The AI Trade Is Shifting: Adoption, Infrastructure Risk, and Hidden Concentration in the S&P 500

In this episode of Lead-Lag Live, I sit down with Kai Wu, Founder and CIO of Sparkline Capital, to break down why the AI boom may be entering a new phase and why investors concentrated in infrastructure stocks could be taking more risk than they realize.

With nearly half of the S&P 500 tied directly or indirectly to AI infrastructure buildout, Kai explains the dangers of crowded positioning, excessive capital spending, and valuation expansion. Drawing on historical technology cycles from railroads to the internet, he outlines why the biggest long-term winners often aren’t the builders, but the early adopters who use new technology to gain efficiency and market share.

We also dive into enterprise adoption rates, proof-of-ROI versus hype-driven AI mentions on earnings calls, and how to distinguish companies generating measurable productivity gains from those simply telling the story.

In this episode:
– Why the AI cycle may be shifting from buildout to adoption
– How only about 10 percent of firms are currently using AI in production
– The valuation risk embedded in infrastructure-heavy portfolios
– Why early adopters may outperform the AI builders
– How advisors can rethink AI exposure without abandoning the theme

Lead-Lag Live brings you inside conversations with the financial thinkers who shape markets. Subscribe for interviews that go deeper than the noise.

#AI #StockMarket #SP500 #NVDA #TechStocks #Tesla

Transcript

Telecom Bust As Cautionary Tale

SPEAKER_00 0:00

These companies spent, you know, billions of dollars building out fiber optic cable. And it turned out that um, you know, when the dot-com bust happened, not that much of these, this, this capacity was being used. So 85% actually, to put a number on it, of uh fiber optic cables were dark. They were not being utilized. And that, just based on basic economics, sent prices crashing. So the price of um internet bandwidth fell 90%. Um, and that's obviously very bad for the telecoms menu, which went bankrupt, like global crossing being a big example. Okay, so that's a risk number one, which is you you're on the hook. You're you're opening your your checkbook to write to write the check to build out this technology. If it works, that's that's that's good. If it doesn't, you're the one who's on the hook.

Giveaway And Show Open

SPEAKER_01 0:49

So it's my birthday, and this year I'm celebrating by giving a special gift away to one of you. And to officially wish a happy my birthday to you, I've had this tote with a few surprises from our signature P2 merch portfolio. And it's not just random merch. Inside, you'll find some items that really at least signal you know exactly what you're doing in these markets. And the tote, well, it won't hedge your portfolio, but it will make you look smarter than anyone pretending they know what risk on means at the grocery store. If you want it, here's what you need to do. Follow Lead Leg Report on X, follow Mela underscore Schaefer on X, subscribe to Lead Leg Media on YouTube, and like and share this video. Only one person gets the toe, but since everyone gets our content, it's really a happy my birthday to all of you. I'm your host, Melanie Schaefer. Welcome to Lead Leg Live.

From Build-Out To Adoption

SPEAKER_01 1:44

Today we're talking about the AI trade shifting from build-out to adoption. We've heard the bubble question getting louder, including a recent warning from Microsoft CEO Satya Nadella, that for this not to become a bubble, the benefits have to spread beyond just the biggest players writing the CapEx checks. My guest today is Kai Wu, founder and CIO at Sparkline Capital. Kai, it's always great to have you here.

SPEAKER_00 2:09

Thanks for having me back, Melanie.

SPEAKER_01 2:12

So it was mentioned recently in the Financial Times, actually, that Nadella's view is basically if AI gains don't spread uh into the real economy, the boom risks turning into a bubble. Uh when you say we're entering the adoption phase, what exactly do you mean by that and what's coming next?

SPEAKER_00 2:32

Yeah, so this uh nomenclature comes from um the work of Everett Rogers, who um kind of coined this diffusion model of technology. Um and you can kind of visualize it as an S-curve where technological adoption starts off with a small group of innovators and then eventually spreads more mainstream before asymptoting and kind of plateauing once the uh the market's saturated. And we've seen this across many historical episodes, ranging from electricity to the internet to cell phones, um these kind of S S-curve-shaped adoption cycles. Um, and so you think about AI where we are today, and you know, really a lot of the energy, um, you know, both at the corporate level, so the CEO of Microsoft, of course, as as well as amongst investor focused, has been on the build-out, um, what I call the infrastructure phase of the cycle.

S-Curves And Infrastructure Phase

SPEAKER_00 3:20

Um, and that's all been about, hey, can we actually marshal the resources necessary to put together this um the physical infrastructure to run GPUs and run AI models at scale? Right. And that's been named mainly the work of uh NVIDIA um and the hyperscalers, um, as well as kind of utilities and more kind of hardware companies uh building out enough uh capacity. Now that that narrative really um kind of took off maybe last year and accelerated um into um into this year, right, with 2026 CapEx expected to be almost double what it was last year. So we have you know trillions of dollars now being spent um over the next few years in kind of plan construction around AI data centers. That's kind of the first phase of the boom that we've seen again repeat in historical cycles too. But if you go back to history, what else, what happens next tends to be a little bit different, which is at some point, people say, wait a second, we have the build out secured. Now the question is, will they actually buy it? Right? Will there be enough end user demand um such that the uh initial outlays we're making now will be justified and you know returns of investment are are adequate amongst the end users will then justify spending more. And then ultimately that will flow as revenue to the infrastructure builders. And what's quite interesting is you can track the returns of stock, of stock prices um through the cycle, and it tends to mirror where you are in the cycle. So, for example, in the very beginning, um it's the infrastructure builders who do the best, of course, um, as CapEx um plans, you know, are kind of revised higher and higher and higher. Again, it's just the same in the dot-com boom and you know, has been the case so

Tipping Point To Adoption Winners

SPEAKER_00 5:00

far today with the infrastructure infrastructure stocks and AI doing the best. But you know, if you look historically, there's also been a tipping point. When you shift from phase one to phase two, the adoption phase, what actually happens is it's the adopters of the technology that end up doing doing better than the builders. Um, and you know, if you look at the long-term history, that that this tends to be kind of the way things play out, where the long-term winners of you know infrastructure build-outs and technology booms are not actually the builders for a variety of reasons we can get into, um, but actually the adopters of the technology, able to use it um as a way of gaining market share over their competitors who may be a little bit less tech savvy.

Why Enterprise AI Adoption Is Slow

SPEAKER_01 5:38

Yeah, and so uh to tap into that a little bit more in your recent investor letter, which I I urge our viewers to sign up for, you say that the the in the data that's only about 10% of businesses are using AI in production right now. Why do you think that adoption is still so low? And what do you think changes first as that number starts to rise?

SPEAKER_00 5:57

Yeah, look, I mean, let's talk about two things. So, first is consumer adoption, right? People have been very, very impressed at the rate at which OpenAI's Chat GPT application has been able to gain users, this exponential increase. One of the fastest apps to reach, you know, whatever, million um users. You know, one interesting side point though is that only 5% of those users approximately actually are paid subscribers, right? So 95% of folks, you know, don't necessarily see enough value in actually paying, right? So, and then you go to the enterprise side. So, what are businesses actually doing to your question? You know, enterprise adoption is always going to be slower than um consumer. And it's not even about like um AI specifically. Enterprise adoption of any technology or any new system is just slow because you know, for big companies, it takes a long time to move the battleship. Um, and you know, there's a lot of um, you know, money and jobs at stake in in these cases. So they don't want to kind of rush into things. So even though ChatGBT was was released three and a half over three years ago, we're we're we're still seeing um you know firms take a c a generally cautious stance toward towards adoption. Um so 10%, you know, and it kind of mirrors where I think we are in the cycle. So if you go back to that um Everett Rogers model, he defined the kind of early phases being like the top, the tip of the spear is like two and a half percent. And then early

Measuring Real ROI In Earnings Calls

SPEAKER_00 7:12

adopters are around, you know, 12 to 15%. Right. So the 10% number kind of makes sense as, you know, uh as you know, roughly mapping to um where we would be on the cycle if it were the case that we're now reaching this adoption phase um, you know, post-build-out.

SPEAKER_01 7:27

Yes, talking about uh the cycle, you make a point in the in the letter about needing proof of ROI, which you just spoke a little bit about, and not just AI activity. When you look across earnings calls, what separates the companies uh with real AI-driven gains from the ones that are just sort of telling, telling the story?

SPEAKER_00 7:45

Yes, and this is something I've been you know paying a lot of attention to over the years. Um, you know, these these AI adoption metrics are, you know, have been kind of something I've been tracking, you know, real times, you know, for last even before ChatGPT was launched. Um and you know, one thing that's happened is that as AI has become very overhyped, um, you've seen a tendency for CEOs to want to just talk about it because, you know, it'll boost their stock price and it's kind of what you're supposed to do. So the the the big the trick became how can we go about, you know, this is obviously useful data, right? There's thousands of companies that are talking about um their businesses on these urgence calls. So we want to be able to use it. We also don't want to fall trap to um being kind of played by corporate spin. How do we go about um disentangling what is um you know just talking about AI versus actually actually doing stuff? Um and so what we decided to do was to look at um mentions that included tangible numerical um uh uh numerical ROI or um efficiency or cost gains. And so what I mean by that is if if a company says, if a CEO says, oh yeah, we're using AI to transform our business, that's very vague, it doesn't mean anything. Every of course everyone's gonna say that. But if they say, actually, you know, we used this new AI product called X, and it helped um increase the asset utilization of our um of this thing

Concrete Use Cases Across Industries

SPEAKER_00 9:10

we have by you know 12.15% or you know, some kind of numerical amount, it's more likely that they're actually citing tangible gains. So that was kind of the way we went about doing this. And of course, we use a large language model itself to go through all the Ern Call transcripts and and classify them as being in into one of three buckets. The the broadest bucket is just any mention of a company using AI, um, even qualitatively. The middle bucket is mentions of companies using AI, um, where um they can link it to a specific reduction in cost, expansion of um you know revenue or some sort of tangible metric with a number attached to it. And then the final bit was um actual ROI, return on investment, which means that not only can you show the gains, but you can show the gains relative to the costs. Because again, if even if you have a you know tremendously useful AI product, but the product itself costs a billion dollars to buy, that's not that that useful. Um, so those are kind of the three categories that we use to parse all these arguments calls.

SPEAKER_01 10:09

Yeah. So when you're uh collecting all that data, and this you know, question is probably for investors who are listening from the outside, what are some of the cleanest examples of AI showing up in fundamentals in terms of revenue lift or productivity cost savings, as you mentioned?

SPEAKER_00 10:23

Or yeah, I mean, there there are hundreds of interesting examples, and I I did ex include one exhibit which we should put up where I highlighted a few cases. So, like for example, Target talked about how they were using um, you know, kind of AI within the within the warehouses to help improve the efficiency of picking um, you know, uh goods. There were examples around um, you know, managing inventory around lots of different SKUs, a biotech use case around RD, you know, uh employee um management systems. So I think there was a uh a storage company that was using a system to help um you know manage their workforce. So their kind of label labor costs were being reduced and um retention was it was increased. Um examples of the kind of the classical example of, of course, is Meta using it to improve their ad conversion rates, so for digital marketing. Um but yeah, there's a ton of different um you know applications. And and I think the key to think the key thing to you know focus on is the fact that it's not just meta, right? It's not just 10 companies. It actually is starting to seep, as Nadella you know hopes, with this the fusion concept into the real economy, into businesses outside of just technology.

SPEAKER_01 11:32

Yeah, so I want to put it a little bit, just talk about your framework, which tries to avoid being overexposed to the most crowded.

Risks Of Concentrating In Infrastructure

SPEAKER_01 11:38

I as I talked about the first question, KPEX heavy, part of the AI trade. In in plain English, what's the risk investors take on uh Kai when their exposure is basically concentrated in the infrastructure layer?

SPEAKER_00 11:50

So I think there are two main risks. The first is um the fact that these companies are spending tens, if not hundreds, of billions of dollars on building out these uh big data centers. And this is money that they expect to earn back over the next, say, five plus years. Um but the problem, of course, is that's a speculative um bet. If you go back to the history of um these build-outs, let's talk about the dot-com boom. In the dot com boom, telecoms were the you know utilities that put up built built out the the backbone of the internet. These companies spent you know billions of dollars building out fiber optic cable. And it turned out that um, you know, when the dot-com bust happened, not that much of these, this, this capacity was being used. So 85%, actually, to put a number on it, of uh fiber optic cables were dark. They were not being utilized. And that, just based on basic economics, sent prices crashing. So the price of um internet bandwidth fell 90%. Um and that's obviously very bad for the telecoms menu, which went bankrupt, like global crossing being a big example. Right. So that's a risk number one, which is you're on the hook. You're you're opening your checkbook to write to write the check to build out this technology. If it works, that's that's that's good. If it doesn't, you're the one who's on the hook. Um, the second risk is just a valuation risk. Um, one thing I looked at was the extent to which there's a group of companies, which I call AI infrastructure stocks, which are the mainly the Magnificent Sevens with NVIDIA and the hyperscalers, as well as you know, a few other folks like you know, Oracle makes it in and Core Weave companies like that. These companies, um their valuations went from, say 10 years ago, a small premium to the market to now almost 2x the market. Right. And what I'm saying is, you know, oftentimes we discovered this in the dot-com boom as well, which is the internet became a thing. The technology was successful. The problem is that investors, you know, who came in at the wrong time made no money. In fact, they lost money because valuations matter. Because if you buy any investment at too high a price, it's not a good investment. And that that's kind of what's happening here, potentially, with these stocks running up so highly that anyone who wants AI exposure knows where to go. It's just too obvious a trade. And so, because of these elevated multiples, they're really priced for an outcome that if it works out, it is good. But you know, there's a lot of downside and a lot of risk in the event that um, you know, you that there's an air pocket or there's a you know mismatch in and when the revenue ultimately comes, if it even does, relative to

Market Mispricing And Early Adopters

SPEAKER_00 14:12

um the amount of money they're spending right now and trying to build um for this um you know future potential demand. Yeah.

SPEAKER_01 14:19

So where specifically then is the market mispricing uh AI adoption?

SPEAKER_00 14:24

Yeah, so I I think the I think the opportunity is actually not in the infrastructure firms. Um, and we've been rotating out of those companies over time into a group that I call early adopters. And so when I say early adopters, what I mean is so non-infrastructure companies, you can kind of take that, you know, 80 to 90 percent of the universe and segment it into firms that are investing in AI and those who are not, which I call the laggards. Um, and you know, my my thesis here is that um, you know, as AI becomes more and more powerful, um, and if it indeed will be transformative, which I think many people, including myself, expect it to be, and including Satya Nadella, who expects it to um, you know, he says, bend the productivity curve to really increase margins for the companies that are using it. Then the opportunity is more in looking at this massive dispersion across names, where you're you find in any given industry a handful of early adopters that are really aggressively pursuing the opportunity. And then the majority of companies are laggards. They're just doing nothing. They're kind of playing wait and see and or or whatever. And we're seeing this with stock prices the past week or or even the past year with software stocks going down. You know, a couple of days ago it was uh wealth management stocks going down at, you know, this kind of disruption fears that are roiling the markets. You know, the the the market's finally catching up to this idea that um AI is actually here to stay and it will change the world. And it that people are, I don't think they're doing it correctly, but they are recognizing that there will be separation between winners and losers due to AI. Now, the really interesting thing is, you know, to my point about indiscriminate selling, that if you look at the valuations of the laggard basket relative to the early adopter basket, they're basically the same. So say as I mentioned, go back 10 years, infrastructure stocks valuations have increased a lot. The other two groups, early adopters and laggards, have basically stayed steady relative to the market. And there's really very little difference between the two, implying that the market is basically saying, I think either everything will be disrupted or nothing will be disrupted. But is not, you know, being being thoughtful about how there's some dispersion that maybe some companies will be winners and others losers, and not really rewarding companies for the investments they're making in AI, even though we expect those to translate into tangible gains as we move forward.

SPEAKER_01 16:37

Yeah. So then for an advisor who's thinking about uh how to fit on things into their portfolio, or if someone wants AI exposure but doesn't want to be all in on the most expensive, most crowded names, as you've mentioned, what does a reasonable allocation uh mindset look like?

Portfolio Construction And Diversification

SPEAKER_00 16:52

Yeah, so so first of all, I think just to kind of level set, I think the the challenge that we face today, advisors face today, and any uh anyone, even retail investors, is that the standard indices, let's take the SP 500, is just they're just massively overloaded with AI infrastructure stocks. So for example, the SP 500 is 33% Magnificent 7, which we all know. But then if you add in the oracles of the world, the rest of the infrastructure stocks, you're talking about only 50% of the index. So you think you're making a passive um market bet on the market, but really you're just taking a huge, um, you know, a huge exposure, concentrated exposure to this, the sustainability of this uh boom and the trillions of dollars being spent um, you know, in building out these data centers. And if it works out, that's fine again. But if it doesn't, you know, you actually have a lot on the line. And I think people don't, I think, correctly appreciate the extent to which they are making that implicit bet. Um, so what are the solutions? Um, I think there's two things people have generally done. One is they said, you know what, like I'm so scared of this AI bubble that I'm just gonna completely um go the other way, buy you know value stocks, buy international stocks, buy you know small caps that have less exposure to um to AI. And you, you know, you could you couldn't do that. The problem is that then you're basically taking the opposite risk, which is that what if AI does continue to advance and the technology does improve, well, then you're missing out on the entire thing, right? That's like someone who who who owns no internet stocks in the dot-com boom. Okay, fine, you missed the bust, but you can also miss out on this revolutionary technology. Um, and so you just get chock full of laggards if you do that. Um I think the the solution is, you know, kind of framing this not so much as a dilemma, but there's kind of this middle ground, this this third option is how I've talked about it, which is these early adopter stocks that, you know, offered a lot of the upside of AI if it works out. These are the kind of second-order beneficiaries of AI technology, right? The examples would be, you know, uh retailers in the railroad boom of the 1860s that were able to use the technology to ship their goods more efficiently, or Netflix, Meta, Google in the internet era, they were the big long-term winners because they were able to use the um the internet, the subsidized internet because of the price of bandwidth falling so much to build these huge businesses, right? So the early adopters, you know, have historically been the kind of winners in the endgame. Um, and you know, where the infrastructure builders end up being utilities. I'm not saying it will necessarily play out that way, um, but you know, you should keep that in mind when, you know, you should keep that in mind when you think about where should where should I be putting my money when I'm betting on AI, and then pair that with two more pieces of information. One is that the early adopters are um a lot less overvalued, as I mentioned. Um, two, they're a lot less capital intensive. In fact, they're actually more capital light now than they were 10 years ago because they're offloading a lot of the uh infrastructure building costs to the uh cloud providers. And I guess third, you know, this third point would to, you know, in their favor is just diversification, right? Like the infrastructure firms are like 85% technology companies, whereas the early adopters, they have some tech bias. Obviously, tech companies tend to be more um, you know, uh progressive when it comes to adopting technology. But the majority of the early adopter basket is actually industrials, healthcare, financials, kind of these old economy companies. Uh, and it just happens to be the leaders in each category in each respective sector that, you know, from a relative standpoint um are believed to um have a potential advantage due to their you know forward-thinking investments in AI. So I think that the early adopters you know are you know really an underappreciated um but um you know potentially clever way to play the AI boom, where you can get a lot of the upside, um, where you can while also mitigating some of the risks of a potential bubble.

SPEAKER_01 20:39

Yes, okay. And I I mean a lot of the information that we've discussed here is

Where To Find Kai’s Research

SPEAKER_01 20:44

in uh your investor lever letter. Where else can people go uh to find out more about you? Where can they sign up for the investor letter and where can they go uh to to be in contact with you?

SPEAKER_00 20:53

Yeah, so the investor letter is is obviously um about the funds themselves. So it lives on our fund website. So it's just etf.sparklinecapital.com. Um and you just go to the website, you'll see this year's letter, last year's letter as well. Um and if you want to just find me more in general, like I have a website where I just post kind of more more of the more academic research I do called sparklinecapital.com. Um and I'm also available on on Twitter and LinkedIn at CKI Woo.

SPEAKER_01 21:21

Awesome. Well, Kai, thank you again so much for joining me, and thanks to everybody for watching. Be sure to like, share, and subscribe for more episodes of Lead Light Live.