Lead-Lag Live

Boom or Bust: Kai Wu on AI CapEx, Concentration Risk, and the Next Phase of Tech

• Michael A. Gayed, CFA

Share episode

In this episode of Lead-Lag Live, I sit down with Kai Wu, Founder and CIO of Sparkline Capital, to break down the AI-driven capital cycle reshaping markets.

From Nvidia’s record-shattering deals to hyperscalers pouring trillions into infrastructure, Kai explains why today’s AI boom echoes past capital cycles—and why investors may be missing the real risk.

In this episode:
– Why mega-cap tech is shifting from asset-light to asset-heavy
– The historic link between rapid asset growth and underperformance
– Why concentration in the “Magnificent Seven” is an overlooked AI risk
– Lessons from the dot-com and railroad booms that apply to AI today
– How to position across the full AI adoption cycle

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

#LeadLagLive #KaiWu #AI #TechStocks #Nvidia #Investing #Markets

Start your adventure with TableTalk Friday: A D&D Podcast at the link below or wherever you get your podcasts!
Youtube: https://youtube.com/playlist?list=PLgB6B-mAeWlPM9KzGJ2O4cU0-m5lO0lkr&si=W_-jLsiREjyAIgEs
Spotify: https://open.spotify.com/show/75YJ921WGQqUtwxRT71UQB?si=4R6kaAYOTtO2V

Support the show

SPEAKER_01:

These firms are starting to take on more debt, you know, often in the forms of these kind of off-balance sheet SPVs, which are a little concerning. And then you mentioned in the very beginning, kind of the circularity around the cross-holdings of firms investing in other firms to invest in other firms to buy the chips, which is, of course, something we saw in the dot-com boom, which was, you know, um, you know, indicative of shenanigans.

SPEAKER_00:

I'm your host, Melanie Schaefer. Welcome to Lead Lag Live. Now, AI stocks are writing a seismic shift in capital flows. Nvidia alone cinched a$100 billion deal with OpenAI, sending markets into overdrive. Multiply that by the cascading ripple effects through infrastructure, software, and chipmakers, and what felt like hype is now shaping multi-trillion dollar valuations. My guest today is Kai Wu, founder and CIO of Sparkline Capital. Sparkline Capital is an investment management firm applying a cutting-edge machine learning and computing to uncover alpha in large, unstructured data sets. Kai, thank you so much for joining me today. It's great to be here. So let's start with the big picture. Um hyperscalers are plowing through hundreds of billions or are plowing hundreds of billions into AI infrastructure. What do you what do you think these capital flows tell you about where the next leg of the opportunity could lie and what risks are you watching?

SPEAKER_01:

Yeah, I mean, this is something I'm watching very closely. Um, you know, as we all know, over the past 10, 15, maybe 20 years, the entire stock market has been driven by the performance of the so-called magnificent seven stocks. Um, you know, Google, Apple, Nvidia, Microsoft, these guys. Um, and, you know, really their success has been, at least to me, um, on the back of their business models being in kind of asset-light sectors, right? So think of Google Search as this perfect business that just throws off cash and doesn't require too much capital to keep afloat. Um, you know, my my work is, my research is primarily in the field of intangible assets. Um, trying to understand how items like brand equity, network effects, human capital, and intellectual property allow firms like these Magnificent Seven to earn supernormal profits. And, you know, for a long time, these companies have really been the poster child of asset-like business models, um, being able to just really dominate in terms of um, you know, earnings and profitability, and hence, you know, the share prices have followed. Um, and to a large extent, they've become a significant component of the stock market, right? I think about a third or more now of the SP 500 is now um, you know, held up just by these seven stocks. And so what I've been watching closely is this transition away from the model that's made them so successful, right? What we've what we've seen is these companies now set to invest, you know, trillions of dollars into AI data centers. And that's you know, these physical infrastructure required to power AI models, primarily GPUs and chips, but also the you know, physical data centers and um, you know, power and all this other infrastructure. Um, what we find is, you know, think of like the best example might be Meta, right? So Meta is, of course, um, you know, the former Facebook, social network, very asset-like business. They are now putting um about 35% of revenues into physical capital expenditures, right? And that basically puts them um in line with the average utility. Uh so we we now we're now seeing these these um formerly asset-like companies transition into a more kind of asset-heavy utility-like model. And I think to, you know, that that's concerning for a few reasons. I think the first is is simply the fact that if you look over the history um of stock returns, you'll find that the asset heavy companies have generally underperformed asset-like companies. Um, and these these stocks are again holding up the stock market and you know, are trading at valuations that are still reflecting their former glory. And I think the second concern is around the the delta, right? The growth in capital spending. And we can probably get into this later, but I mean, if you study the history of capital booms, going back to the dot-com boom, the railroads, um any big infrastructure build-out um around technology, it doesn't always end well for the folks building out the uh the pipes.

SPEAKER_00:

Yeah, so that's what I wanted to ask you about next. The the AI cycle is sort of echoing past uh capital cycles. Where do you see parallels today between this AI CapEx wave and earlier booms, like as you mentioned, the dot com and and where do you think the differences online?

SPEAKER_01:

Yeah, so I guess on the similarity side, um, you know, we we we're seeing a small group of concentrated players, in this case the hyperscalers, um, you know, really drive the the build out, right? We saw the telecom firms um do the same with the internet. And, you know, of course, there's a cadre of uh companies that were um laying out the tracks for the railroad boom um, you know, over 100 years ago. Um and so I think that there's that there's that similarity. Um I think the differences, you know, are primarily on the balance sheet side, right? So think of like global crossing as being perhaps a good example in the dot-com era of a company that um, you know, was over levered and um you know went bankrupt. And I think a lot of the railroads as well went bankrupt because of the way they are financed and they didn't really have existing cash flows. The difference, of course, is that the Magnifist and 7 are, as I mentioned, you know, supremely profitable businesses with up until today sterling balance sheets. Um now, this is starting to change, right? Where there are indications that um, you know, these firms are starting to take on more debt, um, you know, um often in the forms of these kind of off-balance sheet SPVs, which are a little concerning. And then you mentioned in the very beginning, kind of the circularity around the cross-holdings of firms investing in other firms to invest in other firms to buy the ships, um, which is of course something we saw in the dot-com boom, which was, you know, um, you know, indicative of shenanigans. Um so you know, I think we're coming from a better starting point for sure with these companies. And I don't think anyone is forecasting that, you know, Google is going to go bankrupt um because of the because of overinvesting in in um AI infrastructure. But I think it that the concern is more around, you know, just the dilution of their business model, right? If you're averaging something that's very profitable and something that's like not that profitable, you're not, you know, ending up in as good a place as you started off. Um, and and just around the risks um, you know, given market concentration, um, you know, around you know, how much is relying on this one, you know, this handful of companies and and more importantly, the one theme of AI. Um, you know, when AI does well, the economy does well, and when AI does poorly, the economy will do poorly. And so I think that is definitely an area of concern today.

SPEAKER_00:

Yeah, and academic research often shows as well, like firms with very high asset growth underperform. What does that tell us about pulling back the hood on uh growth today? And how do classic signals fail in this environment?

SPEAKER_01:

Yeah, no, I'm glad you brought this up. Um, yeah, there, you know, my my background is as more a quant researcher, kind of, and and and a lot of the work I've done, you know, especially earlier in my career, was studying a lot of the factory literature. So think of things like value, momentum, quality, and the characteristics, characteristics of companies that lead them to succeed on average and and you know, not. So there's a factor, you know, in kind of the um, you know, main models that people use called the investment factor, which I find is kind of a confusing name, but so I think it's better um termed the asset growth factor. And what that um what that shows is that companies that on a trilling 12-month basis, you know, in percentage terms grow their assets, so grow their balance sheet um, you know, more than average, tend to subsequently underperform. Right. And this is a pretty robust finding across markets and across time. And more importantly, it's not just about um technology, right? Of course, investment cycles can be um, you know, tech-led, where an innovation like AI or the internet comes around and everyone wants to hop in, but it can be as mundane as you know, cod fishing, or um, you can think of examples in banking with a financial crisis and housing, right? That's just the idea of um, you know, the capital cycle, people kind of invest too much and then realize subpour returns. Um, what's interesting with this factor is I've actually been doing a bit more work on it, trying to disentangle the different components. Because um firms, when they rate increase their balance sheet, they can be doing so um to accomplish a couple of different goals. Um, you know, the most important being are they spending the money on building a physical infrastructure or simply on RD marketing and perhaps creating other uh you know um more intangible assets? And one thing you find is that you know, it's really the physical capex that is most linked to the underperformance. And I think that has to do with just the the long-lived nature of a lot of this um this infrastructure, right? Like, you know, it's the these things, you know, semiconductors, for example, which is of course what we're talking about to some extent, is a notoriously um cyclical industry where if you look over the past few decades, you'll find these big booms and busts because you know, companies um see it see good times, they overforecast demand, they then overbuild, it takes a few years for the supply to come online, it comes online, they've overbuilt, there's a collapse, and so on and so forth. Right. And so this is just a history that we've seen in these kind of capital-intensive industries. Um, and you know, it's I think linked largely to what what this the academics find as well, with regards to the underperformance of these stocks on average. Um, the other thing you can do is you can disentangle the in this uh asset growth anomaly into two dimensions. So one is like the effect of just a macro, which is you know, you have these waves of um of investment um, you know, both both across the market, like we did with the dot com, but even within the individual sectors, like through airlines, let's say, and then there's consolidation and expansion. Um, and then also within each industry at the company level. And what you find is that the effect persists on both levels. So both a macro effect, where when an industry experiences a lot of capital inflows, they tend to underperform, as well as when individual companies experience um, you know, aggressive ramp up in their capital expenditures relative to their peers, those companies also tend to underperform. And again, like these findings by the academics are not meant to um be indicative of any one company. Like we're not specifically picking on meta in this case, but um, you know, just is a kind of average finding across sets of companies through time, just saying, hey, this tends to not work out that well if if applied, kind of if you were a betting man, right? And wanted to apply it um kind of consistently across names.

SPEAKER_00:

What does this then potentially mean for the stock market as a whole? I mean, like we see extreme concentration with uh a handful of megacap tech names carrying the market. Do you view that as a new AI risk factor? And how do you manage concentration risk uh in your own framework?

SPEAKER_01:

Yeah, and I think look, anyone investing in stocks is generally using as a starting point a cap-weighted index, right? So they're going to be buying stocks in proportion to their market caps. And of course, the Magnesin 7 stocks are have become so big that they comprise, as I mentioned, over a third of the index. Right. So right off the bat, you already um, you know, have your fate in the hands of this um idea. And then, of course, there are, you know, there is exposure just uh at the economic level to AI, right? That I think about the some people have estimated that about half of the GDP growth of the US last year was due to um in investment in AI, right? Because that's of course a component, private sector investment is a big component of AI. And then there's, of course, like the labor market impacts and and so forth of the technology. And so to the extent where you know investors are already concentrated in the Magnificent Seven by dint of their exposure to indices, um, you know, my thought as an investor would be, you know, try to find ways to diversify. Like you never, like maybe maybe AI will end up becoming the like the the real deal, and we will end up see seeing us, you know, a revolution in the way um the world works. And I and I do think so. Um, but as we've seen in past um cycles, oftentimes that can take a long time, right? Like with the internet, the adoption curve was very long. It took decades, right, for um, you know, the S curve to fully materialize. And in the interim, there were plenty plenty of periods, um, such as you know, the dot-com bust in 2000, you know, 2002, where stocks um, you know, investors were disillusioned and stock prices um, you know, troughed. So I think, you know, as investors, you want to be, you know, pretty cautious with regards to exposure to any one theme. And, you know, while, you know, even four or five years ago, when I was, you know, first doing research on this topic, I was very bullish on AI saying that, look, this is the real deal. We should all just kind of be doing AI. Most folks I know are underexposed to AI. I think it's now flipped, where, you know, fast forward four or five years, ChatGBT has now released, everyone's talking about AI. I think the risks are actually now the opposite, where everyone's now overexposed to AI. Um, and in particular due to just the cap-weighted indices um favoritism towards the larger stocks. Um so as an as an investor, I mean, my my thought would be to look for ways to um invest outside of the Magnificent 7, outside of these companies that are not only exposed most to AI, but are in fact, you know, exposed most of the kind of ticks and shovels uh capital expenditure component of AI. If you go back to the dot-com and railroad booms, you know, the the folks who actually built out the railroads, the folks who actually built out the telecoms, those guys did did really poorly. Um now there are some firms um that did well off those booms, but um, the guys who actually built out the physical infrastructure in in both cases did not do well, right? Um, and so I think that's in particular kind of the epicenter of risk for the AI boom. That look, the AI AI may continue to um be transformative and may ultimately be adopted um, you know, in well by enterprise and by by consumers, in which case, um, you know, the beneficiary, there will be many beneficiaries and also might not. But even if AI, you know, disappoints um, you know, in the in the investments that these companies are making in data centers underwhelm, that's actually almost better for the for the adopters, right? Like to, you know, when when the dot com bust happened and fiber collapsed um and was unused for a long time, companies could then, um, you know, like Netflix, for example, could then utilize that fiber at a below cost in a subsidized way to build other businesses. Um, and so I'd say, you know, focus more on the beneficiaries of AI as opposed to the picks and shovels. Um, I think that's probably a slightly safer way um to deal with the um you know the consolidation of risk around this single theme um in the current market.

SPEAKER_00:

So to just get in a little more deeply into this, could you walk uh investors through how they could position themselves um through the full tech adoption cycle if if that comes uh topics?

SPEAKER_01:

Yeah, I think the best way to think about this is, you know, by analogy to the dot-com boom, because we actually have a full cycle of history, we can kind of say, hey, at least with hindsight, how would I have done that? Um so basically with the dot-com boom, you kind of saw the adoption of the internet unfold in waves, right? Whereas in the very beginning, the say mid-90s, early 90s, mid-90s, you know, there was a lot of the innovation was just by, you know, these kind of pure tech companies, these pure play internet names, um, you know, AOL, Cisco, those would have been the best investments. But what happened was everyone figured that out. And, you know, these stocks were now trading at, you know, price to sales ratios above 30. Um, and I use price to sales because in some cases these companies were not even profitable. Um, so fantastic valuations. Everyone was betting on the you know new um economy, and these companies got bit up. So, what would have then been the right play, again, with hindsight, would have been to then rotate into phase two of the game plan, which is to not buy the innovators, but to buy the early adopters. Right. So, early adopters in this case, I mean companies that were positioned to benefit from the usage of the internet um in their own business lines, but weren't necessarily kind of pure play internet names. They're not, you know, telecoms or or dot-com companies. Those companies were still trading at relatively um affordable multiples, not too different from the market in general, um, yet still had a lot of upside um from the um further adoption of the internet. Um, right. And then and then from there, once those names got bit up, you kind of went on to kind of more mainstream companies that again, you know, we all got a GDP bump from AI um from the internet, um, and and those guys would have been the kind of the third stage. Right. So you can kind of see how this plays out. Um, you know, as adoption um goes, you want to kind of uh always be moving, right? If you only bought AOL, you would have done really well in the beginning and then done really poorly and end up you know losing money. Right. So I think the key here is to be dynamic and to be value conscious because value, valuations in a way, kind of help you organically navigate the cycle. It helps you kind of determine, hey, when am I starting to get to the point where you know the we go from phase one to phase two, right? So they're gonna help you naturally do that. And that that's how one would have first uh kind of um done the first cycle. Um, I think in AI, we are at least in my mind, starting to reach the tipping point between um phase one and phase two, where Nvidia is, you know, at this point, everyone knows it's an AI stock, right? It's no secret anymore. You know, if you bought it five years ago, you were probably ahead of the curve, you're no longer ahead of the curve by doing that. And you know, to a large extent, you know, the both the upside and downside of AI continuing to outperform or underperform is now is currently kind of perfectly priced into NVIDIA. And so if you really want to profit from um the continuance of AI, you want to look for the next wave of guys, um, the kind of early adopter category. Um so that's kind of where I see us currently positioned. Um, but again, like who knows, it could take longer or shorter to play out um than in past cycles. Um, history only, you know, rhymes doesn't repeat.

SPEAKER_00:

Yeah, and I think that's some really great advice uh and reminders for investors. Just lastly, Clyde, for viewers who uh want to learn more about your work and even connect with you, what's the best way to do that and where can they go to learn more?

SPEAKER_01:

Sure, yeah. I I I have a website, um sparklinecapital.com, and I post a bunch of different research papers, including some of the research that I've alluded to on this um you know talk. And then in addition, you can just reach out to me directly on social media. Um I'm on both Twitter and LinkedIn under the handle C Kaiwoo, C K A I W U.

SPEAKER_00:

Okay, thanks again for joining me, and thanks to everyone for watching. Be sure to like, share, and subscribe for more episodes of Deep Light Live.

People on this episode

Head Shot

Michael A. Gayed, CFA

Host