DeepSeek's New Model Hammers Chip Stocks as Leveraged ETFs Draw Record ~$10 Billion in a Single Week
nashnova research
DeepSeek released a faster, more compute-efficient AI model, sparking fears of lower memory-chip demand and an 11% weekly drop in the SOX index — yet retail investors poured a record ~$9.5 billion into leveraged ETFs, betting against a historic institutional sell-off.
Why did Asian chip stocks sell off first?
DeepSeek says its new model is faster and uses less compute. This means → if training AI requires less raw power, orders for memory-chip makers could shrink.
South Korea's KOSPI fell as much as 3.2% intraday; Samsung Electronics and SK Hynix each dropped over 4% before paring losses. Japan's Nikkei 225 slid roughly 2%, led by chip-equipment maker Advantest.
Life Asset Management CEO Kang DaeKwun said "there is too much uncertainty to call a bottom for Korean equities," citing U.S. AI-project funding as the core risk. In plain terms = Korean chip companies' profits depend on whether U.S. firms keep spending on AI.
How "surgical" was the semiconductor sell-off?
The Philadelphia Semiconductor Index (SOX) fell 11% last week — the worst weekly decline this year. The semiconductor ETF (SMH) posted 9 straight days of losses, the longest streak in its 15-plus-year history.
Goldman Sachs ETF strategist Chris Lucas noted that SPY, XLK, and QQQ outperformed SMH for 10 consecutive trading days, despite constituent overlaps of 20%, 37%, and 45% respectively. This means → selling was aimed squarely at semiconductors, not a blanket tech liquidation.
The trigger was not a single event: Broadcom CEO Hock Tan warned capital — not components — may be AI's next bottleneck → Goldman flagged a supply-chain "air pocket" → China's CXMT rattled Korean memory stocks with a DRAM-related announcement → DeepSeek's model release added the latest pressure point.
How much leverage are retail traders using to buy the dip?
Roughly $9.5 billion flowed into leveraged ETFs last week — the largest single-week inflow since Goldman began tracking the data in 2006, about twice the prior record.
The 3× bull semiconductor ETF (SOXL) alone drew about $7 billion, a product record, accounting for 74% of all leveraged-ETF net inflows. The 3× bull Nasdaq-100 ETF (TQQQ) came second at roughly $1.3 billion.
In plain terms = across every down week in SOX history, a ~12% drop typically attracted no more than $2 billion into SOXL. Last week's $7 billion data point "sits alone in the upper-left corner" of the scatter plot — a completely different order of magnitude.
Are leveraged-ETF holders always contrarian?
Lucas notes that leveraged-ETF holders historically buy about 56 basis points of prior-day AUM on major down days (moves beyond two standard deviations) and net-redeem about 17 basis points on major up days.
This reflects the most consistent contrarian cohort in the market — they add on drops and pull back on rallies.
But "consistently contrarian" does not mean "consistently right" — that depends on whether the bet pays off this time.
What were institutions doing at the same time?
Goldman's prime-brokerage desk reported the highest net-selling ratio for global tech on record. The firm's global AI basket saw cumulative selling from October 8–15 comparable to the late-July deleveraging episode — the largest since March 2022.
In Asia alone, Goldman's net selling reached $7.8 billion, concentrated in Taiwan, where foreign net selling also hit an all-time high that day.
This means → retail is buying with record leverage the very sector institutions are selling at record scale — the two sides are moving in opposite directions.
Who will be proved right?
Goldman Asia strategist Benny Quek cited the Korean leveraged-ETF market as a cautionary precedent: Korean leveraged-ETF AUM has fallen from a peak of $53 billion to roughly $15 billion — retail leverage does not always win.
KOSPI just posted its worst weekly performance since October 2008, yet it is still up more than 30% year-to-date. In plain terms = the short-term drop looks severe, but on a full-year view the index remains deeply in the green.
The standoff between retail and institutions ultimately hinges on one variable: whether AI capital expenditure continues to deliver.
市场有风险，内容仅供研究参考，不构成投资建议。