Han Dieperink: The Rotation Correction
This column was originally written in Dutch. This is an English translation.
By Han Dieperink, written in a personal capacity
Following the strong rally in IT stocks, investors are now looking for opportunities in other sectors. Industrials, financials and even utilities are benefiting from this shift. Some interpret this rotation as the beginning of the end for technology, but that is a misconception. At the same time, sceptics have introduced a new argument: it is not stock prices that are in a bubble, but corporate earnings themselves. This so-called "earnings bubble" deserves a critical examination.
A Correction in Chip Stocks Is Part of the Cycle
In July, the Philadelphia Semiconductor Index (SOX) entered official correction territory, falling more than 10% from its peak. That may sound dramatic, but it followed a gain of nearly 128% over the previous twelve months. After such a rally, a pause is both normal and healthy. Remarkably, the entire AI boom has been driven almost entirely by earnings growth rather than higher valuations. In fact, the forward price-to-earnings ratio has declined in recent years, even as profits have risen sharply. The market is behaving more cautiously than euphorically.
Korea's Leverage Amplifies the Sell-Off
Why, then, has the correction felt so severe? Much of the answer lies in South Korea. Korean retail investors have accumulated a record KRW 38 trillion (around €22 billion) in margin loans, heavily concentrated in large KOSPI stocks, particularly memory chip makers Samsung and SK Hynix. On top of that, many investors hold leveraged ETFs linked to individual stocks. As share prices declined, margin calls triggered forced selling, accelerating the downturn. The KOSPI has now fallen roughly 27% from its June peak, despite Korean exports surging nearly 71% in June—the fastest pace in almost fifty years—driven largely by demand for memory chips. Forced sellers do not respond to fundamentals; they respond to debt. As a result, Korean large-cap chip stocks now trade at only about seven times forward earnings, while sector valuations have roughly halved since 2023.
The End of Subsidies Marks the Beginning of Profits
Sceptics argue that the era of subsidised AI is coming to an end. For years, AI developers offered their services well below cost. Increasingly, however, providers are switching to usage-based pricing. While some view this as negative, the opposite may be true. Investors have spent years complaining that AI generated enormous costs without meaningful revenue. Now, those business models are finally beginning to deliver.
At the same time, Jevons' Paradox remains highly relevant. When technological advances make the use of a resource more efficient, total consumption often increases rather than decreases. William Stanley Jevons observed this in the nineteenth century, when more efficient steam engines led to greater—not lower—coal consumption. The same principle applies to AI today. As computing power becomes cheaper, demand for chips and electricity is likely to accelerate as businesses deploy AI across an ever-growing range of applications.
AI Adoption Is Accelerating, Including in China
Another concern is that inexpensive Chinese open-source AI models will turn artificial intelligence into a utility, much like electricity. For investors across the broader AI value chain, however, this is more opportunity than threat. Electricity also became a commodity, yet demand for power plants, electricity grids and electrical equipment continued to expand. Chinese AI models still rely on computing power provided by data centres filled with advanced processors and memory chips.
The criticism that large technology companies are merely generating revenue from one another is equally unconvincing. Behind the hyperscalers stand millions of businesses and consumers adopting AI at remarkable speed. Global token consumption—a direct measure of actual AI usage—has increased from just over 11 billion tokens per week in 2024 to more than 15 trillion today, an increase of nearly 1,400-fold. The rapid revenue growth reported by hyperscalers simply reflects this extraordinary expansion in real-world AI adoption.
Supply Will Remain Tight for Years
Anyone who believes the semiconductor industry is weakening should look at company results. Micron delivered one of the strongest quarters in its history, with gross margins of around 85%, and expects demand for memory chips to exceed supply at least through 2028. SK Hynix has suggested that 2027 could become the tightest supply year the industry has ever experienced. ASML has already raised its revenue guidance twice this year, while its memory-related business expanded by more than 53% during the first half of 2026.
Industry dynamics are also changing. Manufacturers are increasingly signing five-year supply agreements that include advance payments and agreed pricing mechanisms. The semiconductor cycle has not disappeared, but it is becoming longer and more predictable.
History provides useful perspective. During the internet revolution, global semiconductor sales increased fivefold in real terms between 1992 and 2000. Since the beginning of 2025, the industry has grown by roughly 80%. Compared with the internet era, today's AI-driven semiconductor expansion still appears to have a long runway ahead.
Embrace Broader Market Leadership, but Don't Abandon AI
The current market rotation should be welcomed. The correction in semiconductor stocks has been driven primarily by profit-taking and leveraged positions in South Korea rather than weakening demand. There is no evidence of an earnings bubble. Corporate balance sheets remain healthy, and investors continue to resist paying higher valuation multiples despite robust profit growth.
Meanwhile, AI business models are beginning to generate meaningful revenues, demand for computing power is expected to exceed supply for years to come, and the hyperscalers remain the most attractive gateway to long-term AI growth. Investors should use the current market rotation to broaden portfolio exposure—but remain invested in the structural growth story of artificial intelligence.
