To The Moon

Last night, the U.S. chip stocks plummeted: Nvidia fell 5%, losing its crown as the world’s most valuable company by market capitalization. SK hynix plunged more than 7.4%, while ASML declined over 5.8%.

The panic extended into today. SK hynix and Samsung Electronics both fell more than 13%, while South Korea’s KOSPI index triggered circuit breakers multiple times. Meanwhile, NAND flash memory leader Kioxia plunged 17%, crashing nearly 60% from its recent all-time high.

In Hong Kong stocks, AI large-model leader Zhipu dropped 16% today, falling from its record high of HK$2,980 to HK$1,070 in just over a month, representing a decline of approximately 64%. $Z.AI(02513)$

This sharp decline appears difficult to explain as a normal market correction. So what exactly are markets worried about regarding these semiconductor stocks?

News reports have mainly attributed the decline to two factors: Nvidia’s announcement yesterday of hundreds of billions of dollars in revolving financing guarantees, and the upcoming mass production of China’s DUV lithography machines. Investors are concerned that Chinese chips could adversely impact Western companies. $NVIDIA(NVDA)$

However, in my view, these explanations do not fully address the fundamental issue. I would like to share my own perspective.

Firstly, the fundamental driver of this chip stock crash is lofty valuations.

In my investment framework, even if a company’s fundamentals remain flawless, when its valuation reaches historical highs, there are usually only two possible outcomes:

1.The stock enters a prolonged period of sideways consolidation, allowing earnings growth to gradually justify the valuation;

2.The stock experiences a significant correction, bringing the valuation back to a more reasonable level.

From a fundamental perspective, semiconductor giants such as Taiwan Semiconductor Manufacturing Company and ASML have recently released their second-quarter earnings reports. Both companies raised their full-year guidance, indicating that semiconductor stocks' fundamentals haven’t deteriorated at all and continue to improve. $ASML Holding NV(ASML)$

However, the market remained unconvinced. The main concern is that valuations have become excessively high. Take TSMC as an example. Its current price-to-book ratio is around 10 times, among the highest valuation levels in its history. $Taiwan Semiconductor Manufacturing(TSM)$

Someone may point out that memory chip companies are trading at a P/E ratio of only around 10–20 times. However, those who value semiconductor companies purely based on P/E ratios often overlook the industry's cyclical nature and may only have started following the sector after the AI boom.

The cyclicality of the semiconductor has been clearly demonstrated throughout history. After the rise of AI, demand appears extremely strong and the memory chip cycle seems to have weakened. However, the reality is that demand surged so suddenly that supply could not keep pace, causing memory chip prices to soar. SK hynix has achieved operating profit margins above 70%, even surpassing NVIDIA Corporation in profitability.

Let's think about it carefully. Is this level of profitability sustainable? From a long-term perspective, the answer is absolutely no. If such profit levels continued indefinitely, a disproportionate share of the AI industry’s profits would end up being captured by South Korean memory chip manufacturers.

Looking at the broader value chain, it is abnormal that upstream AI model developers are struggling to generate profit, while suppliers of chip components are earning extraordinarily high profit. Once AI demand slows down a bit and memory chip capacity expands, $SK hynix(SKHY)$ will once again be viewed as a cyclical stock.

Therefore, whether it is Taiwan Semiconductor Manufacturing Company or SK hynix, their valuations have become very high — and there are no exceptions.

With this high-valuation risk in place, the market is prone to panic at the slightest sign of trouble, because there are simply too many profit-taking positions, and no one is willing to bear the risk of drawdowns at such stretched valuations.

Beyond high valuation risk, there are two other major factors at play. The first factor is that major technology companies, including Google, Amazon, Microsoft and Meta, are expected to increase their capital expenditure next year. However, the growth rate is likely to slow down significantly compared with 2026.

From what I recall, tech giants’ capital expenditures grew by roughly 80% to 100% year over year this year, while growth rate is projected to fall back to around 20%.

For growth stocks, a slowdown in growth momentum can be extremely damaging. Even if profits continue to rise in 2027, a lower growth rate combined with historically high valuations could create significant pressure on stock prices.

The main reason constraining further aggressive capital expenditure growth is that their free cash flow has already turned negative. They can no longer fully fund these investments through operating cash flow alone. As a result, major technology companies have increasingly relied on debt financing, and Google even issued equity to raise capital.

Debt-funded expansion carries significant risks. On one hand, companies must bear substantial interest expenses, with U.S. inflation rebounding this year and the U.S.-Iran war pushing up financing costs. On the other hand, massive investments in data centers go mostly toward buying GPUs, which have extremely rapid technology cycles. The resulting depreciation expenses are enormous and may create significant pressure on the profitability of technology giants.

Overall, AI spending of tech giants is not an unlimited source of capital, it is constrained by many factors.

Beyond tech giants, other companies’ capital expenditures account for roughly half of Nvidia’s revenue. So the key question is: if major technology companies face spending pressure, can smaller companies step in and fill the gap?

In theory, the answer is yes. However, the AI model industry is also becoming increasingly competitive, with price wars starting to emerge. The emergence of the Kimi K3 model is a good example. No one expected that Chinese AI companies could develop models comparable to those of Anthropic, despite having limited access to the most advanced chips. Even more surprising, these models are offered at roughly half the price of Western companies' offerings and are even open source.

Lower AI model costs will certainly stimulate broader adoption. However, in recent months, many companies have also tightened their token quotas as they reassess whether the productivity gains generated by employees using AI can truly justify the additional costs.

As a result, the competitive landscape for large AI models remains unstable. At the same time, U.S. semiconductor giants have become deeply tied to OpenAI and Anthropic. If they lose ground in the competitive race, the risks associated with revolving financing could become significant.

Adding to these concerns, just one month ago, investors in South Korea and Taiwan were aggressively bidding up semiconductor stocks. From a market sentiment perspective, the enthusiasm appears to have reached irrational levels.

Therefore, with this sharp decline in semiconductor stocks, investors should remain cautious about the possibility of a bubble unwinding.