Summary of Key Points
Recently, global AI-related assets (semiconductors, storage, cloud computing, etc.) have experienced a sharp decline, with the Philadelphia Semiconductor Index entering a technical bear market. Some individual stocks have seen returns of over 50%, and leveraged funds have suffered even more severe losses. Strangely enough, the fundamentals of the AI industry chain are actually quite strong: TSMC has raised its revenue forecasts, and SK Hynix has seen its profits surge by 500%. The main cause of the decline is the "mass liquidation of momentum trading funds," coupled with concerns about three key issues regarding the future of AI—sustainability of capital investment, new commercialization scenarios, and potential for further price increases in storage.
Detailed Analysis
1. Why are stocks falling despite good performance? It's due to the mass departure of momentum trading funds
You can think of "momentum trading" as a mechanical version of "chasing gains and cutting losses": algorithms automatically buy the stocks that have risen the most (such as AI chips and storage) and sell those that have performed poorly (such as consumer and pharmaceutical stocks). Over the past year, the AI sector has skyrocketed, and these trading funds have continued to buy more, creating a cycle of "rise → buy → even greater rise." However, this approach's fatal weakness is that it collapses quickly when market trends change: once AI stocks start to fall, the algorithms automatically execute stop-loss orders, leading to a vicious cycle of "fall → sell → further fall." This was precisely what happened this time. Data from Morgan Stanley shows that the momentum factor in TMT (Technology, Media, Communications) sectors plummeted by 40%, both in terms of speed and depth, making it one of the most extreme episodes since the dot-com bubble burst in 2000. In short, it's not that AI has become ineffective; rather, it's the momentum traders who have collectively panicked and sold their positions.
2. The "fast and slow clocks" within the AI supercycle: The market is driven by the slowest one
The AI industry consists of three different timelines with vastly varying speeds:
- Model clock: Updates occur monthly (with new versions like GPT-4 and Claude 3).
- Capital clock: Valuations are reassessed quarterly by institutions.
- Hardware clock: Construction of facilities (such as chip factories and data centers) takes 1–2 years.
The market will ultimately be aligned with the slowest "hardware clock." The current issue is that the growth rate of hardware development may be slowing down. For example, cloud companies like Microsoft and Amazon had high capital expenditure growth rates in 2026 (76%), but these rates dropped to 25% in 2027 and will further decline to 6% in 2028. Even though the total investment remains substantial, the slowdown in growth rate is causing concerns about a decrease in upstream orders. Companies like NVIDIA and optical module manufacturers rely on revenue generated during the construction phase, so a slowdown in this area will lead to lower valuations.
3. Does capital expenditure need to slow down? Tech giants can't keep spending recklessly
AI requires significant investment in data centers and chips, but these companies cannot afford to spend indefinitely; they must demonstrate that their investments are profitable. Currently, only a few leading firms like OpenAI and Anthropic are actually generating substantial profits. Most cloud companies have invested billions of dollars without yet converting them into cash flows. Meta's recent decision to rent out excess computing power indicates that they need to monetize their investments. If the growth rate of capital expenditure peaks, orders from upstream chip, storage, and optical module manufacturers will also slow down. This is one of the reasons for the market's anxiety: the AI industry's "expensive phase" may be transitioning into a "return validation phase."
4. What else can generate revenue beyond just programming in AI? Without new applications, growth could stagnate
The most profitable area for AI in the past year has been programming tools (such as Cursor and Claude Code), where developers are willing to pay for their use, leading to rapid revenue growth. However, market forecasts suggest that the penetration rate of these tools will soon reach 30% (with a global market value of $2 trillion by next year, with OpenAI and Anthropic accounting for 30%). If there are no new "blockbuster applications" (such as widespread adoption in education, healthcare, or office productivity), overall AI revenue growth will slow down. The market is already worried about the next source of profit and may see stock prices fall if no clear alternatives emerge.
5. Can the storage stock "myth" continue? The room for price increases is limited
Storage companies (such as SK Hynix and Samsung) have been major beneficiaries of the AI boom, with profits reaching absurd levels (equivalent to 1.4 times South Korea's government debt over the next three years). The market once portrayed storage as a growth stock due to AI, driving valuations from single digits to double digits and causing stock prices to soar. However, this narrative is now in question: storage already accounts for a significant portion of data center costs, making it difficult to see further price increases. Additionally, downstream companies (such as PC and smartphone manufacturers) cannot afford to pass on the increased costs to consumers and may reduce their orders. As a result, the market is re-evaluating whether storage remains a cyclical asset; if so, the current high valuations are unlikely to be sustained.
Conclusion
While the long-term transformation driven by AI is far from over, short-term stock prices have been significantly affected by factors such as momentum trading and concerns about three key issues. For AI assets to recover, the market needs to see new developments—such as stable capital expenditure growth, the emergence of new commercialization scenarios, or the validation of storage's growth potential. Otherwise, volatility is likely to continue.