美股

Foreign Investors Sold 35% More SK Hynix — So Why Did It Rise? Korea Memory Flow Data, 2026-08-25

2026-08-25 · 10 min · AlphaGBM
SK HynixSamsung ElectronicsKRXforeign investorsinstitutional investorsretail investorsKoreamemory semiconductorDRAMNANDMicronSanDiskfund flowsinvestor category datamarket structureorder flowprice discoverysemiconductor reboundDARTNaver Finance

The Short Answer

At the 2026-08-25 Korean close, foreign investors sold a net 1,234,309 shares of SK Hynix — 35.2% more than the prior session (08-24), amounting to 30.75% of the day's volume, the most intense foreign selling of this stretch.

The stock did not fall. It closed at ₩1,678,000, up +0.42%. The prior session, when foreigners sold less (22.86% of volume), it dropped 3.41%.

Same stock, consecutive sessions, near-identical volume (+0.5%). Foreigners sold harder and the price went the other way. "Foreign investors are fleeing" has zero explanatory power for price direction across these two sessions.

Data timestamps: Korean figures are 2026-08-25 close, settled final (KRX closed at 14:30 Beijing time). US figures are intraday readings at 22:32 Beijing / 10:32 ET on 2026-08-25 — roughly 62 minutes into the session, NOT settled and will change by the close. Read every US-tagged number as intraday.

1. Foreigners Sold More, and the Price Rose

SK Hynix 000660 08-24 (settled) 08-25 (settled) Change
Close ₩1,671,000 ₩1,678,000
Session move −3.41% +0.42% Direction reversed
Volume (shares) 3,994,599 4,013,938 +0.5%
Foreign net sell (shares) −913,265 −1,234,309 +35.2%
Foreign net sell ÷ volume 22.86% 30.75% +7.89pp
Domestic institutions (shares) −173,328 +197,025 Flipped to net buy
Foreign + institutional combined −1,086,593 −1,037,284 −4.5%
Foreign ownership 50.92% 50.75% −0.17pp

The line that deserves attention is the second from the bottom: combined net selling from both tracked investor categories was nearly identical across the two sessions (−1.087M vs −1.037M shares, a 4.5% difference), on nearly identical volume. The input did not change. The output reversed.

Three extractable facts:

2. So Who Bought? — A Limitation That Must Be Stated

The investor-category data used here breaks out only foreign and domestic institutional flows. Summed, both names remained net-sold:

08-25 net (shares) SK Hynix Samsung Electronics
Foreign −1,234,309 −5,003,317
Domestic institutions +197,025 +203,484
Two categories combined −1,037,284 −4,799,833
Residual (necessarily net buying) +1,037,284 +4,799,833

⚠️ That residual contains individual investors and other unbroken-out categories. My data source does not separate them, so I will not attribute it further. Claiming "retail absorbed the selling" requires the individual-investor line item, which I do not have here. I stop at "the residual was a net buyer."

And that is precisely the point: when people discuss "who is selling," the line almost always cited is the foreign one — because it is the easiest to obtain and the most narratively satisfying. But price is set by both sides, and across these two sessions, the side that determined direction is not in the line being cited.

3. Samsung Electronics: A Different Path, the Same Conclusion

Samsung closed 08-25 at ₩257,000, 0.00%, flat against the prior session (which was −8.70%).

Samsung Electronics 005930 08-24 08-25 Change
Session move −8.70% 0.00% Stopped falling
Volume (shares) 32,451,940 21,596,130 −33.5%
Foreign net sell (shares) −6,882,379 −5,003,317 −27.3%
Domestic institutions (shares) −6,269,240 +203,484 Flipped to net buy
Foreign + institutional combined −13,151,619 −4,799,833 −63.5%
Foreign ownership 46.74% 46.65% −0.09pp

Samsung and Hynix took two different routes:

What they share: neither stock stabilized because foreigners stopped. Foreigners were still net sellers of both. And in the 08-24 session, where Samsung fell 2.55x harder than Hynix, the differentiating variable was again domestic institutions (19.32% vs 4.34% of volume), not foreigners (21.21% vs 22.87% — nearly identical). Across two consecutive sessions, the institutional line moved with price and the foreign line did not.

4. US Confirmation: Stabilized, but Storage Has Retraced the Least

Following Korea's 08-25 stabilization, US semis opened higher that evening. Figures below are intraday at 22:32 Beijing / 10:32 ET, ~62 minutes into the session, NOT final:

Tier Ticker 08-24 (full session, settled) 08-25 (62 min intraday) Retraced
Broad SPY −0.29% +0.15% 51%
QQQ −1.00% +0.47% 47%
Semis TSM −2.11% +1.36% 65%
SMH (sector ETF) −2.43% +1.26% 52%
AVGO −2.63% −0.32% −12%
NVDA −2.91% +1.60% 55%
AMD −3.49% +3.84% 110%
Storage MU (DRAM) −5.83% +1.69% 29%
SNDK (NAND) −6.45% +0.61% 9%

⚠️ The "Retraced" column divides a 62-minute intraday move by a full prior session — two different window lengths. It will change before the close. Use it only to compare names against each other under the same pair of windows; do not read it as a trend or a completion percentage.

Within that limitation, two extractable facts:

1. The bounce is broad, not storage-specific. Eight of nine observations turned positive, from broad index to semis; AVGO is the only one still down. This is the same structure as the 08-24 decline running in reverse.

2. But the two tiers that fell hardest have retraced the least. MU has retraced 29% and SNDK just 9%, well below the 47%–65% range for broad and semi names. "Korea stabilized, therefore US storage is recovering its losses" is not what is happening right now. What is happening: storage stopped falling, but it is a laggard in this bounce, not a leader. AMD retracing 110% (fully recovered) against SNDK's 9% shows the rebound is not retracing in proportion to the 08-24 decline.

5. Conclusions

  1. Foreign net flow alone cannot determine the daily direction of Korean memory stocks. On 08-25, Hynix foreign selling hit a stretch high (30.75% of volume) while the stock rose — a direct counterexample.
  2. Across both 08-24 and 08-25, the variable that moved with price was the domestic institutional line. On 08-24 it explained the Samsung–Hynix gap; on 08-25 it was the only one of the two categories to change sign.
  3. Both categories combined were still net sellers; the absorber is an unbroken-out residual. Any conclusion of "capital is fleeing" drawn from the foreign line alone omits the side that set the price.
  4. US storage has stopped falling but lags the semiconductor rebound (as of 10:32 ET: MU retraced 29%, SNDK 9%, vs 47%–65% for semis).

6. What Would Falsify This

Next hard date: SK Hynix, 2026-09-18 — the re-disclosure deadline for its response to the exchange's rumor inquiry (the original response was filed as "undetermined," rcp 20260821800524). This is the only confirmed disclosure event on the sector calendar right now.


Sources: KRX / Naver Finance investor-category trading data (Korea, 08-25 settled); Tiger Brokers real-time quotes (US intraday, 22:32 Beijing / 10:32 ET); DART electronic disclosure system (Korean filings).

Disclaimer: This is public-market data analysis. It contains no positions, trade actions, or investment advice. All figures are timestamped and marked settled or intraday; intraday readings are not final.

想要更深入的期权数据与实盘分析?了解 AlphaGBM →