South Korea's stock market suffered a sharp pullback on October 12, with the benchmark KOSPI closing down 3.65% at 6,384.25 points, as semiconductor names led the decline.
Samsung Electronics fell 4.39% to close at 250,500 won, while SK Hynix dropped 4.52% to 1,605,000 won. The selloff continued into the October 13 session, when program selling was briefly halted during intraday trading.
The decline marks a reversal from the June 22 peak of 9,385.59 points, with the KOSPI having fallen nearly 32% over roughly four months. While this downturn can partly be explained as a technical correction triggered by profit-taking at elevated levels, shifting focus from the broader capital boom to the micro-level nodes of the AI supply chain reveals that market pressure may extend beyond short-term capital withdrawal.
Three underlying concerns stand out: first, whether the continued expansion of AI infrastructure investment can smoothly translate into actual revenue and profit; second, whether rising prices for upstream equipment and components are reshaping return expectations across the chip supply chain; and third, whether the rapid revenue growth of leading frontier model labs can be sustained. Once earnings expectations begin to loosen, adjustments to elevated valuations could reinforce foreign capital outflows. Combined with margin trading and leveraged funds amplifying market volatility during downturns, this can easily trigger margin calls and forced deleveraging, reigniting liquidation fears among retail investors.
Since the release of ChatGPT 3.5, the total market capitalization of the global semiconductor and AI supply chain has surged 6.2 times, with the chip sector becoming the core winner of this AI supercycle.
Anxiety Over a Cyclical Peak
The trigger for the recent weakness in South Korean stocks and the memory sector came from a surprisingly strong earnings preview. On October 8, Samsung Electronics released its preliminary third-quarter results: consolidated sales of approximately 195 trillion won (estimated range of 194 trillion to 196 trillion won) and consolidated operating profit of about 10.74 trillion won (estimated range of 10.73 trillion to 10.75 trillion won), representing substantial year-over-year growth and landing at the upper end of analyst expectations. The strong performance was mainly attributed to robust AI-driven demand for HBM and DRAM. Samsung Electronics and SK Hynix jointly dominate the HBM and DRAM markets, fully benefiting from the structural dividends of global AI infrastructure investment.
Yet the market did not respond with the expected applause. Following the earnings preview, Samsung Electronics shares fell, dragging down the KOSPI index and semiconductor heavyweights including SK Hynix. The reason investors reacted cautiously to record figures, or chose to lock in profits at highs, lies in a reassessment of the sustainability of the memory boom cycle. Multiple institutional analyses noted that third-quarter profit growth may have approached a cyclical peak, with fourth-quarter sequential growth expected to slow to about 8%. As memory price increases flatten and new industry capacity gradually comes online, whether supply shortages can persist until 2028 has become a core market question. In plain terms, the fear has shifted from memory shortages to the possibility of ample supply. This reflects a form of anxiety about a cyclical peak, given that memory is a highly cyclical industry.
The "Messy" Bill
The "messy" revenue calculation methods of frontier labs also served as a key trigger for this pullback. According to the Financial Times and multiple financial media outlets, as of the end of September 2026, OpenAI's actual annualized revenue was close to $50 billion. Compared with the $70 billion figure that had circulated on Wall Street and in some media outlets in previous weeks, there exists a gap of approximately $20 billion. In other words, the market had overestimated the revenue-generating capacity of frontier labs. The key to this discrepancy lies in calculation methods 鈥?peers such as Anthropic record sales through cloud partners on a gross basis, while OpenAI only counts its own revenue, meaning the net amount after deducting channel revenue sharing. Some primary market investors adjusted OpenAI's data to create a direct comparison between the two companies, arriving at the $70 billion figure.
After the news emerged, on October 8 US Eastern Time, US semiconductor and cloud services stocks broadly declined: NVIDIA fell 2.94%; Oracle, which has computing partnership expectations with OpenAI, dropped more than 6% to close down 5.48%; Micron Technology fell 4.79%; SanDisk declined 4.90%; and CoreWeave dropped 7.77%. The Nasdaq Composite extended its decline to 1.25%. As a core link in the AI hardware chain, memory stocks experienced particularly pronounced sentiment contagion.
According to the Financial Times, as of the end of September this year, OpenAI's annualized revenue was approximately $50 billion, far below the widely expected $68 billion. Similar situations exist at other AI companies. On October 9, people familiar with the matter told Bloomberg that OpenAI still expects annualized revenue to reach or exceed $70 billion by the end of this year, driven primarily by enterprise business growth. This "news" partially alleviated short-term panic, but the market has raised higher demands for the comparability of AI revenue narratives. Analysts noted that this turbulence largely stems from information asymmetry. OpenAI's revenue data as a private company is already sufficient to affect the stock prices of public companies such as Microsoft, Oracle, and NVIDIA, demonstrating that it has become an important reference for global AI asset pricing.
Brokerage research shows that cloud infrastructure vendors have not yet shown signs of cutting capital expenditure. In other words, the brief selloff in secondary markets appears more like profit-taking at highs and emotional stampeding rather than a comprehensive fundamental collapse. Reports from institutions including Morgan Stanley and Goldman Sachs still project that combined capital expenditure by global core cloud giants will reach $1.2 trillion to $1.4 trillion by 2027. There is an inherent "sedimentation time lag" between the massive capital spending on AI hardware infrastructure and the monetization speed of upstream AI software applications. As some have summarized, the memory supercycle is still here, but the era of buying with eyes closed is over.
Valuation Discounts and Heavy Capital Expenditure
Despite short-term market sentiment pressure, strong demand for high-performance memory from AI infrastructure development continues to provide fundamental support for memory manufacturers. However, the three major players have different benefit pathways and potential challenges. Micron mainly benefits from HBM and server DRAM demand growth. SanDisk benefits from growing demand for enterprise SSDs in AI data centers. As an important player in the HBM market, SK Hynix, with its first-mover advantage and partnerships with leading customers, has strong demand visibility.
However, strong AI demand is only the starting point for growth; converting demand into sustained profits and cash flow is what matters. As earnings season approaches, memory manufacturers' capital expenditure, advanced process investments, and advanced packaging expansion plans will become important indicators for assessing upstream semiconductor equipment demand prospects. Bearish voices and market concerns are equally evident, focusing on extremely high capital expenditure pressure and upstream cost pressures 鈥?to compete for dominance in HBM3e and next-generation HBM4 markets, memory giants have become embroiled in the most intense equipment and capacity arms race in semiconductor history.
Samsung Electronics has committed to investing over 110 trillion won (approximately $82.6 billion) in capital expenditure and R&D in 2026, the highest annual budget in company history. SK Hynix is fully expanding its Cheongju M15X facility dedicated to HBM4 production and has spent $3.87 billion to establish an advanced packaging base in Indiana, USA. Industry data shows that to meet squeezed HBM demand, by the end of 2026, as much as 25% of global memory giants' DRAM wafer capacity will be forcibly locked for HBM production. Massive capital expenditure significantly increases future equipment depreciation and amortization costs. More critically, upstream supply chain cost pressures are further eroding profit expectations. Previously, lithography giant ASML explicitly notified that starting January 2027, it will uniformly raise prices for EUV and DUV lithography machine replacement parts supplied to Samsung Electronics and SK Hynix by 10%. Meanwhile, lithography machine price increase plans are also imminent.
Against this backdrop, SK Hynix's current dynamic price-to-earnings ratio is only 6.96 times, notably below the semiconductor industry average. On the surface, a valuation discount exists, but whether the discount is justified largely depends on whether the company can maintain the quality and sustainability of future profits under the dual pressures of high capital expenditure and rising operating expenses.
Retail Investors in Liquidation Fear
The risks of high-leverage trading among South Korean retail investors had already raised alarms earlier this summer. A 24-year-old Seoul university student, Lee Seung-ho, previously reported by Reuters, became the most vivid early warning and typical example of this leverage risk. According to Reuters, Lee accumulated 20 million won and participated in 5x leveraged trading through a trading app, with his account assets once surging to 300 million won (approximately 1.55 million yuan). But as market volatility intensified in May, margin calls and forced liquidations followed one after another, and he lost all profits and principal within just four weeks. In the report, he described his physical and mental state at the time as "literally unable to breathe."
However, the risks accumulated from high-leverage trading were not promptly resolved. Instead, with the growing popularity of single-stock leveraged products among retail investors, they persisted and spread to the present. On October 12, South Korean stocks suffered a brutal selloff, and this long-suppressed panic and anxiety erupted on social media. After the close, South Korean investors flooded social platform X with confusion and dissatisfaction. Some retail investors compared the South Korean stock market with US and Japanese market trends, questioning why the South Korean market continued to face pressure. Others pointed to continuous foreign capital outflows, arguing that the South Korean market failed to fully benefit when global markets rose but bore greater pressure during corrections.
These posts reflect retail investors' strong dissatisfaction with market performance, but behind the sentiment are more noteworthy capital flows. According to recent trading data, on October 12, foreign investors net sold nearly 2 trillion won worth of South Korean stocks in a single day, with cumulative net selling over the past month reaching 29.7 trillion won (approximately 156 billion yuan). If relevant statistics are confirmed, this means foreign capital withdrawal from South Korean stocks has become an important source of pressure in this round of market adjustment.
If foreign capital outflows explain why the market faces pressure, leveraged trading relates to why the decline may be further amplified. Controversy surrounding South Korean single-stock leveraged ETFs has been fermenting in investor communities recently. Some posts cited estimates suggesting that retail investors' asset losses in single-stock leveraged products reached 54 trillion won (approximately 283.5 billion yuan). This figure reflects the enormous destructive power of high-risk derivatives during market volatility, but it has also sparked considerable discussion about its statistical methodology and calculation samples.
Leveraged products significantly amplify stock price volatility. When the market declines, shrinking positions may trigger margin calls, forcing investors into passive liquidation and exacerbating downward pressure. Whether the daily rebalancing mechanism of single-stock leveraged ETFs amplifies selling pressure on underlying stocks on specific trading days requires case-by-case analysis combining product structure and actual capital flows, and cannot be generalized.
From Lee Seung-ho's liquidation to zero in July, to the exclamations on social media about the massive 54 trillion won loss, the extreme fragility of high-leverage trading during market reversals has been exposed. Therefore, this South Korean stock market adjustment reflects both investors' reassessment of AI industry growth prospects and the vulnerability of high-leverage trading when markets reverse 鈥?when stock prices no longer rise continuously, the profit advantage brought by leverage can quickly transform into risk. For some retail investors, what is cruel is not the temporary loss, but being forced to exit due to insufficient funds before the market rebounds. Former OpenAI researcher Leopold Aschenbrenner, hailed as Wall Street's AI stock prodigy at just 25, suffered a four-times leverage liquidation at the end of July, collapsing just before the tech stock rebound 鈥?a typical example.