In AI’s Shadow

As I write this letter, the major U.S. indices are hitting new records. Every time this happens, pundits call the top, as though what goes up must come down in markets. I am here to say that now, as it always is, timing the market is a fool’s errand. Just as it is difficult to hold investments as they crash down, it is equally difficult not to hold them as they skyrocket higher. Just as it is a grave mistake not to sell when a crash is justified, it is a grave mistake not to buy when a company’s future prospects shift categorically.

As flawed human beings with imperfect information, the best we can do is have a process and stick to it. Discipline. That is the antidote to market timing.

KB Capital has always been a tech investor, and so have I. My earliest investments, in Apple and Alphabet, were core to my success. I have learned not to bet against U.S. technology. However, I have also learned not to chase things, as Ed used to say. I have never liked crowds; they make me nervous. Perhaps that is why, in my investments, I have always shied away from “crowded trades.” AI and all things to do with it are like a train in Tokyo at rush hour. I am not taking a position on whether valuations in AI names are justified. This is an incredible technological shift, and by some metrics I believe we have already reached something like AGI — though I will concede there is no agreed definition, and reasonable people disagree. I suspect many of the older folks who really run markets are still struggling to comprehend just what this technology will be capable of a year from now, let alone in ten years’ time. My portfolio is very much long AI, having been early to this train. What I am saying is that the AI trade is taking the spotlight from other very impactful trends that will ultimately be rewarded by the market when it inevitably shines its love light elsewhere, in search of new growth.

AI Is Eating AV

While we can argue about whether AGI has arrived (and I recommend the application of Mustafa Suleyman’s modern Turing Test if we wish to argue this point¹), we cannot argue about whether autonomous driving has arrived. I was a passenger in my friend’s Tesla about a year ago as it drove itself from Cherry Creek to my house. It did so with no input from my friend, and along the way it braked sharply as a woman ran a red light in front of us, preventing what would have otherwise been a serious accident (one that was entirely her fault).

AGI is nebulous, debatable, and academic, whereas autonomous driving is tangible, undeniable, and obvious. As a car guy, and someone who still anachronistically drives a stick shift, I am almost sad to see it arrive. But its benefits are undeniable: more productive time, fewer collision-repair bills, likely lower insurance rates, and removing that damned lady from the road. In fact, I am watching robotaxis quietly take market share from human drivers at an ever-increasing pace. To me, though, robotaxis are the least pure expression of where AVs will see the quickest adoption and the highest ROI for investors. That honor goes to long-haul trucking.

It feels like a market failure that only a handful of public companies are addressing this goldmine of a TAM.

Truckers are the backbone of our modern society, especially in the U.S. Trucking is a nearly $1-trillion-a-year industry by revenue, and trucks account for over 72% of all freight tonnage moved in the country.² The industry employs over 3.5 million professional truck drivers, and millions more jobs across the economy are tied to it. It is also highly fragmented: about 96% of trucking companies in the United States operate ten trucks or fewer. With that backdrop, it feels like a market failure that only a handful of publicly traded companies are addressing this potential goldmine of a TAM. The purest of them is Aurora; Tesla comes at it from the side; and Kodiak, now public via SPAC,³ rounds out a very short list — one whose history is littered with first movers, like TuSimple and Embark, that did not survive.

Aurora Innovation is a lovely company with what I see as a durable first-mover advantage in this space, now that Google/Waymo has stepped back from autonomous trucking to focus on robotaxis.⁴ I believe they have taken the correct approach to this problem. I believe Tesla has not. Tesla is approaching this market by producing a truck (and we have seen how bizarrely that can turn out, vis-à-vis the Cybertruck). What’s more, we have seen how few people or businesses are willing to shell out for a massive electric vehicle that depreciates faster than Usain Bolt scalded by a hot poker. In this fragmented industry filled with undercapitalized small family businesses, a very expensive truck is the last thing they need. What they need is a bolt-on solution to their existing fleet that lets them move their already scarce drivers toward last-mile delivery while the long, arduous miles are handled by a system that does not get tired, drunk, or fed up with low pay.

While it is my belief that AUR will be bounded for some time by a lack of GAAP profitability — and by the overhang of Uber’s $1.0 billion notes, which are exchangeable into Aurora shares at $8.50 and cap the stock’s near-term ceiling more than they dilute it⁵ — I also felt that at $4 a share, and with a viable path to free-cash-flow profitability by the end of 2027, I could not wait around either. The path from here to full-scale adoption of AUR’s solution is much like the roads it seeks to drive: winding, uncertain, and full of hazards. But I feel the current political administration provides one of the most supportive environments for AV regulation in some time, especially given its seeming indifference to the concerns of labor and unions. In short, I see an inflection point in this company, and it would appear the market does too, given the roughly 60% gain since I purchased the shares only a few months ago. But like AI, AVs face a common bottleneck. We all know that trucks need oil to run (Tesla’s excluded); what every AV needs is data.

Data Is the New Oil

In the past, access and proximity to oil have defined our entire geopolitical picture. Germany’s inability to secure the Caucasus oil fields in its bloody struggle with Russia in World War II hastened the demise of the Third Reich,⁶ and is a large part of why I sit in such comfort in America writing this letter today. I believe the same dynamic will play out with data. But what is data? Each press of your phone screen, movement of your eyeball, letter typed, and link clicked is a datum: a valuable and underappreciated asset in the age of AI.

Again, I am not a technical expert on the subject, but it is one I have read a lot about and discussed at length with experts in the field. The way AI “learns” is via the ingestion of data. While there are various forms of AI — labeling, deep learning, backpropagation — they all fundamentally rely on data to learn and become “smarter” (more accurate). What’s more, given the incredible pace of AI adoption, we are on the verge of running out of it.⁷ So vast was the Dyson of AI training that almost every human work has been hoovered down its cavernous depths. This has led to the use of “synthetic data,” which might be more accurately termed bullshit.⁸ Synthetic data is data the AI makes up. To be fair, in narrow, checkable domains like math and code it genuinely works; it is in the open-ended, real-world stuff that it tends to fold in on itself. Imagine that in my economics course I had tried to predict the movement of property values from the number of trees in a neighborhood. Had I run out of neighborhoods to survey within driving distance, I might have used the trendline to invent more data along it. You can see how that would quickly gut the predictivity of my model, right? All of a sudden, new neighborhoods with actual trees — and the data that comes from them — are incredibly valuable to me in my pursuit of a higher R².

I believe we are in this moment now. What’s more, I believe the quality of the data, and its uniqueness, is a moat the market is undervaluing. As a friend of mine in the space says, “the winners in AI will be those with the data.” As Stalin reportedly once put it, “quantity is a quality all of its own.”⁹ Those with the most data (Reddit, Alphabet¹⁰) will be rewarded. Where I think the market is not paying attention is to those with the highest-quality data: the most unique and hard-to-replicate datasets.

I continue to explore which companies most embody this insight — and trade at a price that does not fully reflect it. One early expression of the thesis is Caris Life Sciences. Caris does not have the largest molecular dataset in oncology; Tempus, among others, can lay claim to breadth.¹¹ But I believe Caris has the deepest. It runs whole-exome and whole-transcriptome profiling on essentially every case — a far richer per-patient picture than most competitors capture. Its growth in treatment and diagnosis is inflecting, and I believe it is on track to reach free-cash-flow profitability within its next annual report. As such, like AUR, I am not sitting around to find out; I have taken a nibble of the stock, as Ed used to say.

Caris also embodies another trend I feel is being ignored in AI’s shadow: the aging of the American population.

Boomers Don’t Want to Be Doomered

Possibly the most universal desire is to keep living. What we have now is the wealthiest generation of old people ever to walk the earth — Americans aged 55 and older hold roughly 73% of the country’s household wealth, and the boomers alone hold about half.¹² They, too, do not want to die. That combination — vast wealth and an unwillingness to go quietly — will drive an astonishing level of healthcare spending. U.S. healthcare outlays already dwarf AI capex; what interests me is the incremental wave of longevity and personalized-medicine spending layered on top, and I want the fund positioned to benefit from it. Personalized medicine is a burgeoning field, and the pace of breakthroughs is, to my eye, something neither the market nor humanity has fully priced in. We may soon see life expectancies that look impossible today. And I would not bet against us curing many cancers within my lifetime — perhaps even within the decade.

This is a theme I continue to explore with the help of experts, including MDs. I hope to find the same data moats, the same process moats, and the same wonderfully positioned businesses that have brought me such success in tech.

Investing Is Hard

I have not found that yet within this theme, outside of Caris — and Caris is far from a bargain at $15 with no earnings. What makes investing so hard is that it is not enough to identify a great business, with a great moat, a stellar management team, and demographically driven demand. I also have to pay a reasonable price for it to make any money. I’ll leave you with an experiment I gave a friend at lunch recently that captures my predicament. Pick a large-cap stock at random — be a monkey with a dartboard — and I will bet you $100 it trades above 20x forward earnings.¹³ Write to me if it is not, and I will pay; if it is, send me a check, please. In this market, the odds favor the house: the index sits near 21.5x forward earnings, and the megacaps that anchor it trade richer still.

Finding a good deal right now is harder than it has been since I started investing at 16 (for reference, I bought my first shares of Apple at less than 10x forward earnings). That is why my cash pile keeps growing. I expect to underperform the market for the next year, two years, maybe three. But I also expect that patience will be rewarded in the end. For now, I am in no rush to deploy cash. I am taking my time — doing the research, identifying the great businesses, and watching for the chance to buy them at what I consider fair prices. To you, that may sound like market timing. To me, it is called discipline.

Thank you for reading my thoughts today. This is not investment advice, and it is likely worth what you paid for it.