted so much assets, especially since I guess the fourth quarter of 2022 when OpenAI kind of launched. Um and so I want to talk a little bit more about well, first just like, you know, um and that is a fact, like it's been a huge driver of performance when it comes to SP 500 returns until very recently, correct?

SPEAKER_02 20:28

Yes. Um and then I think 85% of the returns of the SP 500 have been driven since ChatGPT's release have been driven by these infrastructure stocks, and and more particularly, even just like the Mag 7 and a few other AI link stocks.

SPEAKER_01 20:42

And you know, that's where discussions about concentration of the biggest companies being a bigger and bigger portion of um indexes has come in and just the economy in general. Um but that you go further in um in your paper to also kind of explain this that um that what you really want to be looking at when if if your thesis is correct for um investments going forward is that change in terms of capital expenditure. Um and you spend some time researching um and explaining to us what um heavy or intensive CapEx companies, um, how their investment results are different from lighter capital expenditure companies. Can you talk a little bit about what you found there?

SPEAKER_02 21:35

Yeah, so basically capital intensity is a question of how much capital does a company need to like run its business. Um, on one extreme are utilities, which are extremely capital intensive. You're always kind of putting more money in to replace depreciating assets that are either either becoming obsolete due to technological technological change or just through wear and tear or like wearing down. It's a really hard business to be in, and um there are Very limited moats, right? Like the only moat you have is the ability to kind of spend more money, which isn't not truly a moat. Um, in contrast, there are what are called asset life businesses or non-capital intensive businesses, I guess, um, which are kind of the opposite. Um, the best example would be like Google, say 10 years ago when they had, you know, this, when they were focused on like say search. At that point, like they had raised very limit, very little relative money in terms of getting to IPO. So they basically didn't need to raise that much money to run this business where they just printed money, right? Their their return on invested capital was and is is but was insane on that business. Where they're very high margins.

SPEAKER_01 22:35

So they they make a lot of profit off of a lower amount of investment.

SPEAKER_02 22:40

Right. They don't they don't have to keep spending money to just run this machine that keeps cranking out profits, right? So that's an asset example of an asset-like business, right? They use network effects, they have you know strong intellectual property, a good brand, um, a lot of human capital. They built like this culture that for, you know, I think it was seven or eight years straight, they were the glass door's best place to work, right? So you think of Google in their heyday, that was kind of like the perfect business, like super asset-light, like, you know, they were just, you know, pure profit.

SPEAKER_01 23:07

Um that really, I mean, I like this is getting a little bit off topic, but didn't that allow them to go into many more businesses? Because with that profit, they invested into more ways to make money that were because they didn't have to just put it back into the ground to build up another power plant like a utility would.

Capital Cycles And Overbuild Risk

SPEAKER_02 23:28

Yeah, that's right. I mean, we saw today that you know, uh, Waymo, which is a uh a self-driving car company that they incubated, um, just raised some outside money at like a very significant valuation. Um, and and why does Waymo exist? Because Google was able to, through the cash flows generated by their main business, fund some side projects, some of which didn't work out, but but many of which did, um, such as Waymo. Um so it puts you in a great place to be to be asset light. Now, the problem is that Google to take Google as an example, they are becoming more and more asset heavy over time. So if you look at like the capital intensity of these businesses, so the main metric would be um how much do you spend on CapEx per year divided by your revenue? So what percentage of your revenue do you need to spend on um capital expenditures each year? Um historically, these companies ran about 4%, um so basically nothing. And now they're at 15. These are the Mag 7, so the you know, the largest seven tech companies, including Google and Microsoft. Um, you know, the extreme is Meta. So Meta is spending 35% of their sales on building out AI data centers, which is a you know crazy amount of money. For context, ATT at the height of the dot com boom, they were at, I want to say 18%. So, or maybe maybe even less in the teens, right?

SPEAKER_01 24:38

So these these are and another traditionally asset-light company where you know they have a website that's running that's getting data that they can use to sell ads to people to kind of you know um get into our living rooms, our our you know, hands. Um, so a huge shift um in terms of what got them here versus what they think will get them to the next phase.

SPEAKER_02 25:01

That's right. Yeah. Meta, Facebook, they're another perfect example of an asset-like business, right? Like a social network that was purely software driven, um, that had massive network effects and you know had you know really high profit margins and didn't require too much capital to run. Um, so you know, these are all examples of businesses that you know were kind of the perfect business, but are now transitioning because what's happening is that this AI opportunity is creating this like kind of competitive dynamic whereby these firms all view AI as an existential threat. They're all like, okay, you know, we used to kind of divide the kingdom up, right? It was used to be an oligopoly where like Meta had search and then Google had Google had search, meta had social, so on and so forth. But now what they're saying is, wait a second, whoever wins AI wins everything.

SPEAKER_01 25:45

And therefore, and we're gonna be disrupted, we're gonna be the blockbusters if we don't do whatever it takes to win this war.

SPEAKER_02 25:53

Right. So you have guys like you know, the CEO of Google or or Meta saying, look, I'd rather go bankrupt than lose this race. It's almost a good thing.

SPEAKER_01 26:00

That's an actual quote, right? Yeah, that's it.

SPEAKER_02 26:02

I mean, I think that was at Larry Page who said that, right? So like these people are actually saying, I need we're gonna go after this no matter what, right? Meta, they actually had a pretty decent quarter when they announced, um, you know, was it last week? But um, you know, people are definitely looking at that CapEx number being like, hmm, that's a little bit a little bit scary, right? Like, let's hope he doesn't do the same thing he did with the metaverse.

SPEAKER_01 26:22

Um well, and that's changing, you know, the equation because uh a lot of it is early stage funded out of cash flow. But um, and this can be an entire different episode topic. We're not gonna solve all the problems of this here, but um, borrowing is happening. Some of the borrowing is at a higher numbers so that it stays off balance sheet and private debt deals. Um, there's all sorts of implications that are complicated. Um, and I'm getting us off track, but like what did you find about these capital-intensive companies relative to their peers that are more capital-like?

SPEAKER_02 26:56

Yeah, so one of the nice things about being a quant is that I have you know decades of data and over thousands of stocks. And I can look at not just the big, you know, macro booms and busts that we discussed already, but even like kind of more micro cases. Like, for example, you know, how would a how would a materials company that was investing heavily in CapEx in a given random year have done relative to its peers who are investing less heavily?

SPEAKER_01 27:21

And it turns out that both And even in utilities, there might be more or less capital-intensive companies, even if they invested more than they index in general, right?

Winners: Users Versus Builders

SPEAKER_02 27:30

Yeah, and I found that, you know, whether we're talking about, you know, just in absolute, who's spending the most or sector neutral. So within financials, within technology, within utilities, who's spending the most, there's a pretty consistent pattern where those firms that invest heavily in capital expenditures tend to subsequently up underperform their peers. Um, and that's that's not to say that there's no like, you know, example counterexamples of firms who you know manage to pull it off. But just if you're a a if you're a betting person, right, you're just playing the odds. You say, wait, well, you know, this the base rate here is not exactly in your favor. Um, and that's both for firms that are investing heavily in aggressive balance sheet growth, aggressive capital growth, and for those that who are just more capital intensive in general, like um, you know, just have a higher capital to sales ratio. In other words, they're not necessarily investing for growth, they're just investing to tread water to maintain. They're just worse businesses. And, you know, the the concern, of course, is that um the biggest companies, the Magnificent 7, which I mentioned, um, as well as many other companies, are becoming more capital intensive as a result of AI. So AI boom is happening, whether we like it or not. It's now causing many firms, including and most notably big tech, to really push into this. And here's the problem, which is if it were just one small segment of the market doing this, fine, right? But the problem is that these companies are most of the index. Um, so the SP 500 index, one third of the index is the Mag 7 stocks. Um, and then if you include also like other infrastructure companies like you know, Broadcom, um you get to a number that is around 50% of the index. So if you're a normal person, you put your money in the SP, you're like, oh, whatever, it's a passive index, like I'm just gonna diversify by the market. Well, you're not really diversified because half your money is in basically a massive bet on this infrastructure build out being successful. And the success of the buildup, again, is not about the technology being successful. That's obviously one requirement, but it's not sufficient. What you also need is for the thing to be successful and for end user demand to be to pick up fast enough before the depreciation of these assets. So this build out is happening, we're spending all this money up front and building out the data centers. If the demand materializes quickly, then we're good. But if it's like the internet or the railroads, where it takes maybe a little bit longer than you might, they may have thought. Maybe it it still happens, it just happens in say five years, not three years. Well, then there's this huge gap in profitability. And that could mean potentially this overcapacity issue, the capital cycle turns, and maybe we'll end up in a situation like we did in historically, where you have where it's you know not so good for the builders of the infrastructure, which represent half the index, as I point out.

SPEAKER_01 30:10

Well, I I could talk about this all day, but I want to talk to about some of the investment implications. Um, I have some thoughts, and I know you do as well. Um, one of the, let's just start with the way you've chosen for kind of your core investment strategies that you offer for investors, um, how you've approached um uh, you know, kind of building a portfolio and what you're focusing on.

Mapping The AI Value Chain

SPEAKER_02 30:37

Yeah. So the key insight that we have at Sparkline is that value is not just the tangible book value of a company, right? So so if you go back to like the history of investing, like Warren Buffett, Ben Grimm, all this stuff, um, you know, there's this idea that book value is kind of the proxy for fundamental value. But what's happened is that over the past century, basically, the economy has transformed massively and we've become very asset-light. The best businesses are like the ones I just described. Um, and so it's intangible assets, um, which I define as intellectual property, um, human capital, brand equity, and then network effects. Those are kind of the four pillars that are actually driving value today. That drives, let's say, half of the value of companies in the market in general, and then in the most profitable sub-segments of the market, like technology and the large gaps, 80% or more, right? So most of the value of companies is no longer tangible, but intangible. And I think that creates a huge opportunity because most traditional investors, especially quants, are still very fixated on kind of these old school metrics. And for them, they're never gonna buy a company like NVIDIA or or whatever, because they're always gonna seem expensive if you don't give them credit for their intangible assets. Right. And conversely, your so-called growth investors are gonna always own these things because they think that there's always opportunity, but they have no idea when to sell out. They have no self-discipline, which means that when the bubble comes and it's our maybe we are in one, um, it's it's gonna be a little bit hard for them to sidestep um the potential um uh losses from from um you know compression um of uh multiples and such. So we're kind of trying to basically take this old school idea of value investing, buying stocks that are cheap relative to intrinsic value, but then adapting the intrinsic value to a concept to include a more holistic definition of you know intangible plus tangible assets that allows us to take a framework into technology stocks and things like that, that most traditional value investors tend to askew.

SPEAKER_01 32:30

And I'll just this is a full circle episode moment because in the beginning you talked about some of the ways that you're doing that type of research, which is you're using some of the tools that wouldn't have been available before the large language models and you know big data sets existed, um, and applying your quantitative trading skills or investment skills, research skills to that. So I appreciate you know, kind of that coming full circle. Another, you know, kind of thing, just as a wealth manager, I think about is um with the Magnificent Seven, which is um necessarily, you know, intertwined with his conversation about um a pivot to higher, more capital-intensive activities. Um, there are millionaires, decamillionaires, centimillionaires who can um attribute much, if not almost all, of their, you know, kind of success and net worth to being um, you know, kind of riding those horses of the Magnificent Seven into the sunset of so far, um, where you know, there's been huge, um, huge wealth building um from the returns of these companies. And it's so difficult when you know, you know, the recent history to think about diversifying away from them. Um, but this is, you know, kind of a cautionary note of taking careful consideration for how you diversify. Um, there's much less AI exposure around the world than there is in the US if you look at markets like the way um, you know, you can articulate investments. And there's a lot of cautionary tales. I'll use the um tech wreck of the early 2000s and the internet boom, the first internet boom is an example where um, you know, kind of what got you here doesn't always um keep you safe going forward. So I think that there's definitely some considerations for those of you who have, you know, significant exposure um to single company or the cohort of the Max 7. I don't know if you have any thoughts on that.

Concentration Risk In Indexes

SPEAKER_02 34:37

Yeah, look, what what I would add is is yes, I I agree with that. I mean, our our funds when we launched initially were actually overweight, a lot of these infrastructure stocks. So we're not like a traditional old school value investor. We actually said, look, NVIDIA at a 100 P ratio is actually cheap because of the prodigious value of its uh IP and its network effects and and human capital and such. Right. So, you know, we're we were very much on that trade. But what's happened is that over time we started to just naturally rotate out based on the valuation framework I outlined earlier as we start started to see these companies appreciate and value what we believe to be too much relative to the prospects. Um and and what I would add to your point is look, I completely understand what you're saying, which is it's very psychologically difficult for someone to cut cut your winners, right? That seems like the wrong thing to do. But on the other hand, I mean, these kind of blessings, right? Like, you know, as an index investor in the US or as any investor in the US, you we've likely benefited from, you know, probably above expect above expected gains, right? Much more than historically um would be expected over the past decade due to the rise of AI and technology in general. Maybe take some chips off the table. What I would say is this to your other point, which is you know, one of the challenges um with navigating the current environment is it's oftentimes framed as this kind of like dilemma, a dichotomy. Either we invest in AI, so US large cap stocks, and we get AI exposure, but with um kind of concentration around infrastructure capital risks and around high valuations, or we go abroad, we go to value stocks, we go to small caps, we do equal weighted SP, and these things give you diversification away from the Mag 7, but we lose AI. We are now invested in a bunch of kind of old economy companies, laggards in the AI race to use the technical term, who are likely going to be disrupted as you know the the technology pushes through. What I would argue is that there's a third option, which is what I call like the AI early adopters. I alluded to them earlier. These are folks, like I mentioned, you know, some around the dot-com boom and some around the the uh railroads, folks that like in the kind of they are not um they're not the picks and shovels. They don't sell the infrastructure, they're not NVIDIA. But you know, in the those companies, only only are about 10% of companies. The other 90% of companies are what would you consider users, and within that, there's a huge amount of dispersion. There's some companies that are really aggressively investing in AI. They're really trying to capture the opportunity they see around the corner, they know this is gonna be a big thing, and they're doing everything they can to position to benefit. And then there's you know, some guys in the middle, and then the the rest of the folks who are kind of just head in the sand, not even they don't even care, just want to kind of milk their current position and and hope it just goes away. Right. And so what I would say is that there's this middle category, the early adopters, or actually, I think the best positioned, because you know, they stand to benefit from the use of tech technology, and for them it's only a relative game. Can they do a better job using AI than their competitors, who, as I mentioned, many of whom are just not even looking at this at all?

SPEAKER_01 37:29

Um so, like a couple anecdotal examples. Um, tell either add in or elaborate. I know I often see Caterpillar mentioned there, yeah, Walmart. Um, what are some like these are not like brand new companies? They're just using the new technologies to adapt, right?

Capital Intensity And Underperformance

SPEAKER_02 37:48

That's right. And you see these examples across industry. So it's industrial, as you mentioned, um, you know, JP Morgan, banking, like you see this kind of all over the place in in healthcare, pharmaceutical, drug development. There are all these little use cases for AI, whether it's something as basic as like um you know making call centers more efficient, to like in the case of Walmart, like logistical and kind of warehouse inventory management of like a wide long amount of SKUs. Like you see there's a lot of different interesting use cases. Um, actually, the paper that I'm writing up now has been.

SPEAKER_00 38:17

Oh, that was gonna be one of my questions. Okay.

SPEAKER_02 38:19

Yeah, where we just kind of scan uh public company earnings calls. So this is all public information and look at what it is that people are that companies are saying, what percentage are reporting AI usage, what percentage of those are reporting positive ROI from their usage. So not just, hey, we're trying this new pilot, but actually, oh yeah, we actually managed to cut costs by 20% or you know, increase our ad conversion rate by five. Um, so these things are popping up all over the place now and at a kind of accelerating rate. So so from a low base, admittedly, it's not like a ton. I think it's like 15% of companies that are reporting or reporting some positive um benefits, economic benefits from AI. But that number is increasing pretty rapidly. And you look at how the technology is evolving, just being someone who's kind of in the weeds, you know, on all the on all the Twitter and and Reddit forums or whatever, and it's kind of seeing how fast this stuff is moving, and you can kind of easily extrapolate out that S curve and how it goes. But the point is that these companies um, you know, exist in all industries. It's not just technology, it's not just um, you know, um the kind of common spaces, it's not just Palantir, that these companies they may be in kind of sleepy industries, but that's almost good because one, it means that competition is weaker. And two, their advantage has never been their technological prowess. Their advantage is their installed base, the fact that they have you know all these customers and all this proprietary data they've been collecting over the years about these customers and you know, have um the ability to kind of then leverage any technology, you know, over this larger installed base, talking about a 2% improvement on a huge amount of assets where profit margin is 5%, right? Like that's a huge addition to their um to their economic value. Um, and I think these companies are really interesting because if you look at the valuations of them relative to what I call the laggards. So what I did was I said, let's look at all the companies that are not infrastructure, and we'll divide it just for simplicity into like the adopters, early adopters, and then to the laggards, these two different groups. Well, it turns out that the valuations of these two groups are basically the same: price to earnings, price to book, price to sales. And what that means is the market's not distinguishing between companies that make AI investments, which is not, which is kind of the opposite of what's in infrastructure. In infrastructure, companies that make investments, when Oracle announces that they're doing a deal with um open AI, their stock goes up, right? That doesn't happen in in this other space. But people are investors are kind of, I think, sleeping on the opportunity in the caterpillars of the world, um, companies that you know are actually positioned well for AI and have other advantages that they can kind of you know synergize, I guess, um, with AI, yet aren't trading at the kind of crazy valuations that we see in the infrastructure space.

SPEAKER_01 40:48

Well, that's interesting. It may um foretell part of the reason why there continues to be such strong and positive forward earnings expectations um from some of the best companies out there, which aggregate into the SP. Um, and I think that like leads into my third kind of investment lesson, which is that everything's not binary, everything's not black and white, it would be unrealistic and um I would um likely ill-advised to scrub your exposure, for example, to the seven largest companies and the S P 500. Um, but thinking about um, you know, kind of diversifying, hedging your bets, um, avoiding concentrated stock issues, as well as recognizing that no one has a crystal ball. Um, things will continue to evolve. Um, well, Kai, you may have had special insight into kind of um the AI story unfolding. It wasn't, you know, something that everybody knew, but you know, around the corner is some other unexpected um, you know, curveball, but also um innovation and opportunity. Um, and so paying attention to your investment discipline so that you're, you know, kind of facing forward and not resting on your laurels, I think just like those companies that are innovating and and finding their productivity by staying with the times is important for your investment portfolio.

Asset-Light To Asset-Heavy Shift

SPEAKER_02 42:13

Exactly. Right, exactly. I think, you know, the if your goal is to, you know, kind of maintain your wealth, you need to be diversified. Um, and again, it cut it cuts both ways, but you know, for most investors, and you know, this is I'm generalizing, obviously, is we've kind of run in our winners, and our winners have been US large cap stocks, in particular the the largest ones. Um, you know, and and the trades become so concentrated, so crowded in in a single theme, which is will this AI boom, um, capital boom continue? That like it is in a way is kind of crowding out all the other ways of diversifying. And so I yeah, I would say generally, yeah, do think a little bit about kind of thematically what else there is out there, um, whether it's the value stocks, the small caps, the international stocks, um, you know, different ways to get exposure that isn't just um you know all leverage to a single trade.

SPEAKER_01 43:02

Well, I think I've alluded to it, um, Kai, but where can people find more? Read your papers and read the other papers that um you've written.

SPEAKER_02 43:10

Yeah, so I you can just go to my website, sparklinecapital.com. That's s p-ar-r-k-l-in capital.com. And so yeah, I did my last piece on AI CapEx. I did one in Warren Buffett, um, you know, because he's retiring, obviously, and kind of thinking about his the evolution of his own journey as an investor from being kind of more old school, buying, you know, textile mills, um, all the way up to kind of buying Coca-Cola and then in the Naple, of course. So the extent to which his investment process also mirrors kind of the the journey that I've been on, trying to add intangible moats um to the to the framework. And then the the trade wars, and I did one on kind of uh partisan investing. The upshot is don't invest based on partisan politics.

SPEAKER_01 43:51

Um don't yeah, don't invest on who's in charge. Yep.

SPEAKER_02 43:55

Yeah, it turns out that that's not a good idea. Um, but there are other ways you can you can actually take advantage of trading. Around elections, namely around um well, this is gonna sound bad, but like regulatory capture. So companies that do a lot of lobbying and do a lot of um, you know, political donations, they actually do better, right? So they're building an intangible asset that's their political influence, and you hate to see it, but that asset has value. And in fact, like the the craziest stat I I found is that the ROI, so the return on investment in a dollar of lobbying, is like an order of magnitude higher, so 10 times higher, than a dollar of investment in like a RD. Right? So you hate to see it, but you know, it is what it is.

SPEAKER_01 44:33

But you'd rather know than just um have that information floating and without your ability to use it, I guess.

SPEAKER_02 44:39

Yeah, I mean, as a as an investor, these things all matter, right? We're just trying to find, you know, edges and trying to understand the value of companies. And some companies, that's kind of the main point of the company, is that they have um, you know, a special regulatory monopoly. And it is what it is. That's just, you know, you can't, you know, you know, what are you gonna deny the fact? If it's a fact, it's a fact.

SPEAKER_01 45:00

Well, thank you for bringing these important and just thought-provoking topics um to the table. Kai, I appreciate you joining us today.

SPEAKER_02 45:08

Thanks so much for having me on.

SPEAKER_00 45:21

Thank you for listening to the Women's Money Wisdom Podcast. If you found value in this episode, the best way that you can support the podcast is to forward an episode to a friend or leave a review. Go to ProPlan.com and the podcast link to get all the resources and links mentioned. This presentation by Pro Planning is intended for general information purposes only. No portion of this presentation serves as a receipt of or a substitute for personal investment advice from ProPlanning or any other investment professional of your choosing. Copies of Pro Planning's current rent and disclosure brochure and form CRS discussing our advisory services and fees are available upon request or on our website platform at proplan.com. The information that we share is meant to educate and inspire, not serve as personalized financial advice. Everyone's situation is unique, so be sure to consult with your own financial professional for guidance that fits your life. And just so you know, the opinions shared in this podcast are Melissa's own and those of our guests. They don't necessarily represent any organizations with which Melissa is affiliated. For more important disclosures, please go to our webpage at proplan.com.