Han Dieperink
Dieperink.png

The training landscape of artificial intelligence has taken an unexpected turn—one that challenges classical economic principles. More data leads to better performance, with no signs of saturation.

Where we would normally expect the law of diminishing returns or marginal utility to kick in, AI demonstrates the opposite: more data consistently results in better outcomes. Even more than the arrival of the PC, the internet, or the iPhone, the potential of artificial intelligence remains severely underestimated.

The DeepSeek revolution

Earlier this year, the launch of DeepSeek caused a seismic shift in the AI world. This model drastically lowered the barriers to training AI, making the technology accessible to a much wider audience. Marc Andreessen aptly compared this moment to the launch of Sputnik in 1957. Just as that first satellite sparked a space race between the United States and the Soviet Union, DeepSeek has triggered a new AI race—this time with China as America’s primary rival, rather than Russia.

The democratization of AI training has led to an explosion of Chinese AI models, which now outnumber their American counterparts. This phenomenon illustrates a paradox we often see with technological efficiency improvements: just as more fuel-efficient cars result in more miles driven, and water-saving technologies lead to higher water consumption, cheaper AI training has driven demand so high that the scale of investment has surged exponentially. This dynamic also explains why attempts to reduce environmental impact through efficiency gains often fail. With AI, the effect is even more pronounced: every drop in the cost of entry leads to exponential growth in both usage and applications.

A unique economic position

What sets artificial intelligence apart from other technologies is the nearly infinite demand for intelligence. Normally, demand for a product or service eventually plateaus.

With AI, however, each improvement in efficiency not only lowers the cost per unit of intelligence but also unlocks new use cases that were previously unimaginable. In addition, every advancement in AI models expands their capabilities, continually raising the ceiling on what is possible.

Shifting market dynamics

Against this backdrop, it’s noteworthy that the stock prices of the “Magnificent Seven” (Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, and Tesla) have come under pressure. The reason is not declining demand for AI services—far from it—but rather the erosion of their perceived monopoly on enabling artificial intelligence. The democratization of AI training means these giants are losing their exclusive grip on the technology.

At the same time, this shift opens new doors for the Magnificent Seven. These companies are likely to earn significantly more from developing and monetizing AI applications than from training the underlying models. After all, demand for intelligent applications far exceeds that for training infrastructure alone.

The last AI monopoly

Amid these shifting power dynamics, one company remains untouched: ASML. As the sole producer of advanced lithography machines for AI chip production, the Dutch company holds an enviable monopoly. Former competitors like Nikon and Canon are no longer relevant players in this race.

ASML’s dominance in EUV machines—capable of producing chips with seven-, five-, and soon three-nanometer technology—means that the growth of AI is currently constrained by ASML’s production capacity. In other words, ASML provides the essential tools for today’s AI gold rush.

Geopolitical developments are further fueling demand for ASML’s technology. Whereas previously a handful of Asian customers accounted for the bulk of demand, the trend toward de-globalization is now driving new chip fabrication plants in the United States and Europe. This is resulting in even greater demand for ASML’s cutting-edge machines.

Just the beginning of the AI cycle

The cheaper AI training becomes, the greater the demand for AI applications. The greater that demand, the greater the need for AI chips. And the greater the need for chips, the stronger ASML’s position as the undisputed supplier of critical production technology. In this self-reinforcing cycle, there is no diminishing return or marginal utility in sight. In a world where data hunger only grows, the supplier of the most advanced chip-making machines may hold the most valuable position of all.

That said, don’t count out the Magnificent Seven. Their stocks have declined sharply this year, a sharp contrast to recent years. However, profits continue to rise, while both stock prices and interest rates have fallen. Valuation only really matters when prices fall, and far less so when prices are rising—especially in the tech sector. Additionally, most of these companies earn the bulk of their revenue outside the United States, meaning they benefit from a weaker dollar. And historically, a bull market rarely changes leadership mid-cycle. It takes more than a mere correction for that to happen.

Han Dieperink is Chief Investment Officer at Auréus Wealth Management. He previously served as Chief Investment Officer at Rabobank and Schretlen & Co.

Author(s)
Categories
Access
Members
Article type
Column
FD Article
No