A 24-year-old turned a contrarian bet on the AI buildout into a $45 billion hedge fund. Four trading days and a wave of margin calls wiped out almost all of it.
Leopold Aschenbrenner ran a fund called Situational Awareness LP. At the start of July it managed $45 billion. By July 30, according to Bloomberg, that figure had fallen to roughly $10 billion after Ken Griffin's Citadel agreed to buy the leveraged public stock positions still sitting in the book, at a discount to where they'd been marked days earlier.
The fund didn't blow up because the trade was wrong. It blew up because of how much borrowed money was riding on it.
Aschenbrenner is 24. He worked as a researcher at OpenAI before departing in 2024, then wrote a widely circulated essay series called "Situational Awareness" that argued AGI was coming faster than almost anyone in finance was pricing in. He named his fund after it and built a portfolio around the thesis: go long the companies supplying the chips, memory, data centers and power behind the AI boom, and short the software incumbents that looked most exposed to being disrupted by it. His biggest long positions, according to a widely shared breakdown from Bloomberg and trading desk chatter, included SK Hynix, Nebius, SanDisk, Micron and CoreWeave. His short book leaned on names like Adobe.
It worked, spectacularly, for a long time. The fund returned more than 1,000% since inception and 439% in the first half of 2026 alone, based on figures reported by the Financial Times. That kind of run doesn't happen on conviction alone, and it didn't here. CNBC and Bloomberg both reported the book was leveraged roughly four times over, financed through prime brokerage arrangements with Goldman Sachs, JPMorgan Chase and Bank of America.
Then the trade reversed, hard. From its June 22 peak, the Philadelphia Semiconductor Index fell 28.6%. Morgan Stanley's sector-neutral Momentum Index dropped 17.4% over four trading days, its worst such stretch on record, worse than the momentum unwinds that followed the dot-com bust, the pandemic crash and the 2022 inflation-driven bear market. The bank's narrower Momentum TMT Index fell 53.5%. Aschenbrenner's long book and his short book lost money at the same time. SK Hynix and the rest of the AI infrastructure names sold off, while software stocks, Adobe among them, rallied. On a 4x leveraged portfolio, a reversal like that doesn't just hurt. It forces action.
By Thursday morning, July 30, Goldman Sachs, JPMorgan and Bank of America had issued margin calls the fund couldn't meet without selling. Citadel stepped in and bought the bulk of the leveraged public equities, SK Hynix and CoreWeave among them, at prices that reflected the seller's desperation more than the assets' worth. As a public-equities vehicle, Situational Awareness effectively ceased to exist that day.
Here's the thing that gets lost in the headline number. Aschenbrenner didn't lose everything. Situational Awareness held its public stock positions on margin, but its private holdings, including a stake in Anthropic reportedly worth around $5 billion according to Bloomberg, weren't pledged against the same margin lines and weren't part of the Citadel deal. TechCrunch reported the fund kept that stake intact even as the public book was liquidated. The roughly $10 billion left in the fund is now mostly that private portfolio.
Jim Cramer had, almost eerily, flagged the risk just two days before the unwind. On July 28, he told CNBC viewers to get out of any AI infrastructure trade financed with borrowed money. When the fund collapsed two days later, he called it a clearing event, the kind of forced selling that can mark a bottom rather than the start of something worse. Whether he's right about the bottom is a separate question. What isn't in dispute is that the mechanism he warned about, leverage amplifying a momentum reversal on both sides of a book at once, is exactly what took Situational Awareness down.
None of this means the underlying AI infrastructure thesis was wrong. Chip demand, data center buildouts and power constraints haven't gone away because one fund got margin-called. What broke wasn't the idea. It was betting on that idea four times over with someone else's money, in a market that can move 28% against you in five weeks.
Also read: Meta Stock Falls 10% as Reality Labs Losses Blow Past $80 Billion • Chinese Chipmaker CXMT Surges 466% to Overtake ICBC on Debut • Saudi Arabia's PIF Takes Control of Electronic Arts in $55 Billion Buyout