1. "Record-High Earnings, so Why Is the Stock Price Falling?" — The TSMC Paradox

This is the TSMC paradox. TSMC, the world's largest foundry company, announced record-high earnings for the second quarter of 2026, with revenue of 1.27 trillion Taiwan dollars (approximately 58 trillion won) and a net profit of 706.6 billion Taiwan dollars (approximately 32.4 trillion won). This net profit represents a whopping 77% increase compared to the same period last year, significantly exceeding market expectations.

But what happened to its stock price? It fell 2.3% in the U.S. market on the day of the earnings announcement. Reuters analyzed this, stating it means "that expectations have risen that much." In other words, it was not because the performance was poor, but because excessively high growth expectations had already been priced into the stock. This serves as a warning sign that simply delivering an "earnings surprise" may not be enough to satisfy the market in upcoming earnings announcements from major AI semiconductor companies like Nvidia.

2. "A $1 Trillion AI Investment, Will It Actually Make Money?" — Fear of Return on Investment

The real issue is not demand, but the speed of return on investment. TSMC raised its 2026 capital expenditure plan from the previous 52 billion to 56 billion dollars (approximately 75 trillion won) to 60 billion to 64 billion dollars (approximately 92 trillion won). This means they will expand production facilities further due to strong demand for AI semiconductors. However, investors have begun to ask: "How quickly will the massive funds poured into AI data centers translate into actual revenue and profit?"

Reuters summarized this adjustment as "questions about the sustainability of a nearly $1 trillion spending boom in the near future." Meta projected its 2026 capital expenditure at 125 billion to 145 billion dollars (approximately 181 trillion won), while Google's parent company Alphabet set its target at 180 billion to 190 billion dollars (approximately 275 trillion won). Amazon CEO Andy Jassy stated that the investment of approximately 200 billion dollars (approximately 290 trillion won) in 2026 is "not speculative, but based on customer commitments." Although the numbers are massive, the market remains uncertain about when these investments will turn into profits.

3. "A Single Shot of K3 Fired by China" — The Shock of Moonshot AI

The open-source model "Kimi K3" released by Chinese AI startup Moonshot AI pulled the trigger. Moonshot stated that this model is close to cutting-edge closed models such as OpenAI's ChatGPT and Anthropic's Claude Fable 5.

The market reacted sensitively for two reasons. First, the revenue models of U.S. AI companies, which have charged subscription fees for closed models, could be eroded by free open-source alternatives. Second, in this case, the large-scale infrastructure investment plans of AI companies would become complicated, ultimately shaking the semiconductor demand outlook. This is the same pattern as when the release of China's DeepSeek model in January 2025 caused the New York stock market to plunge. However, at that time, the U.S. stock market recovered quickly, and Big Tech's AI infrastructure investments continued. Open-source models are alternatives that allow companies to save significantly on infrastructure costs and build their own customized AI because the source code is publicly available.

4. "Will HBM Demand Also Decrease as Open Models Increase?" — The Truth of the Structural Bottleneck

The proliferation of open models does not necessarily mean a decrease in memory demand. On one hand, if cheaper and more efficient open models spread, the excess profits and aggressive capital expenditures of expensive closed-model operators could face pressure. In that case, the market's tolerance for "how much can be invested in AI infrastructure" may lower. On the other hand, as models become cheaper and more widely adopted, total usage could increase, potentially expanding the demand for HBM instead.

Recent academic studies point out that Large Language Model (LLM) inference still strongly exhibits memory-bound characteristics, where the saturation of DRAM/HBM bandwidth, rather than the computational power of GPUs, becomes the bottleneck. In other words, HBM is not a demand that will disappear, but is closer to an essential yet expensive bottleneck resource.

5. "38% Evaporated in a Month" — What Happened to SK Hynix

Why did SK Hynix shake more sensitively? SK Hynix, which hit an intraday record high of 2,987,000 won on June 25, closed at 1,842,000 won on July 16. It plunged 38.3% in less than a month. Samsung Electronics fell 31.9% from its intraday high on June 19 to July 16, meaning SK Hynix suffered a larger drop. The reason is clear. In this cycle, SK Hynix has been priced not as a simple DRAM manufacturer, but as the representative beneficiary of the HBM premium.

The problem is that while HBM is a powerful growth story, it is also the area where expectations have been most heavily priced in advance. Even if demand does not deteriorate, the moment the market judges that the "speed of profit growth is slower than the speed of expectation pricing," the stock price correction can widen.

6. "Bought Samsung, Sold Hynix" — The Choice of the Experts

In the plunging market, the responses of high-yield investors were mixed. According to Mirae Asset Securities, as of 11:00 AM on July 16, the top net purchase among customers in the top 1% of investment returns over the past month was Alteogen. Samsung Electronics and Samsung Electronics Preferred Stock ranked second and third, respectively. On the other hand, the top net sale was SK Hynix. Their responses completely diverged even when dealing with the same large-cap semiconductor stocks. As SK Hynix plummeted 11.53% in a single day after surging 8.83% the previous day, it is interpreted that moves to realize short-term profits and selling to avoid further declines occurred simultaneously. Conversely, funds viewing the sharp drop as a buying opportunity flowed into Samsung Electronics and Samsung Electronics Preferred Stock. An industry insider explained, "Even among the same semiconductor stocks, investors' choices seem to have diverged depending on the recent stock price gains and valuation burdens."

7. "Why Does the KOSPI Shake More Violently?" — The Trap of Semiconductor Concentration

The South Korean stock market is structurally more vulnerable. According to foreign media reports such as Bloomberg, the KOSPI, where Samsung Electronics and SK Hynix hold an absolute weight, is structured such that the entire index shakes heavily due to the sharp fluctuations of these two stocks. Under this structure, shifts in global memory and AI semiconductor sentiment easily amplify into volatility for the entire Korean market. In fact, on July 16, the KOSPI plummeted 6.37% to close at 6820.60. Although it had recovered the 7000 level by surging 6.24% the previous day, it fell back to the 6800 level in just one day. During the same period, the Nasdaq index had rebounded after falling about 7% from its peak. The OECD pointed out that South Korea's value chain is highly exposed to external shocks, and its dependence on inputs from non-OECD countries is among the highest in the OECD.

8. "Is Chey Tae-won's Advice Right?" — The Crossroads of Long-term vs. Short-term

Shall we look closely at Chairman Chey Tae-won's advice again? At the Jeju Summer Forum, he said, "AI is still like a four-year-old child, but to become an adult, memory must be used. That demand is bound to grow exponentially." He added, "The fact that stock prices suddenly rose tenfold is also due to this phenomenon," explaining, "When the outlook improves, it goes up, and when it seems a bit off, it drops sharply. Because it rose too quickly, there is also a process of adapting it to reality."

From a long-term perspective, this is a valid assessment. Indeed, TSMC CEO C.C. Wei stated during the earnings call that "AI-related demand is still extremely strong," and the World Semiconductor Trade Statistics (WSTS) projected that the global memory sector would grow by approximately 250% year-on-year in 2026, reaching over 800 billion dollars (approximately 1,160 trillion won). In the short term, however, stock prices can move more heavily as a function of expectations rather than actual performance.

9. "Where Did the Retail Investors Go?" — The Escape from the Domestic Market

Retail investors, exhausted by extreme volatility, are leaving the domestic stock market. According to data from the Korea Securities Depository, from June to July 16, the net purchase amount of U.S. stocks reached 1.79509 billion dollars (approximately 2.6 trillion won). In contrast, during April and May, the selling amount was larger, resulting in a net selling amount of 1.4087 billion dollars (approximately 2 trillion won). This means that "Seohak Ants" (Korean retail investors investing in U.S. stocks) who had returned to the domestic market have headed right back to the U.S. The KOSPI fell 25.16% from its closing high on June 2 (9114.55) to July 16. Furthermore, daily fluctuations ranging from around 5% to as high as the 9% range are dampening investor sentiment. On the other hand, the Nasdaq index has rebounded after falling about 7% from the peak it recorded on June 2. In particular, thanks to the resilience of the existing "Magnificent Seven," such as Apple rising 17% so far this year, it is holding up despite the impact of sharp fluctuations in the semiconductor index.

10. "What Should We Watch Now?" — Future Checkpoints

Here are the things to watch moving forward. First, the real signal of a demand slowdown would be a downward revision of guidance by key supply chain companies like TSMC, Micron, and SK Hynix, or a reduction in capital expenditures by hyperscalers. For now, TSMC, Meta, Alphabet, and Amazon all officially maintain an expansionary stance. Second, we must observe whether open models like "Kimi K3" are not just one-off events but are continuing to erode the pricing power of leading U.S. models. Third, we need to see if HBM's status as a bottleneck is maintained. According to academic data, LLM inference still strongly exhibits memory-bound characteristics. As long as this assessment holds, HBM's strategic position is highly likely to remain unshaken. Fourth, we must watch whether the concentration of the Korean market eases. Because the weight of Samsung Electronics and SK Hynix in the Korean large-cap index is extremely high, the possibility that changes in foreign semiconductor sentiment will be amplified in the Korean market remains high.


Deep Dive Q&A
Q1. Even though TSMC's earnings are historic, what is the fundamental reason for the stock price correction centered on tech stocks?

A1. It is due to the market's "pricing-in" mechanism. Stock prices grow based on future expectations rather than current performance, and expectations for the semiconductor supercycle and the AI boom were priced into stock prices too quickly. Combined with the fear of return on investment (ROI doubts) regarding whether the massive capital expenditures of Big Tech (Meta, Alphabet, etc.) can be recovered with clear "profitability" in the short term, a volatile period has occurred to adjust expectations to reality.

Q2. Does the emergence of open-source AI models (such as "Kimi K3") decrease SK Hynix's HBM demand in the long run?

A2. While it may put downward pressure on short-term investment sentiment, it is unlikely to lead to a structural collapse in HBM demand. If the monopoly excess profits of closed Big Tech companies decrease due to the spread of open models, there is room for the pace of infrastructure investment to be adjusted. However, technically, Large Language Model (LLM) inference suffers from a "memory-bound" problem, where it is limited by the speed of data transfer with memory rather than the computational power of the processor. As open-source models become widely distributed and total AI usage increases, the need for high-bandwidth memory like HBM may actually expand.

Q3. Should we refer to Chairman Chey Tae-won's advice to "just hold on" or the "sell SK Hynix" strategy of the top-performing experts?

A3. The two perspectives should be understood as a difference in investment time horizons. Chairman Chey Tae-won's remarks represent a "long-term position" from a fundamental perspective, suggesting that memory demand will trend upward until the AI industry reaches maturity. On the other hand, the top 1% of investors chose "tactical asset allocation," avoiding short-term valuation burdens and high volatility while rebalancing their funds into alternatives that have risen less (such as Samsung Electronics). Responses should differ depending on whether an individual's investment style is long-term value investing or momentum trading.