After more than a month of sharp corrections in South Korean AI memory stocks, just how pessimistic has the market become?
A report from HSBC on August 3 does not continue the discussion of whether price-to-earnings ratios are cheap. Instead, it works backward to deduce what long-term earnings expectations are actually embedded in current market prices. The results show that, compared to traditional valuation metrics, the market is now betting that the AI super-cycle will be significantly shortened, with long-term profitability set to decline sharply.
The most extreme example comes from Samsung Electronics Co., Ltd..
The HSBC model indicates that the long-term earnings level implied by Samsung Electronics Co., Ltd.'s current stock price has fallen to approximately 0.8 times its 2024 earnings per share. This means the market not only believes AI is unlikely to deliver permanent earnings improvements but has also priced the company's cycle-transcending profitability back to pre-AI boom levels. In other words, the long-term AI premium in Samsung Electronics Co., Ltd.'s stock price has been almost completely erased.
HSBC: The real focus should not be on PE, but on how much the AI cycle has been discounted by the market
HSBC points out that a major misconception in investing in South Korean memory stocks currently is that investors still rely on traditional valuation metrics to judge whether stocks are "cheap."
For example, SK Hynix has a forward 12-month price-to-earnings ratio of only 4.3 times, near historical lows. However, its price-to-book ratio is already close to historical highs. These two indicators appear contradictory because the market does not recognize the sustainability of earnings at the current cyclical peak.
In other words, a low PE ratio does not necessarily indicate undervaluation; instead, it may suggest the market expects earnings to decline rapidly. Meanwhile, a high PB ratio implies the market believes the long-term asset value of the company has not yet significantly deteriorated. The essence of this divergence is that the market is repricing future earnings prospects, not current earnings themselves.
Therefore, the key variable is not next year's earnings performance but rather how the market assesses the duration of this AI super-cycle and how much profitability companies can retain after the cycle ends.
To answer this question, HSBC did not directly forecast profits. Instead, it worked backward to deduce the long-term earnings path implied by stock prices. The model uses consensus earnings expectations for the next three years, runs 100,000 Monte Carlo simulations, and identifies the 15-year earnings trajectory that matches the current stock price. This reveals the market's true expectations regarding the length of the earnings cycle, the speed of its decline, and the final long-term earnings midpoint.
The final conclusion shows that, compared to short-term earnings fluctuations, the market has truly downgraded the long-term AI earnings expectations for South Korean memory stocks.
Samsung Electronics: The market almost assumes AI will leave no lasting earnings
Within this framework, Samsung Electronics Co., Ltd. is the clearest case of concentrated pessimism.
Since its early June peak, the company's stock price has fallen about 25%, while the market-implied earnings cycle has shortened from approximately 3.5 years to 2.5 years. At the same time, the implied compound annual growth rate for earnings per share from years three to nine has dropped from approximately -15% to -35%, hitting the lowest level in history.
This means that, in the market's view, the high profits brought by AI are not only shorter in duration but also face a deeper decline after the cycle ends than in the past.
More notably, the long-term earnings midpoint is also noteworthy.
HSBC uses "implied trend EPS relative to 2024 levels" to measure the market's pricing of long-term AI demand. For Samsung Electronics Co., Ltd., this indicator has rapidly fallen from about 2 times to just 0.8 times.
If the earnings cycle length reflects how long the market believes the AI boom will last, the trend EPS reflects how much "permanent earnings improvement" the market believes AI will ultimately leave companies.
This means the market is gradually eliminating the long-term earnings premium from AI. It not only suggests that the current AI cycle is unsustainable but also indicates that even after the AI investment cycle ends, Samsung Electronics Co., Ltd.'s profitability is unlikely to exceed 2024 levels. In other words, the market has almost entirely priced out any long-term value premium for AI.
SK Hynix is similarly pessimistic, but TSMC still retains an AI premium
The adjustment for SK Hynix is even more severe.
Since its peak on June 25, the company's stock price has fallen 37%. The market-implied earnings cycle has plummeted from about 6 years to 2.7 years, with the EPS compound annual growth rate for years three to nine dropping to approximately -35%. Long-term trend earnings have also fallen from about 6 times 2024 levels to about 2 times. HSBC believes such pricing reflects "excessive pessimism" regarding the downturn cycle.
In contrast, TSMC shows a completely different pattern.
Although its stock price has also corrected, the market-implied earnings cycle remains at about 7.4 years, and long-term trend earnings still represent about 2.3 times 2024 levels. This suggests the market still believes AI can deliver earnings improvements that transcend the cycle, not just a brief boom.
The pricing difference between South Korean memory stocks and TSMC has become the most distinct feature of the current AI sector correction.
The most intense mechanical selling pressure may be nearing an end
Besides earnings expectations, HSBC also sees positive changes in fund flows.
Since the start of the year, foreign investors have net sold approximately $150 billion in Samsung Electronics Co., Ltd., SK Hynix, and TSMC. Of this, about $60 billion has occurred since June alone. Meanwhile, single-stock 2x leveraged ETFs, which had previously amplified volatility, are rapidly deleveraging. The total assets under management of related products have fallen from about $370 billion at the end of June to $120 billion, with their share of trading volume during days of sharp individual stock swings also declining noticeably.
HSBC believes this indicates that the most destructive mechanical selling pressure is easing.
For the market, the real question for the next phase is no longer whether AI demand exists, but whether current stock prices have compressed the AI cycle too pessimistically. If AI demand continues to show resilience, then the long-term earnings assumptions currently embedded in South Korean memory stocks may have room for re-correction.
Comments