The semiconductor sector just endured its worst week of the year, yet retail investors poured in with record-scale triple-leveraged dip-buying, creating a rare standoff against hedge funds' record net selling.
Goldman Sachs Prime's latest data shows global tech stock net selling last week hit an all-time record, with net exposure dropping to the 0th percentile; however, gross exposure remained as high as 213.9%, sitting at the 83rd percentile.
Meanwhile, retail investors used leveraged ETFs to buy the dip against the trend. Goldman Sachs ETF specialist Chris Lucas noted in a Sunday report that nearly $10 billion flowed into leveraged ETFs over the past week, the largest single-week inflow since data began in 2006, with approximately $7 billion going into the triple-long semiconductor ETF SOXL, a record for that product, contributing 74% of the week's net inflows.
Goldman Sachs derivatives and fund flow strategist Brian Garrett judges that positioning is already extremely low, market breadth is improving, and the AI debt burden is not new news 鈥?the next move "depends more on earnings confirmation than anything else." With institutions significantly underweight but not truly exiting, and retail leveraged capital highly concentrated, earnings season becomes the adjudication point for the bull-bear standoff.
Semiconductors hit by massive sell-off
Last week, the Nasdaq 100 closed at its weekly low, down about 5.5% for the week, the largest decline since "Reciprocal Tariff Day, April 2." The Philadelphia Semiconductor Index (SOX) fell for 8 consecutive trading days, dropping about 11.5% for the week.
Goldman Sachs TMT strategist Peter Callahan described in an internal report published Sunday:
It was an extremely difficult week... Institutional fund flows show clear selling pressure in the tech/AI sector, with the market struggling to lower growth expectations 鈥?driven both by pressure from rates and oil prices, and by growing external doubts about linear AI capital expenditure growth.
Semiconductors have retraced about 25% from their peak, with the RSI approaching oversold territory for only the second time in four years. Callahan also noted that Goldman Sachs Research estimates AI capital expenditure currently contributes about half of S&P 500 EPS growth.
Retail's "triple-leveraged dip-buying" collides with institutions' "record retreat"
Facing an 11% weekly decline, retail investors doubled down with triple leverage. Beyond the triple-long semiconductor ETF SOXL attracting about $7 billion, the triple-long Nasdaq TQQQ saw inflows of about $1.3 billion, driving total leveraged ETF inflows for the week to approximately $9.5 billion (October 12-16), roughly double the previous record high.
Goldman Sachs historical data shows that leveraged ETF fund flows move inversely to the market: such products see their strongest inflows on days when the S&P 500 falls more than two standard deviations, averaging net inflows of about 56 basis points of prior-day AUM over the past five years; on the best-performing days, they instead see outflows of about 17 basis points. Lucas summarizes this as "buy panic, sell euphoria."
But history also provides warnings. Goldman Sachs' Benny Quek noted that Korean leveraged ETF exposure has already collapsed, with assets under management shrinking from a peak of $53 billion to $15 billion; the triple-long Korea ETF KORU has hit a new 2026 low, down 80% from its June high.
On the institutional side, Goldman Sachs Prime's global tech stock net selling for the week hit a record, with cumulative selling from October 8-15 comparable to late July levels, the largest since March 2022; in Asia alone, net selling reached $7.8 billion, primarily in Taiwan. Quek believes institutional positioning is "about 6/10, notably cleaner than June and July," with the real concern being fundamentals 鈥?rates and credit, earnings, ROIC.
"Reducing longs" has happened, "shrinking scale" has not
Garrett revealed the subtlety of institutional positioning adjustments: US fundamental long-short net exposure fell 3.2 points in a single week, the largest decline since April 2, dropping to the 0th percentile, meaning "reducing longs" has already occurred; however, gross exposure remains at 213.9%, at the 83rd percentile on a 3-year lookback, and "shrinking scale" has not yet appeared.
In other words, fund managers chose to hedge rather than truly deleverage. They added short hedges on top of longs but did not reduce overall position size. This means those crowded long positions remain fully on the books.
On valuation, the Nasdaq 100 forward P/E has fallen back to 20.5x, the lowest since the AI cycle began in 2023, with its premium over the S&P 500 narrowing to about 1.8x, the tightest in nearly 10 years; the semiconductor sector has de-rated from about 29x forward to 17x. Meanwhile, Friday's S&P 500 put option notional volume reached $3.3 trillion, a historical record, and Goldman Sachs' panic index rose to 7.4.
Will this dip-buying work?
Lucas acknowledges that retail buying the dip is nothing new:
Fund flows within the leveraged ETF complex have historically been strongest when the market faces severe declines (negative 2 standard deviation events), and weakest on the market's best days. Additionally, it's worth noting that inflows during sharp market drops are also larger in absolute terms.
Over the past five years, this logic has indeed worked 鈥?every crash had the Fed providing support behind the scenes.
But Garrett's Asia-Pacific colleague Benny Quek provides a counterexample: Korean leveraged ETF assets under management have already fallen from a peak of $53 billion to $15 billion.
Credit market alarms ring louder than VIX
Garrett points out that the AI debt burden is "not new news," citing an article titled "AI Borrowing Floods the Bond Market," saying, "This sounds like something written just this weekend, but it's actually a Wall Street Journal report from a year ago."
But he also acknowledges that CCC-rated spreads have broken above 1000 basis points, while investment-grade credit default swap (CDX IG) volatility is "far more pronounced than 1-month VIX futures" 鈥?IG spreads have risen to their widest since April, while VIX futures have only edged modestly higher.
Earnings season becomes the adjudication point
Lucas specifically flagged a "loaded spring": last week's scramble pushed leveraged ETF year-to-date cumulative fund flows (excluding inverse products) from about -$12 billion rapidly back to about -$2 billion. Leveraged ETFs currently have about $152 billion in AUM and about $386 billion in gross exposure. If market conditions improve, this batch of new shares could become a multiplier amplifying any rebound; if semiconductors continue to decline, daily rebalancing becomes mechanical selling pressure, with a unit base far larger than a week ago.
Garrett provides a timeline anchor: 14% of S&P constituents report earnings this weekend, 44% next week. Meta, Tesla, and Lam Research will report results on Wednesday (October 21), and on the same morning Nvidia CEO Jensen Huang will deliver a keynote at GTC Berlin. Whether underweighted institutions are forced to cover on positive earnings and whether retail's triple leverage can wait for a rebound will be revealed this week.
The next observation point is the sustainability of SOXL fund flows this week and Wednesday's earnings guidance from Meta, Tesla, and Lam.
Comments