Dear Valued Clients and Friends,
I decided to write about this topic several weeks ago, believing the story of AI, capital expenditures around AI, the role of such companies in the stock market, and where this all goes from here to be among the biggest stories investors have faced in a long, long time. There is a potential inconsistency in what I just said, as I spend a great deal of time [accurately] insisting that the primary determinant of investor success is not something to be found in the headlines. “Noise” in markets, as we often call it, does not drive investor returns. If what I were saying were that there is a lot of “noise” around this AI talk right now (for good or for bad) and how investors “play” the noise will be really important for their portfolio, then yes, I would be guilty of contradictory messaging. But I am actually saying something very different. And what I am saying about the potential of an AI bubble is the subject of today’s Dividend Cafe.
The decision to write about this topic came at some point in mid-October, well before the recent ~10% correction in Nvidia and the 15% drop in AI darlings such as Palantir. Nvidia releases earnings this coming Wednesday, and they could make up all of the recent drop in 5 seconds, or they could go down further, or anything in between. None of this is, or could be if I wanted it to be, timing-sensitive. I have no opinion on what sentiment will be in the short term. Today’s commentary is meant to be much more substantive than a silly short-term market projection (how financial professionals rebrand “rank speculative guesswork”).
I set up this topic differently about a month ago, as the conversation about whether someone should “abandon their philosophy or plan just for a while” was growing louder. I avoided saying then what I was thinking (what I actually know to be true) – that the more people say and ask things like that, the sooner the reckoning will come, and the more substantial the reckoning will be. But nevertheless, I thought that Dividend Cafe covered a lot in terms of what the investment success in AI had actually been so far, trying to shine a light on the circularity of a lot of the story, but also explain why certain speculative assumptions (that may very well play out) were a major part of the thesis, and that I was not even remotely tempted to abandon what I believe about investing to chase this moment. I subsequently wrote about the misunderstandings around AI’s impact on jobs. There are a lot of aspects of the AI conversation that matter – politically, economically, and culturally. None of those things is the subject of today’s Dividend Cafe.
Today’s Dividend Cafe is one of those classic MARKET COMMENTARIES I love writing. As much as I live my life in the nexus of politics, economics, and culture, the Dividend Cafe exists to provide macroeconomic and investing commentary useful to investors. And when I talk about the primary determinant of investor outcomes being something far more important than headlines and noise, I am referring to investor behavior – to avoiding the big financial mistakes that I have seen do unspeakable harm to more people than I could ever count over the years. And that is the subject of today’s Dividend Cafe – where the current AI story lends itself to one of the most classic investor mistakes of all time, one repeated so many times throughout history that you almost believe human nature is incorrigible.
Because human nature is, in fact, incorrigible, we all have a job at The Bahnsen Group. And to that end, we work, not just to illuminate truth in today’s Dividend Cafe, but to guide our advisors away from the big mistakes and toward a financial journey that can be agnostic about these moments in time. And I assure you, my friends, we are living through a really remarkable moment in time.
Let’s jump into the Dividend Cafe …
| Subscribe on |
Starting with The Conclusion
I am very happy to get my bottom line out of the way in case any of you want to stop reading early, I believe the broad AI-capex story that has dominated markets for some time is in a bubble, is going to end very badly, and will do significant damage for those who decide to abandon a more disciplined approach to euphorically join this party. I believe that, for many reasons I will share in today’s piece. And I also believe there will be great investment opportunities as a result of the AI moment. The risk is not, per se, in what AI does, or when, but rather the high-level decision-making criteria that do not differentiate between sensibility and euphoria. I do not believe any decision made from FOMO ends well, but the stakes are far, far higher with this (due to sheer exposure, concentration, and severity of valuation at play).
But more than some prediction of carnage, of people buying the Pets.com companies of 2025 that expose them to 100% losses, I believe the likely outcome in all this is very different, very fundamental, and very intertwined with certain market realities that no one is really talking about.
Timing is Everything When You Look Back in Time and Describe It
I do not believe that saying this story is in a bubble means the bubble will not grow before it pops. It may not – for all I know $212.19 on October 29 was Nvidia’s generational top. And for all I know, it will add $500 billion, or $1 trillion, or $2 trillion to its market cap from here, in the blink of an eye. That is part of the absurdity of the moment when valuations are 50-60x earnings and people talk like “trillions” with a T is normal stuff – Nvidia has lost $500 billion of market cap in the last fifteen days and no one seems to care; and it added $2 trillion in the ten months before that (after adding $1.5 trillion last year), and everyone acted like that was normal. None of it is normal.
I am not picking on Nvidia. It is the hands-down winner of the AI stock story of 2024 and 2025 so far, with no even close second place (well, actually, there is another). You know what is interesting, though? Nvidia was $150 one year ago. So roughly a year later it is $185, meaning on a time-weighted basis it is up about 23% in that window. That is pretty darn good, of course. It is also basically what the S&P has done (not quite, but you get the idea). It is less than many other companies in that same window. Now, I am cherry-picking dates, but the point is that with these huge moves up (and yes, certain huge moves down), there are people whose entry time into a company like Nvidia is going to be the primary determinant of their return there. A stock over one year, two years, or five years, may have a huge return, but the time in and time out may significantly hinder the return real life people get. For someone who bought at $209 two weeks ago, if it hits $170 in this current drawdown, they will need a +25% return just to get back to even. Holding fast during a huge secular move for a stock is one thing. A low basis gives you a very different psychology as a high beta, high volatility stock like that moves way up and way down. But entering at $209? Entering at $150 and then seeing $100 six months later? (down 33% in six months)? It just changes the math substantially. What about people who bought Oracle for $345 two months ago? It is now sitting at $217 two months later (down -37%), and yet still trading at 50x earnings (ay yi yi). The stock now has to go up +59% for people who bought at that level after the “big announcement” just to get back to even!
These are the things that cause investors to throw in the towel. These are things that make people give up. And these are the things that people are buying into right now where sentiment and psychology have surpassed common sense and fundamental logic.
Why did Oracle hit $345 two months ago before dropping -37%? As best I can tell it is because the announcement came that a company with no money was going to give them a bunch of money to make things that would require them to buy a bunch of things from another company and that company invested money in the company that gave the initial order to Oracle. Fine. I find the circularity problem of companies whose stocks go up because they buy things from each other to be problematic. But in this case, my only point was that this story might actually really work out well – Oracle might deliver huge value to Open AI, who got funding from Nvidia, and who becomes the supplier to Oracle – but if you bought at that hype and the whole story creates +59% of gains from here, you are going to make 0%, because of math.
As one client said to me 25 years ago when I explained that a stock has to go up 100% to be back to even after dropping 50%, “that’s not fair!!”
I am not yet talking about what will happen with Nvidia or Oracle or even OpenAI. I am talking about math. What Nvidia did the last five years (some of the greatest individual big cap stock performance in history) is now completely irrelevant for future returns. It is baked in the past and immaterial to the future. And all I am trying to gently remind people is that the “play” here in this whole story has become unbelievably timing-dependent, and not just entry-dependent, but exit-dependent.
Consider Coreweave (CRWV), a darling of AI cloud platform and infrastructure. The stock was $187 five months ago. It is currently $77 and trading at negative 38x earnings (I will let you do the math on what the P/E was in June). The stock now has to go up a measly +142% to be BACK TO EVEN for the massive amount of people buying massive amounts of stock in June. Is this company getting worse or better? It is getting much, much better. But the outlook for investors who didn’t trade out in June is bleak. If the stock is up 20% per year (double the S&P average return) for FIVE YEARS, an investor will … be back to even from June 2025.
I repeat for those in the back: none of this has anything to do with the companies mentioned. Many people bought a long time ago and will let it all ride. Many people own index funds and don’t even know, let alone care about, what I am referring to. I am merely trying to explain the value path for investors now, at this point in time: It is highly dependent on entry and exit timing, where the risk along the way is mathematically daunting. If not done with “speculative play money” – if it is the substantive part of an investor’s core portfolio, it is not just mathematically daunting, but economically existential.
Okay, Math Bath. What are You Worried About?
The premise behind the AI trade is one of two things:
- Hyper-scalers are spending so much money on AI capability build-out that, regardless of how their investment works out for them, there is so much money to be made by those companies receiving their capital expenditures (the expense for some is the income for another) that this AI trade is good enough on that basis alone …
- The money the people investing in AI will make once they convert the computing power into real-life applications is so big that, even though we don’t know how or what they will do, it is worth just going along for the ride as these companies lay the groundwork for huge new profits from some form of AI application and utility.
If you find yourself reading either one of these theories of the case and being tempted, you are not reading them as, ummmm, I intended.
The “build-out” part of AI is where everyone’s focus is now. It is why Nvidia is all the rage. Some future utility of AI is believed to be coming (fair enough), and Nvidia is at the ground zero of AI infrastructure (the chips needed to make it all possible). Many have referred to this as the “picks and shovels” phase. At some point, this has to lead to usage and application. And from there, that usage has to become economically rational, provide a return on investment, and generally increase productivity. If that final phase doesn’t happen, the second phase stops (why do something that doesn’t have a productive, positive impact), and if that happens, the first phase becomes a very expensive mistake.
The problem is that I can’t think of any precedent where the huge anticipation for a “transformative thing” did not go from a build-out phase to a utility phase without a massive purge of dead weight from excess in the build-out phase. Start at the railroads if you want, but go to radio, to fiber optics, to media, to telecom, to shale oil – those transformative parts of modern society find winners once investment gets rationalized, but not before massive, and I mean, massive, excess investment has to be liquidated. It is an economic law that stems from a law of human nature – in this case, the law of FOMO. Excess comes from fear (of missing out), which is to say greed, which is inevitable.
The brilliant Louie Gave distinguishes between “productive bubbles” and “scarcity bubbles.” With a “productive bubble,” you get overinvestment and excess capital spending driven by misperceptions of future profits. Those things have to get re-priced, but when all is said and done, some useful asset is still there, albeit in the hands of different owners than the now much poorer owners who previously owned them. In a scarcity bubble, the assets were not productive, yet the perception of scarcity drove their prices up. When the game ends, you are not left with a productive asset – just something whose value was tied to a narrative that has now gone away. Louie uses tulip-mania, Japanese land, and 1970s gold as examples. I suspect he will soon have another example to use in the annals of history, but I know many crypto-lovers disagree (they disagree so much, in fact, despite their confidence in the investment thesis, that they will devote all sorts of emotional energy to explaining why there is value in something that so obviously has so much value). But I digress … With a scarcity bubble, when it pops, you are not left with something productive that has merely switched ownership hands and been re-priced … The value story just went away.
I believe that current AI investing model skeptics can very well believe we are in a bubble, without believing the end result will be an unproductive asset or technology. In fact, I very much believe the opposite. But I am not writing about the macroeconomic opportunities out of AI productivity right now – I am writing about how it all plays out for individual investors.
AI technology and infrastructure might be in the “productivity” bucket, but a bubble still results in significant value deflation and an outright change in ownership. That is what I want to avoid for my clients.
One of These Bubbles is not Like the Other
The other categories Louie mentions in talking about “productivity” versus “scarcity” bubbles are “equity”- funded versus “debt”- funded. This is a distinction I have written about for many years, and more recently as well. This distinction matters a lot in understanding bubble bursts that become systemic versus ones whose pain is more selectively felt by owners and risk-takers. The greatest part of the dot-com collapse of 2000 was that the pain, as brutal as it was, was truly felt only by those who had gotten in the water. Those on the side of the pool, or those who didn’t even come to the pool party, were pretty much unaffected.
If you want me to describe a PRODUCTIVITY, EQUITY-FUNDED bubble like this AI craze in the same way I would describe a SCARCITY, DEBT-FUNDED bubble like the housing craze of the 2000s, it is simply not going to happen. Lots and lots of tears can be shed in both, but the nature of the pain and damage with debt bubbles is exponentially worse than equity bubbles, not to mention the difference in the residual of the productive asset underlying the whole thing.
This is a pretty sanguine view of the systemic risks posed by the AI bubble fears I am describing. But note the thing I am taking for granted … that it will remain an equity-funded craze. I am not at all sure that is true. Much like a homeowner with massive amounts of protective equity is somewhat insulated from a decline in housing prices, these “hyper-scaler” companies are so incredibly valuable, cash flow positive, and unlevered, that if anyone could ever afford to go overspend on AI capex by hundreds and hundreds of billions of dollars, it is this group of companies (in theory). But wait, there’s more … Not all of the companies that have shown up to the dance print money from the free cash flow machine that Apple, Google, Microsoft, and Meta do. In fact, OpenAI has committed the most funds in the AI world so far, with $1.2 trillion (or so) over the next five years, which strongly suggests they will participate in a debt bonanza.
The capital structure that this all takes on is very, very uncertain at this point across the multiple parties and counter-parties that are sure to be involved. Maybe one big tech hyper-scaler pledges from their own cash-printing earnings for one order of chips, but the commitments to build data centers to create the power to fund the AI need are filled with promises of borrowed money. Will sellers provide financing that represents borrowed money (even if not from an external lender), thereby changing the risk profile of these transactions? Will it get to a point where there is no choice? Counter-parties, adjacent transactions, three-way transactions, debt that looks like equity – there are so many things unknown and on the table right now that all I can say is … my sanguine outlook about this as a productivity bubble funded by equity could easily, easily change.
A List of Concerns
- Selection Risk – One may believe that the AI story and capex behind it is all going to come together, and monetization is going to be found, and there is going to be a big, happy ending, but they cannot possibly believe that all the companies competing for this computing power will win. They can’t ALL win. Some have to lose for others to win because the capex is way, way too high to work out for everyone. For it to work for some, it has to not work for others. This is the math and competition of it, and I think this is the biggest thing missing in the story of AI capex. Investors seem to either want to own all the names, which guarantees some major problems as some [might] succeed, and others [for sure] fail … OR they believe they will correctly pick the successful names and avoid the unsuccessful ones, which is fine, in theory, but doesn’t seem to be properly understood or appreciated.
- China – I do not suggest that China is going to beat the U.S. in AI. I do not suggest that they have tools and applications better than those built in the U.S.. I do suggest that they are trying, and the confidence with which some dismiss that risk is utterly mystifying to me. To those who scoffed at Intel ever losing chip supremacy to some small island country near China … Taiwan Semiconductor called: They said, “Who’s laughing now?”
But Nvidia’s dominance, our hyper-scaler investment, and our programming superiority all mean this is different, right? 40% of Nvidia’s sales are going to Malaysia, Singapore, and Taiwan. What exactly are they buying, and what are they buying it for, if global competitiveness in the AI investment race is not a real thing?
- Investor naivete – I do fear that many people invested in this trade were not around for the late 1990s and the 2000 dot-com implosion. I routinely talk to managers and investors who believe they are bulletproof. This trade is not bulletproof. Many investors believe that strong earnings growth warrants permanent multiple expansion, no matter the price. I do not find that proposition to be true. Narratives work until they don’t. The current investment environment has lent itself to narratives that are not written in stone, because no investment narrative is.
- Revenue growth deceleration – I accept that today’s tech companies are higher-quality businesses than those of 2000. But I do not accept that revenue growth rates will continue to accelerate. In fact, I believe those growth rates will decelerate substantially when capex as a percentage of revenue is flying higher!
- The differences are not all positive – everyone loves to point out that this AI moment is very different from the tech bust of 2000 because the companies these days are higher quality, with much higher earnings and cash flow. That is certainly true. But there are other differences, too, that are not as positive. Prior loved tech plays were extremely capital-light; this new era is brutally capital-intensive. The earnings of a lot of the big tech loves of the last era were gigantic; many of the companies in this AI moment are babies with no earnings whatsoever.
- Rational pricing is out altogether – people like me come off like a broken record when making empirical observations about valuations. Investors tend not to care about valuations when stock prices are going higher. But when investors began changing earnings expectations to ludicrous heights just to make sense of things, or even worse, just abandon trying to make sense of things altogether, history is clear that diminished returns result. Sometimes things break and can’t be put back together (Pets.com type stuff), and other times they just lead to prolonged periods of subpar returns. Both are concerns about this AI moment. I see more companies trading at greater than 10x revenues than ever. I see some companies trading at over 100x revenue. I see things that are actually worse than they were in 1999 in plenty of ways, and then I see mental gymnastics to explain why what I see isn’t what I see. This usually ends poorly.
- Misallocation of resources – if I am right that an excess of capital is chasing opportunities in AI, we have to consider both the visible effect (too much money chasing too few returns in one space) and the invisible effect (not enough money chasing attractive returns in another space). A diversion of capital away from certain needs is a concern, and one has to believe that the sheer size of AI capex is coming at a cost of other economic needs.
- Depreciation concerns are real – I recognize some sounding the alarms here are short-sellers with an axe to grind. But I am not a short-seller, I don’t have an axe to grind, and math is math. But it does seem to me entirely plausible that the lion’s share of the investment hyper-scalers are making is in computing power and technology that will be obsolete by the time they can get a return on them. Their claim is that the infrastructure expense is for the water and the power and the energy, and once that is built, it can depreciate over a long period of time, whereas the chips and other purchases can be replaced. But this strikes me as fanciful. The major expenses are the things that represent Nvidia’s major income. Why are those paying for AI capex boosting earnings by using long depreciation schedules for AI hardware? These chips are not long-dated assets, period. I do not want to overconnect my concerns to this category, but I do believe it is valid, potentially significant, and worthy of inclusion on this list.
Conclusion
I believe AI is in a bubble, and I believe it will be the most important technology of this century.
I do not believe these two things are a contradiction.
I believe the historical pattern of a bubble, then a crash, then a normalization will play out here as it has in all sorts of transformative eras that morphed into a mania.
I believe the above sentence applies to the AI-specific story and not the entire market, as was the case with the Nasdaq/tech crash of 2000.
And I believe that those who experience their first tragic moment as an investor in this AI moment will learn from it, and become a better investor in the future because of it.
But to the extent it is up to us, we prefer people avoid the pain to begin with. To that end, we work.
Chart of the Week
Nothing to see here.
While the chart above highlights Mag7, which are riddled with AI and AI-capex adjacency, 41 S&P 500 stocks can be said to be in this vein. Those 41 stocks have accounted for 71% of the S&P 500’s gain since November 2022 (when ChatGPT launched), with the other 459 stocks accounting for 29% of that gain. And those stocks – 41 out of 500 – are basically HALF of the S&P 500 – HALF!! (47% to be precise).
Quote of the Week
“Your boos don’t mean anything to me. I’ve seen what you people cheer for.”
~ Rick Sanchez
* * *
I thought I would keep today’s Dividend Cafe to 3,000 words. I got near 4,500 and had a lot more I had to leave on the cutting room floor. Forgive me if it went on too long, but there was a lot to say on this topic. It isn’t a topic that is going away any time soon.
This current sell-off in the space is not the subject of today’s Dividend Cafe. This sell-off actually could very well be noise. But I don’t know that. I just know that I believe in what we are doing, and I fear owning a bubble that pops. And so I have to make sense of the unbelievably valuable parts of AI technology without the risk of the things that do not make sense to me.
Thank you for believing in the process. It is a worthwhile journey, I assure you.
With regards,
David L. Bahnsen
Chief Investment Officer, Managing Partner
The Bahnsen Group
thebahnsengroup.com
This week’s Dividend Cafe features research from S&P, Baird, Barclays, Goldman Sachs, and the IRN research platform of FactSet
