One of the most crowded AI trading sectors is undergoing a sharp correction.
During Thursday's US trading session, the memory chip sector experienced a collective heavy sell-off. The Philadelphia Semiconductor Index (SOX), which tracks the overall performance of chip stocks, fell by more than 4% at one point, while the memory chip index extended its losses to 7%.
On an individual stock basis, SanDisk (SDNK) fell more than 10% intraday, Western Digital (WDC) dropped nearly 9%, SK hynix ADR (SKHY) declined nearly 9%, Seagate Technology PLC (STX) fell over 8%, and Micron Technology (MU) was down more than 6% at one point. Over the last two trading sessions, its cumulative loss has exceeded 10%, representing a pullback of over 30% from its intraday high on June 22nd.
This round of adjustment signifies that the memory chip sector, which has been one of the most sought-after beneficiaries of the AI boom this year, is now facing a market repricing. As South Korean regulators tighten rules on single-stock leveraged ETFs, leading stocks like Micron have seen massive prior gains, and investors begin to reassess the supply-demand dynamics for High Bandwidth Memory (HBM), the sustainability of AI capital expenditure, and potential inflection points in the industry cycle, capital is clearly reducing its risk appetite for the high-valuation semiconductor sector.
South Korean Regulatory Crackdown on Leveraged ETFs Serves as Immediate Catalyst
The selling pressure initially began in Asian markets.
On July 16th, South Korea's Financial Services Commission (FSC) officially announced measures to tighten regulations on single-stock leveraged ETFs. These include raising the minimum margin requirement from 10 million won to 30 million won, stipulating that only cash can be used for margin, limiting single-stock leveraged trades to a maximum purchase of 20 shares per transaction, and prohibiting the launch of new single-stock leveraged products.
Following the announcement, South Korean memory chip stocks were the first to be hit hard, with SK hynix plummeting over 11% at one point and Samsung Electronics falling more than 8%. The selling pressure then rapidly spread to European and US markets.
Analysis suggests that South Korea's move precisely targets one of the most active sources of leveraged funding in memory stock trading.
Over the past year, driven by demand for AI servers and HBM, memory leaders like Micron Technology and SK hynix have become key targets for South Korean retail investors and leveraged ETF funds. The regulatory tightening means this source of incremental capital is being restricted, while the deleveraging of these products further amplifies stock price volatility.
JPMorgan analyst Nikolaos Panigirtzoglou noted, "Since peaking in June, assets under management (AUM) for memory chip leveraged ETFs have shrunk by 34%, while total leveraged equity ETFs have declined by only 13% over the same period."
He believes what is more noteworthy is that the AUM of leveraged ETFs for memory stocks accounts for a proportion of the related companies' market capitalization that is about three times that of regular equity ETFs, making them a significant amplifier of industry volatility. Once stock prices enter a correction phase, the daily rebalancing mechanism of leveraged ETFs before the market close further exacerbates the decline.
Market Repricing Driven by Three Key Factors: Liquidity, Fundamentals, and Valuation
However, the South Korean regulations are merely the trigger for this adjustment.
Based on recent discussions on Wall Street, the market is effectively repricing the AI memory sector along three main lines: liquidity, fundamentals, and valuation.
Liquidity: Leveraged ETF Deleveraging Amplifies Volatility
It has been pointed out that a significant driver behind the substantial gains for memory stocks like Micron Technology over the past year has been single-stock ETFs offering leveraged exposure.
These products maintain a fixed leverage ratio through derivatives like options and swaps. Once stock prices begin to fall, they require daily rebalancing before the close, involving continuous selling of the underlying assets. This creates a negative feedback loop of "decline - reduction in position - further decline."
JPMorgan analyst Nikolaos Panigirtzoglou stated that since peaking in June, AUM for memory chip leveraged ETFs has shrunk by 34%, while total leveraged equity ETFs have declined by only 13% over the same period.
He pointed out that the AUM of leveraged ETFs for memory stocks accounts for a proportion of the related companies' market capitalization that is about three times that of regular equity ETFs, making them a key volatility amplifier. South Korea's regulatory tightening is further accelerating this deleveraging process.
Fundamentals: Market Begins to Question Proximity of Supply-Demand Inflection Point
Beyond liquidity factors, the market is also re-examining the supply-demand landscape for the memory industry over the coming years.
It has been noted that Dutch lithography giant ASML recently indicated that its next-generation EUV lithography equipment can further improve memory chip manufacturing efficiency.
The market anticipates that both SK hynix and Samsung Electronics plan to adopt next-generation EUV equipment. This suggests that future capacity expansion for high-end memory chips like HBM and DRAM may proceed faster than previously expected.
Any news that helps alleviate memory chip supply tightness easily triggers market concerns that the super-cycle duration might be shorter than earlier projections.
Simultaneously, a recent report by a research analyst suggests that the global memory chip shortage is expected to peak in the second quarter of 2026, with a gradual return to balance starting in the first quarter of 2027. The industry could potentially re-enter a phase of overcapacity as early as 2028.
This assessment implies that while AI has indeed extended the industry's upcycle compared to traditional memory cycles, it has not completely eliminated the cyclical nature of the sector.
Valuation: Sustainability of AI Capital Expenditure Comes Under Scrutiny
Another factor causing high market sensitivity is whether future AI capital expenditure can sustain its high growth rate.
It was reported this week that AI cloud computing company CoreWeave is exploring the use of financial instruments to hedge against potential future price declines for memory and storage equipment.
While the report did not indicate that CoreWeave expects memory prices to fall imminently, nor were there signs the company plans to cut AI capital expenditure, this news was interpreted by some investors as a sign that the supply chain is already beginning to consider future price cycle changes proactively.
Concurrently, rising geopolitical risks are prompting some capital to shift from high-valuation growth stocks to defensive assets. Ben Reitzes, Head of Technology Research at Melius Research, stated that against a backdrop of declining market risk appetite, capital is flowing out of previously crowded AI trades, with high-valuation memory chip stocks becoming a primary target for profit-taking.
For a stock like Micron Technology, which has risen nearly 700%, after its price has fully reflected optimistic expectations, any news regarding a slowdown in AI demand, increased supply, or changes in the price cycle could trigger a market repricing.
Wall Street Maintains Bullish HBM Outlook, But "Best Expectations" May Already Be Priced In
It is noteworthy that the sharp stock price decline does not signify a uniform shift to pessimism among institutions.
KeyBanc analyst John Vinh maintains an optimistic view on the memory industry, forecasting that NAND prices will rise 30% to 40% in the third quarter, continue to rise about 15% in the fourth quarter, and that HBM prices could double from current levels by 2027.
Similarly, Daniel Morgan, Senior Portfolio Manager at Synovus Trust, stated that in the upcoming earnings season for tech giants, cloud providers are expected to repeatedly emphasize compute power shortages. Data center storage remains one of the core bottlenecks for AI servers, and data center storage prices are anticipated to rise by another approximately 30% in the third quarter.
On another front, a research analyst recently noted that the current global memory chip shortage is expected to peak in Q2 2026, with a gradual return to balance starting in Q1 2027, and the industry could potentially re-enter overcapacity as early as 2028.
This judgment means that, compared to traditional memory cycles, AI has indeed prolonged the industry's prosperous cycle but has not completely eradicated its cyclical attributes.
For the market, the real question is no longer whether HBM demand is strong, but whether current stock prices have already front-loaded optimistic expectations for the next several years.
As AI computing power investment gradually moves into a realization phase, capital is beginning to focus more intently on future supply-demand balance, the sustainability of capital expenditure, and valuation safety margins. Following a year of substantial gains, the memory chip sector is shifting from "trading solely on AI demand" to "simultaneously trading AI demand and cyclical risks," with industry volatility rising significantly as a result.
Comments