Hedge funds sold off 2-year AI positions in five days; Goldman warns of ongoing deleveraging.

Original author: Zhao Ying

Source: Wall Street Journal

Hedge funds liquidated their AI positions, accumulated over nearly two years, within five trading days. The intensity and speed of this selling wave are sparking concerns about broader deleveraging in the market.

According to Vincent Lin, Head of Goldman Sachs Prime Brokerage, in a recent report, hedge funds net sold global AI stocks for five consecutive trading days from October 8 to 14, completely reversing all net purchases made since the beginning of 2025—in other words, every net purchase made over the past two years was erased within a week.

Meanwhile, Goldman Sachs' Prime Book shows that global equities experienced "significant net selling," with a one-day net selling volume reaching -2.1 standard deviations under a one-year lookback, led by declines in North America and emerging market Asia, with South Korea and Taiwan seeing the highest concentration of net selling.

The direct impact of this sell-off has already manifested across multiple markets. The Korea Composite Stock Price Index has entered bear market territory, Seoul has triggered its circuit breaker mechanism, and Taiwan’s market recorded its largest single-day net foreign outflow in history, reaching approximately $9.3 billion. Central banks across several Asian countries have been forced to intervene in currency markets. The TMT directional long/short fund suffered a single-day loss of 2.7%, representing an extreme move of -2.6 standard deviations, with cumulative losses for October expanding to 4.5%.

The scale of this AI stock sell-off is the second-largest in Goldman Sachs' historical data, behind only the crash in July this year. According to Goldman Sachs Prime data, the net selling volume of global AI stocks (constituents of GSXGGLAI) was the second-highest over the past year, with a long-to-short selling ratio of approximately 3.5 to 1—indicating that this sell-off was primarily driven by long positions actively reducing exposure, rather than short sellers increasing their bets.

Looking at the flow chart, the net flow of hedge funds on the AI list (black line) has retreated to the starting point of early 2025. Long flows (blue line) have consistently declined since peaking in February to March this year, while short flows (red line) have remained near a low of approximately -13% throughout the period. Goldman Sachs’ chart clearly illustrates this "back-and-forth": the entire net buying cycle from 2025 to 2026 was reversed within a week, as shorts continued to add positions during the decline.

Just last week, Goldman Sachs noted that hedge funds in its Prime book had a net exposure to the Mag7 accounting for nearly 22% of total U.S. exposure, a historical high. Five days later, the same group of funds collectively exited.

Despite the heavy selling this week, AI positions remain elevated over the longer term, and the scope for deleveraging cannot be ignored.

According to Goldman Sachs data, the net allocation to global AI stocks as a percentage of global single-stock net market capitalization has declined from a peak of 31.7% in early June to 26.1%, placing it at the 49th percentile over the past year—but still at the 83rd percentile over the past three years. The long-short ratio for AI portfolios has fallen from a June peak of 4.34 to 3.27, placing it at the 7th percentile over the past year—but still at the 64th percentile over the past three years.

In short, one-year data shows positions have been relatively unwound, but three-year data indicates hedge funds as a whole remain significantly overweight. Rachel Hu from Goldman Sachs Asia noted that trading feedback from the Taiwan market suggests this selling wave has been less of a "capitulation" than in July, implying further selling pressure may still lie ahead.

JPMorgan's Market Intelligence TPM position indicator is currently in the "neutral" range at approximately the 46th percentile, well above the 19th percentile seen at the low point in late July, suggesting there is still room for "further deleveraging" ahead of key tech catalysts.

This selling pressure was not limited to the AI sector but reflected a broad-based defensive shift. Goldman Sachs Prime Book data shows that 8 out of 11 U.S. sectors experienced net selling, with Information Technology and Industrials leading the decline; all sub-sectors within Information Technology—software, semiconductors and semiconductor equipment, and IT services—were sold off net.

Meanwhile, hedge funds' buying focus has been on healthcare and consumer staples, which Goldman Sachs characterized as a "marginal shift toward defensive positioning." Early Thursday, healthcare and consumer staples each rose over 2%, supporting the Dow, while the Nasdaq slid toward its lowest level in nearly a month; CoreWeave fell another 8.5%, and the Philadelphia Semiconductor Index posted its seventh consecutive day of declines, wiping out approximately $2 trillion in market value.

In this collective loss among human fund managers, quantitative funds emerged as the sole winners. Global systematic long/short funds rose 0.5% on the day, gained 1.9% for October, and led all strategies with a year-to-date return of 21.7%.

Leverage data reveals why this sell-off was so severe. The overall gross leverage ratio of Goldman Sachs' prime book decreased only slightly by 0.4 percentage points to 310.3%, ranking at the 72nd, 91st, and 94th percentiles over one-year, three-year, and five-year horizons respectively—near five-year highs. Net leverage decreased by 1.5 percentage points to 73.3%, with a cumulative decline of 4.3 percentage points in October, placing it at the 6th percentile over the past year.

Goldman's logic is that a fund with a total leverage of 310% and a net leverage of 73% has a much larger actual sellable size than the net exposure suggests. When crowded long positions (AI) and crowded short positions move in opposite directions simultaneously, deleveraging becomes the only option—and the exit is extremely narrow, as everyone rushes through the same door—exactly as happened during the crash at the end of July.

Rich Privorotsky of Goldman Sachs' Delta-One business summed up the current dilemma in an overnight report: "In hindsight, the painful trade was going long AI and short bonds. Now we're seeing both ends unwinding... The market is realizing that demand for computing power can grow exponentially, but the capital available to finance it cannot."

The key divide in today’s market is whether this is a buying opportunity or the beginning of a larger deleveraging trend.

JPMorgan tends to favor the former, believing it presents a buying opportunity on dips. Goldman Sachs’ flow trading desk was still observing positive technical signals at year-end over the past two days. However, Goldman Sachs’ overall assessment is more cautious: AI net exposure remains at the 83rd percentile over a three-year horizon, and total leverage stands at the 94th percentile over a five-year horizon, while foreign capital in Asia is exiting at a record scale—these signals collectively suggest that this deleveraging cycle is closer to its beginning than its end.

Hedge funds have sold all their AI positions purchased since 2025. However, the positions accumulated in 2023 and 2024 remain fully intact on their books.