The losses, in order

Start with the weekly figure, because the daily ones can mislead. A 2.9% loss in a week for diversified stock-picking funds is a bad week, not a collapse, and these funds are still up 14.3% for the year. What makes the note worth reading is the pattern inside it: losses on every day of the week, across strategies at the same time. On Wednesday, October 14, according to Reuters, stock pickers globally lost 0.9%, Asia-focused managers lost 2.6%, tech-focused funds lost 2.0%, and systematic and multi-strategy funds lost roughly 1% each. Goldman also notes the simultaneity: all three broad strategy groups it monitors falling more than 80 basis points on the same day did not occur once between COVID and the end of 2025. It has now occurred three times in 2026 — on 4 February, 28 July and 15 October.

Simultaneous losses point to positioning, not to one bad stock call. When different strategies lose together, the usual explanation is that they held similar trades and tried to reduce them at similar prices. The report ties the unwind to AI positions, with semiconductor and AI supply-chain names hit as inflation data pushed rate expectations up again.

Asia held the concentrated version

The regional split is where the note sharpens. Asia-focused managers lost 2.6% on Wednesday, October 14, more than double the global stock-picker figure, and tech-focused funds lost 2.0%. For the first half of October, Asia-focused stock pickers are down 6.8%. Set against their position at the late-June peak, up more than 40% for the year at that point, the October loss has erased more than half of those peak gains in about two weeks of trading.

The index moves match that concentration. The KOSPI, heavy with semiconductor names, fell 8% on Thursday, October 15, to a six-month low. Taiwan stocks fell more than 3%. The losses in Asia are the observable part. The reading that Asia was also where the AI supply-chain trade was most crowded, and therefore where an exit costs the most to make, is an interpretation: the report does not disclose how crowded any portfolio actually was. Either way, it is a statement about market structure, not about whether Korean or Taiwanese chipmakers became worse businesses this week.

What to watch instead of the headline

Two numbers in the note argue against reading this as a verdict on AI spending itself. The first is the year-to-date gain: at 14.3%, it sets the scale of the week — a drawdown inside a strongly profitable year, not by itself a collapse. It does not say why managers sold, and it cannot distinguish a repricing from an orderly reduction. The second is the trigger the report itself names alongside AI doubts — inflation and the rate path. Expensive, long-duration trades fall hardest when borrowing costs look set to stay high, whatever the sector.

The test for the coming weeks is narrower. If the losses stay concentrated in AI hardware and semiconductor exposures and the funds' broader books stabilise, this was an unwind of positioning. If the simultaneous-loss days keep recurring across unrelated exposures after positioning has had time to clear, the market is repricing the spending itself, and the CDS and chip-index moves this site covered earlier in the week will look like the start of that story rather than a bad week inside it. One boundary on the evidence: simultaneous negative returns are not direct evidence of margin calls or funding stress. The note shows the pattern of losses, not the mechanics behind them.

Sources

  1. [1] Reuters, Summer Zhen and Anirban Sen, "AI selloff hammers hedge funds this week, Asia managers lead losses", Tokyo, 16 October 2026, published 02:13 UTC (Goldman Sachs prime brokerage note figures: weekly and daily fund losses, Asia losses, strategy simultaneity, KOSPI and Taiwan moves, year-to-date gains)Read source
  2. [2] ALFW, "Chip Stocks Fell 4.3% Despite ASML's Strong Demand Outlook", 14 October 2026 (earlier leg of the same selloff week, index level)Read source
  3. [3] ALFW, "AI Default Insurance Just Hit Record Prices. CoreWeave Is Past 1,000 Basis Points.", 15 October 2026 (credit-market leg of the same repricing)Read source