Everything About AI
NextFin News - Marshall Wace's flagship Eureka fund was caught in October's AI rout, a bruising month for the stock-picking hedge funds that spent the first half of 2026 leaning into the same artificial-intelligence winners. The London-based manager, which oversees more than $70 billion, posted a 2.64% gain in Eureka for the month, lifting its year-to-date return to 10.72%, but the fund's mid-month drawdown during the October 22 selloff underscores how quickly a crowded trade can turn against even the industry's most established names. The rout was not confined to one firm or one style: on October 22 alone, hedge funds suffered their worst single-day hit to equity books since the pandemic-era market turmoil, and the week that followed marked the deepest weekly losses for the industry in two years.
The breadth of the damage is the story. Goldman Sachs' prime brokerage desk told clients that all three broad equity strategies it tracks fell more than 1% on October 22, the first time that has happened on the same day since the Covid-19 pandemic.
"If last Thursday felt bad, today has been worse," the desk wrote.
Multi-strategy platforms, engineered to keep firm-wide drawdowns contained, slid 1.9%, their second-worst day since the April 2025 tariff selloff. Fundamental long/short managers lost 1.4%, their worst session since July 16, and systematic long/short books fell 1.3%, their third-worst day since December 2023.
The trigger: AI spending doubts turn into forced selling
The spark was Meta Platforms. Shares fell more than 7% on October 22, their largest drop since July, after the company raised its 2026 AI capital spending forecast to between $125 billion and $145 billion, up from $115 billion to $135 billion, and announced a new class of shares to help fund the buildout. Nvidia dropped roughly 4%, the Nasdaq Composite lost 2.6% and the Philadelphia semiconductor index fell 6.7% as investors asked the question the AI rally had spent two years avoiding: at what point does the bill for the infrastructure exceed the revenue it can produce?
That question hit hedge funds hardest because they were the most crowded side of the trade. Goldman Sachs estimated that more than two-thirds of the hedge funds in its prime book finished October 22 lower, and the losses were spread across styles rather than concentrated in one approach. Telecom, media and technology specialists were down an average of 2.7% on the day.
The damage had been building for two weeks before the October 22 climax. In the week to October 15, global fundamental long-short funds lost money on every trading day, ending down 2.9%, as a broad unwind of AI trades swept through equity books. Goldman said the selling was dominated by long liquidation rather than new shorts, and hedge funds net sold the bank's Global AI Basket of roughly 100 semiconductor, data-center, cloud and power names for five straight sessions from October 8 to 14, reversing net buying accumulated since early last year. Proceeds were rotated into healthcare and consumer staples, the classic defensive move when a growth trade breaks.
The pressure did not stop there. In the week ending October 26, trend-following hedge funds suffered their worst week since 2024, falling 2%, while fundamental long/short strategies fell another 1.1%, according to a Goldman note seen by market participants. The Nasdaq 100 dropped 5.5% for that week, its biggest weekly fall since April 2025, and Meta lost 10% for the week despite reporting strong earnings, after signaling it would need to raise more capital for its AI investments. Goldman's prime brokerage desk reported $20 billion in technology stock sales as the unwind accelerated.
Why the pain was so broad: crowding, leverage and the short squeeze
The striking feature of October 22 was not the size of the loss but its breadth. Losses that hit fundamental stock pickers, quantitative books and multi-manager platforms at once point to shared crowding in the same AI-linked names rather than isolated stock-picking errors. When the same hundred names appear on the long books of fundamental managers, the model portfolios of systematic funds and the pod books of multi-strategy platforms, a reversal in those names becomes a simultaneous hit across the industry.
The second mechanism was leverage. Market participants said the losses forced deleveraging as managers cut exposure to crowded AI trades. When crowded positions unwind, the exit itself moves the market: every fund selling the same names pushes prices lower, triggering risk limits and stop-losses at other funds, which sell into the same weakness. That feedback loop is why a 7% day in Meta can translate into a 1.9% day for multi-strategy firms that are supposed to be insulated from directional equity risk.
The third mechanism cut against funds that thought they were hedged. Concentrated short positions were squeezed in the second half of the October 22 session, meaning managers who had bet against weaker AI names were also punished as the session stabilized. Long books lost money and short books lost money, the Goldman note said. That is the signature of a liquidity-driven unwind rather than a fundamental reassessment: in a fundamental repricing, shorts typically provide a cushion; in a forced deleveraging, every position is a source of cash.
Who was exposed: the toll across the industry
Marshall Wace's experience captures the paradox of the month. The firm finished October positive in both Eureka and its Market Neutral Tops fund, which gained 1.15% for the month to lift its year-to-date return to 14.96%. But the mid-month rout tested the book hard enough to be singled out in the industry's post-mortem, a reminder that in a crowded unwind even well-run books are dragged down by the positions everyone else holds.
Other managers had it worse. Altimeter Capital Management, the Boston fund founded by Brad Gerstner, lost 11% in October, while Balyasny Asset Management and Verition Fund Management lost 1.5% and 1.1% respectively, market data showed. Technology-focused and Asia-focused hedge funds were among the weakest through October 22, down roughly 10% and 9% respectively, Goldman said.
Yet the industry entered the rout with a cushion. Global fundamental long/short funds were down 3.8% month to date through October 22 but still up 8.8% for the year. Asia fundamental long/short managers remained up about 14.2% year to date even after a 9.3% October decline. The question for allocators is whether the drawdown is a cyclical interruption in a still-intact AI investment thesis or the start of a structural repricing of the entire complex.
Cyclical or structural: a cyclical wave on a structural shift
The distinction matters because it determines what happens next. A cyclical fluctuation mean-reverts: positions get cut, leverage comes out, and the trade can resume once the forced selling exhausts itself. A structural shift does not revert on its own: the assumptions that justified the valuations were wrong, and prices must find a new floor.
The evidence points to a cyclical wave riding on top of a structural shift that is not yet disproven. Three historical-cycle comparisons support the cyclical read. First, the July 2026 AI slide produced similar fund losses and then reversed, with the same funds back in positive territory within weeks. Second, the April 2025 tariff selloff produced a multi-strategy day nearly as bad as October 22's, and the industry recovered its gross exposure within a quarter. Third, the pattern of long liquidation dominating the selling, with proceeds rotating into defensives rather than leaving the market entirely, matches the mechanics of a positioning washout more than a permanent capital reallocation.
But the structural leg is real, and it is what makes this cycle more dangerous than the last two. AI capital spending has moved from a growth narrative to a balance-sheet commitment measured in hundreds of billions of dollars. Meta's $125 billion to $145 billion capex range is not a pilot program; it is a multi-year obligation that must be serviced regardless of whether the revenue arrives. If the payoff lags, the spending does not simply reverse. That is the regime change: the industry has locked itself into a capital cycle that will outlast any single earnings disappointment.
The correct read is to separate the two. The October rout itself is cyclical and mean-reverting: leverage is already coming out, gross exposure is being cut, and the positioning washout creates the conditions for a rebound. The structural question is whether the AI revenue curve can catch up to the capex curve, and that will be decided over quarters, not days.
Second-order effect: the repricing the market is not talking about
The first-order effect of the rout is obvious: AI stocks fell, hedge funds lost money. The second-order effect is what comes next, and the market has not fully priced it. When hedge funds deleverage crowded positions, they do not just sell the losers; they sell what works to raise cash and meet redemptions. That means the strongest AI names with the most liquid markets become the last to stabilize, because they are the best source of cash for funds under pressure.
The cross-market transmission runs through prime brokerage. As gross leverage falls, prime brokers tighten financing terms, which forces further de-grossing at the most leveraged funds. That is the propagation chain: Meta's capex surprise triggers a tech selloff, the selloff hits crowded hedge fund books, forced selling widens the drawdown, prime brokers respond by tightening, and the tightening forces another round of selling. The terminal impact is not lower AI stock prices; it is a higher cost of capital for the entire AI complex, which feeds back into the very capex decisions that started the rout.
That is why the rout is self-reinforcing in the short run. Every dollar of AI capex that gets delayed because financing costs rose is another data point for the bears who say the spending was too aggressive. And every delay validates the funds that sold, which sells more. The loop only breaks when the capex curve and the revenue curve converge, or when leverage has been cut enough that forced selling exhausts itself.
The counter-thesis: what if this time is different?
The strongest case against the cyclical read comes from the scale of the commitment. This is not a positioning problem that leverage can solve; it is a valuation problem that earnings must solve. If AI revenue does not grow fast enough to justify $125 billion-plus annual capex at the hyperscalers, then the entire complex is marked to a reality that no amount of deleveraging can change. The July 2026 and April 2025 comparisons fail if the denominator has changed: the industry is not the same size of bet it was then.
That argument has force, and it is backed by the simplest arithmetic in the market. A fund that lost 11% in a month, as Altimeter reportedly did, is not being punished for positioning; it is being punished for being wrong about the payoff. And a firm like Marshall Wace, built on long-short equity selection, does not see its book tested in a rout because of leverage alone; it is tested because the longs it picked and the shorts it avoided both moved against it, which is a research error, not a risk error.
The falsifying signal is specific: if Meta, Microsoft, Alphabet and Amazon collectively guide 2027 AI capex lower for two consecutive quarters while AI revenue growth slows below 20% year over year, the structural-repricing thesis is confirmed and the cyclical read is wrong. Until that print, the evidence favors a positioning washout over a permanent repricing.
What to watch: the signals that decide which story wins
For allocators, the pattern matters more than the single day. The next test is whether managers can rebuild gross exposure without being forced to sell into further weakness. Three signals will decide which narrative wins.
First, gross leverage. If prime brokerage data shows gross exposure stabilizing and starting to rebuild in November, the washout is ending. If it keeps falling, the feedback loop is still active. Second, the AI basket. If Goldman's Global AI Basket stops being net sold and starts attracting net buying, the rotation into defensives is reversing. Third, capex guidance. If the hyperscalers hold their 2027 spending plans at current levels through the next two earnings seasons, the structural commitment holds and the cyclical rebound has a ceiling.
Short-term, the rout is likely to linger until leverage is fully out of the system. Medium-term, the funds that kept powder dry and avoided the crowding are best positioned to buy the rebound. Long-term, the structural question of AI's payoff remains unanswered, and the firms that can separate the cyclical noise from the structural signal will be the ones that make money on the other side.
The October rout is not the end of the AI trade; it is the market's first honest attempt to price it. The funds that treated it as a positioning problem will recover. The funds that treated it as a valuation problem will discover which one it actually was.
Explore more exclusive insights at nextfin.ai.
