Goldman Sachs prime brokerage data reveals that from October 8–14, hedge funds net sold global AI stocks for five consecutive trading days, wiping out in one week the cumulative net buying built up since the start of 2025. Two years of accumulation, cleared in five days. In the same week, the world's 20 largest AI and semiconductor stocks lost over $1 trillion in combined market value. This is not a story about a model falling behind on benchmarks—it is a story about positioning itself.
Quant Funds Take the Year's Heaviest Hit
On October 15, hedge funds experienced a rare multi-strategy synchronized decline—the third of its kind since the pandemic. Quant funds suffered a shock of roughly 3.7 standard deviations, the worst single day since August 19. Asia fundamental long-short funds extended monthly losses to 6.8%, with YTD returns compressed from a June peak of 40% to 16%. Korean AI positions were hit hardest: Seoul triggered circuit breakers, the index entered bear-market territory, and multiple Asian central banks intervened in currency markets.
Per a rough normal-distribution estimate, a 3.7 standard deviation day should occur only a few times per ten thousand days. Three such events in a single calendar year means either the distribution is wrong or the models need replacing. Goldman's Marco Laicini argues the latter: three once-in-five-years events in ten months suggests existing VaR models may systematically underestimate tail risk, warranting a review of industry risk frameworks.
Catalyst: OpenAI's Annualized Revenue $20 Billion Short
Per an October 8 Financial Times report citing investor documents, OpenAI disclosed annualized revenue of roughly $50 billion as of end-September—about $20 billion below the ~$70 billion figure previously circulating. CNBC confirmed the number; a source explained the discrepancy: the widely reported $68 billion included gross partner revenue (including from the Nvidia partnership). OpenAI's investor materials claim 77% total run-rate growth in Q3 and 107% for enterprise; Bloomberg later reported the company expects to reach or exceed $70 billion by year-end. On the news, Nvidia fell ~3%, Oracle nearly 6%, CoreWeave nearly 8%, and AMD and Broadcom ~5% each.
The number strikes at the repayment source of the entire chain: investors fund OpenAI, OpenAI rents data centers from cloud providers, cloud providers buy chips, and chip giants loop capital back via investment guarantees. Only the model company collects money from end users—if revenue misses, the credit narrative risks turning from a growth story into a bill-collection problem. Context: OpenAI is reportedly negotiating a new round at a ~$1.4 trillion valuation seeking $30 billion+, secretly filed for IPO in June (delayed to at least next year), while Anthropic's July run-rate was $65 billion with a possible listing as soon as November.
How Extreme Was Positioning?
Global AI stocks saw six consecutive days of net selling, with cumulative six-day net flow Z-score below −3—the most extreme de-positioning since ChatGPT launched. Goldman analyst Mark Wilson notes sentiment, pricing, and positioning momentum all turned negative simultaneously, and the pendulum has likely swung too far. A trade running for three and a half years—long AI winners, short AI risk assets—gained over 90% within the year but has nearly halved since end-June. On October 16, the Nasdaq 100 fell 2% in a day and 5.5% for the week, the worst week since April 2025; the Philadelphia Semiconductor Index fell eight straight sessions, losing ~$2 trillion—the longest streak since 2008. CoreWeave's decline that day was reported as nearly 12% by one outlet and 15% by another; its five-year CDS hit a record high.
Leverage remains elevated. Goldman warns that as options expire, hedge buffers on some positive-Gamma positions weaken, and further risk reduction by hedge funds could transmit to broader markets.
The Same Week: A-Share Quant's Identical Stampede
The CSI 300 fell 3.4% for the week, with quant index-enhanced products drawing down broadly: a RMB 60 billion-scale firm's products fell ~9% in a week; Hefu Investment's index-enhanced fell 8.2%; Mengxi's CSI 1000 enhanced fell 7.75%; Wenbo fell up to 7.5%; Century Frontier over 6%. The anatomy is consistent: years of excess returns came from micro-cap mispricing; after regulators tightened DMA and programmatic trading reporting, liquidity concentrated in dividend and large-cap blue chips, leaving micro-caps illiquid. Hundreds of billions running similar factors meant every machine sold simultaneously when the wind shifted—small caps hit limit-down, unsellable. Models assume "I can exit"; in extreme conditions that assumption fails first. As Chongyang Investment's Kou Zhiwei puts it: no emerging industry can sustain exponential growth indefinitely; when expectations revert from hypergrowth to linear, trading-structure fragility becomes the primary contradiction.
| Date | Event | Reading | |---|---|---| | Oct 8 | FT reports OpenAI revenue gap | ~$50B vs ~$70B | | Oct 8–14 | Hedge funds net sell AI stocks | Five days erase two years | | Oct 15 | Multi-strategy synchronized decline | Quant ~−3.7 std dev | | Oct 16 | Nasdaq 100 | −2% day, −5.5% week | | Week | Philadelphia Semiconductor Index | 8-day slide, ~$2T lost |
The Puzzle: Crowding Isn't New—But This Time It's the Narrative
Recent posts on this site fit together: one on inference compute failing to yield net trading returns; a 40-asset minute-bar study pinning costs and noise in the denominator; and a factor-crowding analysis in which no alpha among eleven factors survived after removing crowded ones. The novelty of this round: what's crowded isn't a factor but the AI narrative itself—and quant funds are both the makers of the crowding and the first sellers in the unwind.
Sources and caveats: Goldman Prime data is relayed via Securities Times, Wallstreetcn, and Sina Finance; the original Goldman report was not directly verified. The FT report is paywalled; OpenAI figures reconcile FT, CNBC, and Bloomberg accounts. A-share quant drawdown figures come from a Tencent News aggregation page and are directional only. Two outlets disagree on CoreWeave's Oct 16 decline; both are presented without adjudication. This article is not investment advice.