The AI Personification Trade
The Next Leg of the AI Trade is Robots, Not Data Centers
Most everyone with an equity portfolio has heard of semiconductors and AI. Or, rather, everyone in the developed world with a pulse has heard of AI. So much so that people who have no idea what an LLM is seem to know—without question!—that NVDA will only go up. Semis will only go up. Mag7 will only go up. It’s “the future,” for goodness’ sake. Give me the TQs.
The thesis for the AI trade to date has largely been predicated on the race to see who can make the most advanced chips, buy the most of them, and proliferate data center capacity to drive the “AI revolution,” the Holy Grail of productivity. We will see data centers everywhere, AI will do all the thinking for us, businesses will see efficiency gains they previously only dreamt of in their most grotesque capitalistic dreams, and the future is now. Let’s go.
But… is that true?
Is that the way to think about forward-looking opportunity tied to AI in the equity markets?
I don’t think so—at least not entirely.
In this note I’ll lay out why.
While the general thesis above is likely to prove true over the next five years—efficiency gains for business, unlocking new frontiers with data, expediting R&D—those ideas have a high probability of being priced into current market valuations and have been for some time now. How do you know? Because prevailing narratives are, by definition, priced into current equity valuations. That means they form the baseline set of assumptions supporting current market pricing at scale.
But investors may be misreading the future of AI and how that technology will actually be deployed.
What does that mean?
The probability that the hype around data centers is overblown is growing, and the less discussed next leg of the AI and semiconductor journey is, logically, robotics.
That statement rests on a few observations:
Data center expansion is gated by power generation.
Data centers have high potential to become politically untenable if power consumption negatively affects citizens’ daily lives.
For national security reasons, we do not want data center proliferation here, there, and everywhere.
We have yet to see AI and data center consolidation.
It is unlikely that AI will replace as many jobs as people think it will in the next 12–24 months.
Raw materials needed for advanced semiconductor manufacturing are not entirely located domestically or in friendly or uncontested areas, and the supply chain could be easily disrupted for key materials (think neon), much as we observed in 2022.
If America is going to rejuvenate its industrial base, it will need robot labor to be competitive with other countries that are heavily investing in AI-powered automation (China).
These assertions hold based on currently known information, meaning:
Power needs are increasing, but the ability to generate power is not yet keeping pace, and that gap can be closed only for some corporations that are permitted to and can afford small nuclear reactors (SNRs).
To the point above, if data centers proliferate beyond energy production capacity, they will face material legislative headwinds (and already have in places like Oregon via the POWER Act, HB 3546), where the populace cares more about staying warm than about anyone’s compute needs. A single 30 MW data center can use as much power as a city like Eugene, OR.
In an ironic twist, we are supposed to need “more” of the thing (AI/data centers/semis) that allows us to have “less” of other things—more compute, fewer people. But put your .gov hat on for a moment. Do you really want private industry to have 10, 20, 50, 100 independent AI businesses that may overtake your Palantir AI systems and be sold to the highest bidder on international markets? This is not some James Bond scenario; it is capitalism, and the scenario above can, has, and would happen. It must be gated.
This consolidates points 1–3 into, well, consolidation. If you recall the tech bubble, we experienced significant consolidation of internet platforms. AI has a low probability of being any different.
AI proliferation for corporations may reduce the need for various administrative tasks but is unlikely to cause wholesale disruption of white-collar labor in the next couple of years. The AI has to learn, business has to develop confidence in it, and even then it is largely going to be used to replace existing menial tasks in the near future. We are a good way from businesses deploying AI to its full potential—gated by cost, confidence, and AI capability.
Advanced semiconductor manufacturing is… advanced. It is also resource‑intensive, and some of those resources are relatively scarce from an American perspective. We do not entirely control the supply of what we need. This leaves room for disruptions and, despite the current administration’s desire to shore up our resources, we are not there yet. Neon, as an example, is a byproduct of steel manufacturing. Highly pure neon is important for manufacturing advanced chips. Where do we get a lot of our highly purified neon? We get it from Ukraine—and Russia knows that. Sensitive. Pre‑war, depending on the estimate, something like half to the vast majority of semiconductor‑grade neon used in chipmaking lasers came out of Ukraine.
Attempting to run a world‑leading service‑based and manufacturing‑based economy simultaneously is an ambitious goal and not a problem in theory. The functional problem is having the people count to do both at baseline. More people bring more problems as a function of scale, which is expensive all around and runs directly counter to what the AI boom is attempting to accomplish. If we are going to keep our collective national headcount at a level that does not destroy important considerations (budgets, resources, welfare programs, security, etc.) and run an economy that is robust in both service and manufacturing, we are going to need robots. China already figured this out, and the US is behind—which means major investment potential in the good old USA. To this point, China has roughly 470 industrial robots per 10,000 manufacturing workers. The US, by contrast, sits around 295 robots per 10,000 workers. China doubled that capacity in the last four years. The US is behind.
Point seven above is the one to hold onto for consideration as I think forward.
Holding on to point seven is important because, eventually, we will have a correction at the index level, and with it will go semiconductors and “AI.” That presents the opportunity for investors to reflect on their thesis (AI and data centers), and doubly so if they participated in the short‑term decline, as high‑flying positions tend to get smoked in correction scenarios. This down‑market phenomenon fades sentiment for market leaders and the leading narrative as the prevailing thesis is called into question (data centers) or unravels (price action plus goal‑seeked observation reinforcing risk‑off bias). Identifying the mismatch between wrongly faded sentiment and future application (too bearish) is a place you find opportunity.
Related, a way to think about “corrections” or “bears” is to appreciate that markets need to reset to pivot. Not fun, but normal. The current narrative will continue until it doesn’t, and it is not common to see a new big shift emerge while the existing thesis remains strong. By definition, it is a shift—which is doubly hard for most to identify when the “shift” is not wholesale but subtle, like data centers to robotics for AI and semiconductors. Put another way, there is only so much capital, and wise investors think of the future. Down periods in the market allow investors to reflect on deploying that capital into what comes next—not what has been—when the market is “on sale.”
So what could a person do with the above?
Well, corrections hammer the high flyers, which right now include all things AI, and I doubt that correction comes from anything AI‑specific. Retail thinks in terms of “What can I buy to get rich?” instead of “What can I invest in today that will make me a lot of money over the next 12–24 months or beyond if I stay disciplined to a sound thesis?” Retail may blindly BTFD, and some professional money managers (PMMs) will reload on what was hot because of FOMU (fear of material underperformance) or optics, while others will take the opportunity to think ahead to the future application of the AI and semiconductor trade. And that future is the robotics trade powered by existing or near‑existing AI and semiconductor capability.
The rotation to robotics likely remains good for semiconductors as we pivot from “advanced chips” to existing chip technology seeing increased demand (volume versus bleeding‑edge technology driving earnings) as we proliferate robotics in manufacturing, defense, healthcare, and the service sectors to start. Low‑hanging fruit for repetitive tasks is where robots shine—which, coming full circle, means existing AI already has the compute capability for that now, as it is already taking over menial tasks in the digital realm.
Robotics is what I think of as “the AI personification trade,” and that personification has a high probability of being the next leg of the “AI will change everything” narrative.
And, accordingly, the next leg of the AI trade in a new form: robots.
Get ready for the bots, folks.
Disclaimer: As always, this is perspective and not an investment recommendation. You are free to disagree, like, share, shame, generally ignore, or love and laud any of the above. It is here for perspective, which I hope you found thought‑provoking and valuable.
To that end, I am The Unemployed VP by choice on a break from corporate life to spend time with my family, your support is welcomed, appreciated, and will come with more engagements as time unfolds. I am in free agency, but nothing is free, you know? Much more to come.
Thanks for reading the first edition of The Unemployed VP.
To successful investing,
~UVP