Hedge Funds Erased Their Entire 2025 AI Bet in Five Days

What happens when the smartest money in the room decides it wants out at the same time? You get five days of selling, a lot of bruised tech charts, and a brand new debate about whether the AI trade just broke.

Here's what we actually know, and here's what I think is being oversold as a narrative.

The raw numbers

Goldman Sachs prime brokerage data shows hedge funds sold AI stocks for five consecutive sessions in October. By the close of that run, they had undone every single AI purchase they made since the start of 2025. An entire year of accumulation, wiped out in a week.

Morgan Stanley's figures tell the same story from a different angle. Funds have cut more than 8% of their total equity exposure since June. That's the largest retreat in the data series going back to at least 2017, which is about as far as anyone's willing to cite with confidence.

Big tech absorbed the damage. That's where the selling landed, and that's where the prices moved.

Five days isn't a change of heart

Speed matters here. If a fund manager genuinely concluded that AI capex was a mirage, they'd trim over months. They'd fade the trade, write a letter to investors, and let the position bleed out quietly. Selling everything in five sessions is what you do when you need cash, not when you need to be right.

Color me skeptical, but "hedge funds quietly lost faith in AI" is a much easier headline to write than "hedge funds got a call from their prime broker and sold what they could." The second one happens constantly. The first one sells subscriptions.

So which is it? Is this a bubble finally popping, or forced selling dressed up as conviction? I lean toward the second, and I'll tell you why.

The bubble case, and its weak spot

The skeptics have a fair argument. Valuation multiples on AI infrastructure names are generous by any historical measure. Hyperscaler capex keeps climbing. And the revenue that's supposed to justify all of it's still arriving on a lag, which is fine until it isn't.

But bubbles don't usually deflate in a single week of orderly selling followed by a data release. They deflate over quarters, with denial, a bounce, and then the real capitulation. History suggests otherwise five-day panics. Those are usually plumbing problems, not thesis problems.

Here's my read, and it's a strong one: this is de-risking driven by balance sheet pressure, not a referendum on AI spending. Funds got crowded, got called, and got smaller. The direction of AI capex hasn't changed in the last three weeks. Only the ownership has.

The question worth asking: if the fundamentals were intact on September 30, what exactly changed by October 15?

What to watch next

Three things will settle this.

First, the Q3 13F filings, due November 14. Those disclose long positions as of September 30, so they'll be stale by definition, but they'll show how crowded the trade was right before the exit. Compare those filings to the Goldman flow data and you'll see exactly who blinked.

Second, hyperscaler capex guidance on the Q3 earnings calls from Microsoft, Amazon, Alphabet, and Meta. If those numbers hold or go up, the selling was mechanical. If any of them get trimmed, the skeptics earned their payday.

Third, watch whether that 8% gross exposure cut reverses inside four to six weeks. That's the tell. Forced sellers come back fast once the pressure lifts. Thesis sellers don't come back at all.

Time will tell, though I'd bet on a partial rebuild before year end. And if the capex guidance holds while the selling resumes anyway, then I'll admit the bubble crowd was early and right.