VD 106: Garbage in - Garbage out
In this issue:
Garbage in, garbage out
Stock in focus: requested by our readers: Google, Cisco, Bekaert and Floridienne
The Rationality Test: Constellation Software
What I’ve been reading these past few weeks
News from our companies
Doubler Portfolio update
Garbage in, garbage out
The stock market is never boring. After investors flocked en masse toward ETFs and AI, it seems commodities are now having their moment, with gold taking a starring role. Toss a president like Trump into the mix, and you simply don’t know what tomorrow will bring.
Can you invest profitably by chasing one hype after another? If I’m to believe social media, the answer is yes—the massive gains there are practically slapped in your face. However, it’s just not my style. Give me steady business analysis any day, where you actually get to know and understand a company to estimate its long-term prospects.
This calm approach also grants time to think, which is exactly what I’ve been doing these past few weeks regarding AI. Admittedly, this was partly prompted by suggestions sent to me by readers. Many of those companies benefit in one way or another from the current AI boom. They aren’t the AI developers themselves, but the companies providing the tools to keep it running: the so-called “picks and shovels.” This is analogous to the tool sellers during the American gold rush in the nineteenth century.
It’s a clever move, of course—looking at who profits from building data centers, cooling systems, and the necessary materials, rather than trying to predict who will build the best AI.
By now, you know that I view the valuations of many of these companies with a fair bit of skepticism, so I won’t get into that here. Instead, I want to talk about my perspective as an AI user and what I see, based on everything I’ve read, as the element that will determine the winners.
I’ve mentioned before that I primarily see the users of AI as the winners. Companies that successfully implement this technological shift will leave competitors who don’t far behind. Much like the users of steam engines won out over those trying to do the same work with horses and humans.
I don’t see the AI companies themselves as the big winners. People often calculate what AI can save companies, assuming the providers can capture that entire saving for themselves. I think that’s absolute nonsense. If I employ ten people today and, thanks to AI, we can do the work with six, I’m not going to pay my AI provider the equivalent of four annual salaries. If there’s nothing in it for me, why would I go through the trouble of changing a perfectly functional system? Just because it’s “hip” to use AI? Furthermore, am I going to give an AI company that much leverage over my business? Of course not—I’ll only do that if I significantly increase my own earnings as well.
The AI provider will therefore only be able to claim a portion of that productivity gain. How much depends on the competition. Right now, competition is fierce and systems are relatively interchangeable. One day Gemini is slightly better, the next it’s ChatGPT, Anthropic’s Claude, or Mistral AI. They are all in a race to improve their systems, constantly playing leapfrog.
I also read articles suggesting that AI is becoming a commodity, where the provider with the lowest costs has the advantage. That would mean operating on razor-thin margins.
So, is the hardware provider perhaps the winner? People are currently talking about an “NVIDIA tax.” NVIDIA is pulling in 75% margins because their chips are the best on the market, allowing them to quadruple their costs for companies busy building data centers.
I believe that NVIDIA, but also companies specializing in cooling or energy transformation, are currently enjoying golden times. These picks-and-shovels companies can rake in money by the bucketload right now. However, as soon as the great race is over, more competition will emerge and margins will come under pressure. That 75% margin at NVIDIA will likely settle back toward the roughly 45% the company saw when it was “merely” the best developer of graphics chips.
I’m cautious about the value of these companies for that reason. I can’t predict whether they will profit from these high margins for another two, five, or ten years, let alone what the growth will look like. The only thing clear to me is that too much was spent over the past year (and this year looks to be no different) just to stay in the race. At some point, that will lead to a correction.
A lot is also being written about energy or copper as key elements. Energy to run the machines and copper because a hyperscaler reportedly needs three to ten times as much copper as a normal data center. Should we follow that path? Given the sheer volume of articles on this, I assume it’s already baked into the prices. Moreover, it’s just as unpredictable as hardware usage once investments taper off.
By the way, consider the following as the thought process of an interested outsider, not deep technical knowledge. It’s an “outside view” from someone trying to strip things down to their simplest form.
When I ask myself who the winners of the whole AI saga will be (besides the companies implementing it), I see only one factor that is truly distinctive: data. AI models are only as strong as the data they are trained on. We saw this early on with ChatGPT, which pulled a lot of data from Reddit. You can find valuable information on Reddit, but there’s also a lot of nonsense. As they say: Garbage in, garbage out.
AI gets its data from the internet, and who is king there? Google. Their search engine is the best, giving them an index of the internet that is only rivaled by Microsoft’s Bing. Add the data from YouTube, and Alphabet looks like the big winner. The data Meta and X work with is often very personal—useful for ads, but less suitable for feeding a broad AI model.
If other AI companies want to play, they’ll have to find a way to collect data as efficiently as Google. If not, that part of the market is already taken. Below, I’ll dive a bit deeper into the value I place on Google today.
Of course, there are other ways to win as an AI company, such as entering specific niches. AI for doctors, lawyers, architects, or technicians, for example. This requires not just the model, but primarily the specific data to feed it. Countless applications are possible, but they are only ever as valuable as the underlying data.
I am, however, assuming AI as we know it today. If we’re talking about true superintelligence, then I believe the predictions of Terminator 😉 and the rest won’t really matter anymore.