The Fund Was Situational Awareness: What Actually Happened To Leopold Aschenbrenner — And What We Think Happened

Yesterday I wrote that a levered fund got taken down and that most of the selling in unrelated names was coincidental. Now the name is public. The fund was Situational Awareness, run by Leopold Aschenbrenner — the former OpenAI researcher who wrote the 165-page AI manifesto that made him famous in Silicon Valley in 2024 and then turned that fame into a hedge fund.

The numbers, per CNBC's reporting: roughly $45 billion in assets at the peak earlier this month, down to around $10 billion after being forced to sell the entire levered public equity book. Ken Griffin's Citadel bought that book at a discount. Reported leverage ran as high as 400%. Before this month, the fund was up more than 1,000% since its July 2024 launch, according to the Wall Street Journal. Aschenbrenner is 24.

I want to separate two things here, because the coverage is blending them: what is reported, and what I think happened. The first part is the record. The second part is my read as somebody who has been on the wrong side of a margin clerk and lived to write about it.

The trade was one of the most crowded expressions on Wall Street, just run at institutional size with retail-grade leverage: long the physical plumbing of the AI build-out — chips, memory, data centers, power — and short the software companies that AI is supposed to disrupt. Filings as of March 31 showed large stakes in Nebius, Bloom Energy, Sandisk, CoreWeave, SharonAI and IREN. By Wednesday's close, those names were down somewhere between 50% and 78% from recent peaks.

The short leg is where it turned lethal. Software names used as the hedge — Adobe among them — rallied while the long leg was getting destroyed. That is the definition of a factor unwind: both sides lose at once, and the hedge stops being a hedge and becomes a second losing position.

The scale of the reversal is not opinion. Morgan Stanley's sector-neutral momentum index fell 17.4% in four trading days, the worst such move on record per BTIG — bigger than the dot-com unwind, bigger than the pandemic shock, bigger than 2022. BTIG's Jonathan Krinsky called it the largest and fastest momentum crash in modern history and said it wasn't close. MTUM had its best month ever in April and is closing out its worst.

Then the plumbing took over. As the portfolio marked down, the equity cushion shrank, prime brokers asked for more collateral, raising collateral required selling, and selling pushed the same names lower — which required more selling. That is a deleveraging spiral, and it does not care about anybody's thesis on artificial general intelligence. Roughly two-thirds of the book was long/short public equity; the rest was private, dominated by a multibillion-dollar Anthropic stake, which is exactly the part you cannot sell in a hurry. The public book went to Citadel. The investor letter that followed said, in effect, it's not over.

And here is the part that should make every retail trader pay attention: the S&P 500 sat near record highs while this happened. The index told you nothing. The damage was entirely underneath the surface, in the factor.

My read: this was not a bad idea. The AI infrastructure thesis was and is directionally right. What killed it was a structure problem stacked on top of a correlation lie, and then leverage turned a drawdown into a funeral.

Start with the correlation lie. Long AI hardware versus short legacy software looks like a hedged book on a risk report. It isn't. It is one bet — long the AI-adoption factor — expressed twice. Both legs are the same trade wearing different jerseys, so when the factor reverses, your gross exposure works against you on both sides simultaneously. A real hedge loses money when your thesis works. If your short leg makes money in exactly the same conditions as your long leg, you own a leveraged directional position and a risk report that lies to you about it.

Second, the position sizing was the strategy. When your longs are the highest-beta, lowest-float, most narrative-dependent names on the tape — a memory name, a couple of neoclouds, a power name, a bitcoin miner turned data-center play — you are not underwriting cash flows, you are underwriting flow. Those names went up because everybody wanted them. That means they come down when everybody needs out, and there is no valuation floor to catch you because there was never a valuation ceiling on the way up.

Third, 400% leverage against that kind of book is a countdown, not a strategy. At four times gross, a 15% adverse move against the whole factor is your equity. There is no scenario in which you survive a record momentum crash while carrying that. You don't get to be right later; the prime broker forecloses on your thesis at Wednesday's close. That is what people mean when they say the blowup was a question of when, not if.

Fourth — and this is the one nobody in the coverage says plainly — I think he got liquidated partly because everybody knew what he owned. The filings were public. His followers traded off them. Which means the Street knew, name by name, exactly which illiquid positions had to be sold and roughly how much. When your book is public and your leverage is known, the market can front-run your margin call. That is not a conspiracy, it's just what happens: everyone steps away from the bid in the names you have to sell. The gap between the last print and where the block actually cleared is the price of being famous for your positions.

Fifth, I don't think Citadel bought a distressed pile. I think Citadel bought good inventory at a bad-inventory price, which is the whole business. Somebody with permanent capital and no margin clerk gets to take the other side of a schedule-driven seller. That is the single most reliable edge in markets and it has nothing to do with predicting anything.

The kicker on experience: he had never run outside money before July 2024, and his prior finance-adjacent stop was FTX's philanthropic arm. Being early and correct on a technology is a research skill. Sizing, hedging, financing and surviving are a completely different skill, and the market charges tuition for the second one regardless of how right you were about the first.

Yesterday I said the tape had two unrelated things in it: mechanical supply with a deadline, and a real fundamental repricing at Coinbase. Nothing in the Situational Awareness reporting changes that, and one detail makes it sharper — SNDK was in the book. I flagged getting out of SNDK weeks ago and I've been writing about that unwind since June. Some of what looked like fundamental deterioration in those names was a fund being unwound, name by name, on a schedule.

This is also why I refused to read Friday's prints as information about the underlying businesses. When one seller has to be out by a date, the tape shows you the seller's liquidity problem, not the company's earnings power. Nebius, CoreWeave and Sandisk did not become 60% worse companies in four days.

The confirmation showed up Thursday and Friday: AI infrastructure rebounded hard once the largest forced seller stepped away, and the Nasdaq gained two straight days. That is the classic post-liquidation snap-back — the pop on lighter volume that tells you the seller was a clerk, not an investor. Cramer called it a clearing event. I'd call it evidence, not a bottom.

Your hedge has to be uncorrelated to your thesis, or it isn't a hedge. If both legs win together and lose together, you're levered long and lying to yourself. Test it the ugly way: name the scenario where your short leg makes money while your long leg loses. If you can't, you have one position.

Leverage is a time constraint disguised as a return multiplier. Every unit of leverage shortens how long you're allowed to be wrong. Being right on a two-year horizon at 4x gross is functionally being wrong, because you don't own the position anymore by the time the thesis pays.

Crowding is a risk factor, and it is the one that doesn't show up in your P&L until the day it takes all of it. When your longs are the consensus longs of every fund running your strategy, your exit is the same door as theirs and it's narrower than the room.

Momentum reversals are the way concentrated books die — not recessions, not bad earnings. Index calm means nothing. A 17% four-day factor move with the S&P at highs is a reminder that the thing that kills you is under the surface, not on the front page.

And the practical one: forced supply is inventory, not information. When a name is down because a levered holder has to be out by Friday, that is a discount with an expiration date. When a name is down because the business changed, that is a warning. You get paid for knowing which one you're staring at — and you get paid more for waiting for the schedule to finish before you buy it.

The factual record here is drawn from CNBC's reporting on July 30 and 31, 2026 (the forced unwind and the Citadel purchase, the fall from roughly $45 billion to about $10 billion, the reported leverage near 400%, and the March 31 filing positions), BTIG's note on the record momentum reversal, and the Wall Street Journal's prior profile of the fund's performance since inception. Business Insider reported the substance of the investor letter that followed.

Everything under "What We Think Happened" is my interpretation, not reporting. I have no inside knowledge of the fund, its financing terms, or its counterparties, and I am not claiming any.

This is my opinion and analysis as of publication on July 31, 2026. It is not investment advice and not a recommendation to buy, sell, or short any security. Reported figures are attributed to the outlets named above and may be revised as more becomes public. Nothing here alleges wrongdoing by any person or firm named — a levered fund taking losses in a factor reversal is a risk-management outcome, not a legal one. Do your own work before risking capital.