The most disorienting feature of the past five weeks in technology and memory stocks is that the business results have been extraordinary while the prices have gone the opposite direction. Alphabet confirmed Google Cloud grew 82%. Micron reported $41.5 billion in quarterly revenue, up 346% year over year. SK Hynix is expected to deliver operating profit of approximately 64.1 trillion won today, a number that would exceed the company's entire full-year 2025 profit in a single quarter. The AI demand that drove these results has not been disproven by a single piece of fundamental data. Yet the Philadelphia Semiconductor Index fell roughly 25% from its June peak, the KOSPI lost nearly a quarter of its value in three weeks, and SKHY is trading around $130 today, down more than 30% from its Nasdaq debut high. Micron is around $810, down roughly 35% from its all-time high. The DRAM ETF is near $47, less than half its June peak.
To understand this, you have to separate two questions that most market commentary conflates: what is happening to the business, and what is happening to the price. They are not the same thing, and the mechanism connecting them is more complicated than most narratives suggest.
Why Prices Move Before Fundamentals Do
Goldman Sachs has characterized the July selloff in global technology stocks as primarily a momentum unwind rather than a fundamental repricing. The distinction matters enormously for how to interpret the data.
Momentum strategies are straightforward in design. They buy the assets that have performed best over the past 6 to 12 months and sell the assets that have performed worst. When AI hardware, semiconductors, and Korean equities simultaneously offered strong earnings revisions, strong price performance, and high index weights, they became the most heavily overlapping long position in global institutional portfolios. The same thesis held by too many funds simultaneously.
When a position becomes sufficiently crowded, the selling mechanics change. A marginally negative catalyst, not necessarily a fundamental one, can break the price trend. Once momentum strategies begin trimming exposure, quantitative long-short funds see volatility rising and reduce their own positions. Volatility-targeting funds, which mechanically size to maintain a target level of portfolio risk, cut exposure as realized volatility increases. Leveraged ETFs, which must reset to their target multiple at the end of each trading day, become forced sellers in declining markets. Each of these actors is behaving rationally at the individual level. Collectively, they create a nonlinear price decline that is entirely disconnected from any change in the underlying business.
Goldman's global momentum index data showed this precisely. US, European, emerging market, and global momentum indices all peaked simultaneously in late June and reversed sharply. Simultaneous reversal across geographies and asset classes is the signature of a single type of position being unwound, not nine different regional economies deteriorating on the same day.
The Korean Amplifier: Three Layers of Leverage
If the momentum unwind was the spark, Korea's market structure was the kindling. The KOSPI fell nearly 25% from its June 22 peak in approximately three weeks, with single-session declines large enough to trigger circuit breakers multiple times.
The first layer of amplification is concentration. Samsung Electronics and SK Hynix together represent approximately 59% of KOSPI market capitalization, a figure estimated to be at the 99.9th historical percentile. A national index that derives nearly 60% of its value from two companies that are simultaneously the most crowded global AI hardware longs transforms individual stock volatility directly into index volatility with no diversification buffer.
The second layer is retail participation. Korean individual investors account for approximately 46% of daily trading value and 71% of daily trading volume, multiples of US retail market share. Brokerage margin accounts can operate with minimum collateral ratios as low as 40%, implying theoretical leverage of 2.5 times. Margin balance reached 38 trillion won, a figure at the 99.9th historical percentile in absolute terms. Goldman Sachs subsequently observed that the margin balance has fallen from 38 trillion to approximately 33 trillion won, with investor cash deposits rising to 110 trillion won, suggesting the deleveraging is in progress but not complete.
The third layer is leveraged ETFs. Korean domestic and foreign leveraged ETF exposure to Korean equities totaled approximately $46.9 billion, representing roughly 1.5% of free-float market capitalization. The ratio of 2x long to 2x short products reached approximately 206 to 1 at peak. The mechanical problem with leveraged ETFs is not simply that losses are doubled. It is that they must rebalance daily. If a 2x ETF with 100 units of net asset value holds 200 units of underlying exposure and the underlying falls 10%, the net asset value drops to 80 but the stock exposure is only 180. To restore the 2x ratio, the fund must sell 20 units of exposure before the next open. The faster and harder the market falls, the larger the forced selling in the next session. Goldman's updated data showed that Korean leveraged ETF assets have fallen back to pre-single-stock-product levels, and forced liquidation amounts tracked by CICC reached 42.4 billion won, at the 96.9th historical percentile of prior episodes.
The 2000 Lesson: Demand Direction Is Not Stock Return
The most important analytical warning in the transcript above is the reference to 2000. Capital markets correctly identified the long-run demand trajectory for internet traffic. Many shareholders in the companies positioned to serve that demand still lost money, because supply expanded too fast, debt was too high, and competition compressed prices before companies could reach profitability.
That risk is different in the current AI infrastructure cycle, and it is worth being precise about how. The memory incumbents, SK Hynix, Micron, and Samsung, are operating under a contractual structure that did not exist in 2000. Micron has disclosed 16 strategic customer agreements covering approximately $100 billion in minimum revenue obligations at floor pricing that exceeds prior peak cycle margins. The risk of a 2000-style collapse in realized prices is bounded by the contractual floor in ways that internet equipment vendors in 2000 were not protected. But the warning still applies to the non-contracted portion of memory revenue, particularly in conventional DRAM for smartphones and PCs, where Chinese supply from CXMT's newly capitalized expansion program could eventually compress pricing.
The key distinction is between the near-term and the medium-term. Through 2027, the contracted revenue base, sold-out HBM supply, and accelerating hyperscaler capex confirmed by Alphabet's $205 billion full-year guidance create a defensible earnings floor. The medium-term risk, which the market is pricing in prematurely, is whether CXMT's conventional DRAM expansion, combined with the major players' own capacity additions, creates a supply overhang by 2028 or 2029.
SK Hynix Today: The Fundamental Anchor the Market Needs
SK Hynix reports Q2 results Wednesday morning Korean time, with results available before US markets open. The consensus from 14 brokerages projects Q2 revenue of approximately 84.1 trillion won, about $57.6 billion, and operating profit of approximately 64.1 trillion won, representing an operating margin of roughly 77%. If achieved, the Q2 operating profit alone would exceed SK Hynix's full-year 2025 operating profit of 47.2 trillion won.
The earnings result matters, but what matters more is management's commentary on three specific points. First, whether HBM pricing held at levels consistent with Micron's 84.9% Q3 gross margin benchmark. Second, whether contracted volume commitments have expanded as the industry shifts toward multi-year agreements. Third, whether management maintains language about supply tightness extending through 2027.
SKHY is currently trading around $130, down more than 30% from its Nasdaq debut high of $170, and MU is around $810, down roughly 35% from its $1,255 all-time high. These declines reflect the momentum unwind and the Korean market structure, not a revision to the earnings trajectory. SK Hynix's 37 covering analysts maintain a consensus target price approximately 84% above the current Korean listing price. If the Q2 numbers confirm the business thesis, the gap between institutional models and market prices becomes analytically untenable regardless of what the momentum factor or Korean leveraged ETFs do next.
The most useful observation from the Goldman and CICC analysis is this: many of the sellers in the past five weeks have not reassessed 2028 HBM demand. They did not have enough cash to wait until 2028. That is a statement about portfolio financing, not about whether the AI infrastructure buildout is intact. Understanding which of those two forces is driving a price is the entire exercise.
Explore more exclusive insights at nextfin.ai.
