On July 1, Leopold Aschenbrenner's fund controlled roughly $45 billion of stock and had just posted a 439% net return for the first half of the year. Twenty-nine days later, three prime brokers margin called it, and the entire public equity book was sold to Ken Griffin's Citadel in one block trade before the market opened.
The Situational Awareness hedge fund collapse became the biggest Wall Street story of the summer within about six hours, which means most of what you have read about it is a mix of accurate reporting, guesswork, and a conspiracy theory that falls apart under thirty seconds of checking. I went through the coverage from the Financial Times, CNBC, Bloomberg, the Wall Street Journal, and the fund's own SEC filings to sort the confirmed from the invented.
The short version:
-
Situational Awareness LP was running heavy borrowing, reported at up to four times, on a concentrated AI infrastructure book.
-
Its long positions fell 35% to 47% in July while its short bets against software companies rallied against it. Both sides lost at the same time.
-
Goldman Sachs, JPMorgan Chase, and Bank of America issued margin calls. The whole public book, longs and shorts, went to Citadel on July 30.
-
The fund did not shut down. It kept a roughly $5 billion stake in Anthropic and continues as a private investment vehicle.
-
By the fund's own unaudited numbers, it finished July down 67% for the month and still up 80% for the year.
-
The stock picks were largely right. The borrowed money is what ended it.
What happened to the Situational Awareness hedge fund?
Situational Awareness LP was forced to sell its entire public equity portfolio to Citadel on Thursday, July 30, 2026, after its three prime brokers issued margin calls it could not meet. Assets that stood near $45 billion at the start of July fell to roughly $10 billion in a matter of days, according to Bloomberg.
The transaction was one enormous block trade covering both the long book and the short book, sold to a single buyer before Thursday's opening bell. CNBC's David Faber reported the unwind that morning, and the Wall Street Journal identified Citadel as the buyer. Bloomberg reported that Millennium Management and Jane Street also looked at the positions, though accounts differ on whether they declined outright or bid and lost.
Selling to one buyer instead of hitting the open market was a deliberate choice. A piecemeal liquidation would have told every trading desk on the street exactly which positions were coming, and they would have gotten in front of each one.
Who is Leopold Aschenbrenner?
Leopold Aschenbrenner is a German-born investor, still in his twenties, who ran the fund and had no professional trading experience before launching it in late 2024. He enrolled at Columbia at 15 and graduated as valedictorian at 19 with a degree in economics and mathematics-statistics.
He joined OpenAI's Superalignment team in 2023 and was fired in April 2024 over what the company described as an improper disclosure of internal information. Aschenbrenner disputes that framing and has said his dismissal followed a memo he wrote to the board arguing OpenAI's security against foreign industrial espionage was insufficient. OpenAI denies the two were connected.
Two months later, he published "Situational Awareness: The Decade Ahead", a 165-page essay arguing that AGI by 2027 was "strikingly plausible" and that the binding constraint would be physical rather than algorithmic. Power contracts, transformers, memory, land. The essay went viral well outside AI research circles and became the intellectual case for the entire AI infrastructure trade.
Then he built a fund on it. Backers included Stripe co-founders Patrick and John Collison, Nat Friedman, Daniel Gross, and later Jane Street, which almost never backs outside managers. Jane Street is also one of the firms that, per Bloomberg, looked at the distressed book last week and did not end up with it.
What was the fund actually betting on?
The book was a long position in the physical supply chain for AI and a short position against software companies that AI might disrupt. Long the picks and shovels, short the things the picks and shovels were supposed to make obsolete.
The long side held memory and chip names including SK Hynix, SanDisk, and Micron, neocloud providers CoreWeave and Nebius, power and infrastructure plays including Bloom Energy, and a large and fast-growing position in bitcoin miners. That last one surprised people who had not read the essay. Miners own power contracts, land, and cooling, which is exactly what Aschenbrenner argued the bottleneck would be. The fund's first disclosure covering December 2024 listed six holdings worth $254.8 million and not one crypto name. Its most recent 13F, covering positions as of December 31, 2025, listed 29 holdings worth $5.52 billion, with roughly $1.38 billion of it in miners and their data center arms. Core Scientific was the largest of those at $418.7 million.
The short side included software companies such as Adobe.
The performance was absurd. Up 47% after fees in the first half of 2025. Up more than 1,000% since inception. Up 439% net in the first half of 2026 alone, according to the Financial Times' review of the July 24 investor letter.
How did 439% turn into a forced sale in four weeks?
Both sides of the book went wrong at once, in the middle of what technicians are calling the fastest momentum crash on record. The longs fell, and the shorts rallied, so the hedge stopped being a hedge and became a second, separate source of losses.
Here is the timeline that matters:
-
June 22: the Philadelphia Semiconductor Index peaks. It falls 28.6% from there over the following weeks.
-
July 1: the fund's assets reach roughly $45 billion, closing out a first half up 439% net.
-
July 10: SK Hynix, one of the fund's largest positions, lists in the US. The listing sets off an unwind of leveraged Korean equity positions and the stock falls hard.
-
July 10 to July 20: the AI infrastructure complex sells off broadly. Nebius, SanDisk, Micron, and CoreWeave all drop more than 35% for the month. Some names are down 50% to 78% from recent peaks.
-
July 24: Aschenbrenner sends investors a letter reporting the 439% first half, calling the selloff one of the best buying windows since early 2025, and opening a subscription window for fresh capital on August 1. The letter includes the line, "At times we call out opportunities that seem like a particularly good time to add funds, if you have been waiting for one."
-
July 29 into July 30: the Financial Times reports the heavy losses. Once investors know, the market knows. Rival funds press the positions.
-
July 30: Goldman Sachs, JPMorgan Chase, and Bank of America issue margin calls. The public book goes to Citadel before the open.
BTIG's chief market technician, Jonathan Krinsky, measured Morgan Stanley's sector-neutral Momentum Index falling 17.4% in four trading days, the worst such decline on record. Worse than the dot-com unwind, worse than the pandemic shock, worse than 2022. He called it the largest and fastest momentum crash in modern history.
The detail that makes this strange: the S&P 500 stayed near record levels the entire time. This was a factor crash rather than a market crash, and it hit one specific kind of portfolio. Situational Awareness was the most concentrated version of that portfolio in existence.
What did the borrowed money actually do here?
Leverage did not cause the losses. It took away the option to wait them out, which is a different and much worse problem.
Reporting put the fund's borrowing at up to four times, with CNBC describing some of its positions as carrying as much as 400%. That distinction is the piece most of the coverage skipped, and it matters. At four times leverage, a 25% decline in the underlying positions wipes out the entire equity contribution. If the whole book had been levered four times and the core names fell 35% to 47%, July would have been a total loss rather than the 67% the fund reported. The honest read is that the most concentrated positions carried the heaviest borrowing while the book as a whole sat below that ratio, which is likely why there was anything left at the end of the month.
The spiral works the same way at any ratio. Meeting a margin call means raising cash. Raising cash means selling the same falling positions, which pushes them down further, which triggers the next call. Once that starts, conviction stops being an input. Nobody at Goldman Sachs cared what Aschenbrenner believed about 2027.
The part that translates to any business is the loss of control over timing. Most operators have lived through a stretch where the call was right and the calendar was wrong, and the only thing that saved it was the ability to wait it out. Borrowing against a position is how you give that ability away, and the rules say your lender does not even have to call you first.
If you want the full mechanics, we broke down exactly how this works, with real numbers on both the upside and the downside, in How Leverage Works in the Stock Market.
Did Citadel engineer the Situational Awareness collapse?
No, and the theory falls apart in four places. This is the most viral claim about the collapse, so it is worth walking through properly.
The theory goes like this: on July 27, Citadel talked up a surprise Fed rate hike. On July 29 the market sold off, and Situational Awareness got margin called. On July 30 Citadel bought the whole book at a discount. Rug pull.
The first piece is true. Frank Flight, who runs macro strategy at Citadel Securities, published a note on July 27 saying he now expected a hike at the July meeting. Everything after that breaks down.
There are two different Citadels. Citadel Securities is a market maker that buys and sells for other people. Citadel is the hedge fund. They are separate firms.
Flight was not alone. PGIM and Wrightson ICAP also called for a hike. Bond veteran Harley Bassman wanted one twice as large.
The fear predated the note. Bloomberg tied the selloff to oil prices rising after renewed US and Iran tensions, plus a labor market that refused to cool.
The Fed did not hike. At Kevin Warsh's first meeting as chairman on July 29, the FOMC voted 9 to 3 to hold the federal funds rate at 3.5% to 3.75%, with Lorie Logan, Beth Hammack, and Neel Kashkari dissenting in favor of a quarter-point rise. That was the first time since September 2016 that three policymakers dissented in the same direction, which tells you the pressure to raise was real and sitting inside the Fed rather than manufactured by a trading firm.
There is also a simpler tell. One block trade does not move an entire chip index. Microsoft reported its best quarter in years on Wednesday night, printing 43% Azure growth against a 40% consensus and guiding to 45% next quarter. Chip stocks jumped the following day, with the main semiconductor index closing up more than 8%, its biggest single-day gain in about fifteen months. Earnings moved that tape.
Citadel buying a distressed book is not a conspiracy. It is the thing Citadel is famous for. It took Amaranth Advisors' energy portfolio in 2006 and Sowood Capital's book in 2007, in both cases while everyone else was retreating.
What else are people getting wrong about the collapse?
Three claims keep circulating that the reporting does not support.
"The fund lost $35 billion of investor money." Start with the figures, because they are the source of most of the confusion. The Wall Street Journal reported the fund at around $20 billion in assets under management in recent months. CNBC reported it reaching roughly $45 billion by the start of July. The most recent 13F disclosed 29 US-listed long positions worth $5.52 billion. Those numbers are not contradictions. They were reported at different moments, by different outlets, on different bases: an assets under management figure from the spring, an asset figure at the July peak after a 439% first half, and a quarterly snapshot of long US-listed positions from a filing dated December 31, 2025. Subtracting one from another is how you get to a headline claim that $35 billion of investor money evaporated, and no reporting supports that claim.
"Situational Awareness is shut down." It is not. Aschenbrenner told investors the fund was not liquidated. It retains its private holdings, most significantly a stake in Anthropic valued at roughly $5 billion, plus positions in chipmaker MatX and AI data center startup FluidStack. Bloomberg reported that Aschenbrenner had tried to sell private stakes ahead of the Citadel deal; a firm spokesman told CNBC that reports the Anthropic stake was being marketed were "not accurate." Both of those are on the record, and they do not agree, which is worth knowing before you repeat either one.
There is a structural reason the Anthropic stake survived while every public position did not. A private stake cannot be margin called. Prime brokers lend against securities with observable daily prices, and a private company has no daily price to mark against a threshold. Every position that could be called was called. The one that could not be was not.
"He lost everything." By the fund's own unaudited numbers reported after the unwind, it was down 67% month to date and up 80% year to date. Check that math yourself: up 439% through June is a multiple of 5.39. Take 67% off and you land at roughly 1.78, or up about 78% on the year. The reported 80% is consistent with that. A brutal month sitting on top of a spectacular first half is still a good year on paper, which is not the story anyone is telling.
Was the AI thesis wrong?
The thesis was not the problem. Many of the fund's core names were up well over 100% on average even after July's selloff, and the fundamental backdrop got stronger during the crash rather than weaker.
Microsoft's Azure grew 43% and guided higher. Jensen Huang told Bloomberg in late July that the chip industry needs to grow roughly tenfold to serve a world running 100 billion AI agents. At AMD's Advancing AI event in late July, Lisa Su put the company's 2030 server CPU market at over $200 billion, with agentic AI named as the driver, a sharp increase from where AMD had it a year earlier. Chief executives do not move a market forecast like that unless the demand is showing up in real conversations.
Tae Kim, who writes Key Context, put it plainly: had Aschenbrenner used no leverage, or even a moderate amount instead of four times, he would be having a great year and his fund would rank in the top 1% for performance.
That is the part worth sitting with. He was right about the technology, right about the bottleneck, right about the specific companies, and right about the timing of the buying window. He simply did not own the positions anymore when the rebound came.
What happens to Situational Awareness now?
The fund continues as a private investment vehicle built around the Anthropic stake. Anthropic was last valued at $965 billion in a Series H round in May and filed confidentially for an IPO on June 1, with reporting suggesting a listing as soon as October. If that IPO prices well, a windfall on the private book could offset a meaningful share of the public market losses.
Aschenbrenner has also engaged existing investors and lenders about raising fresh capital, and per the Financial Times, some investors have been offered the option to buy assets out of the portfolio.
Whether the AI infrastructure names recover is a separate question with three variables the market is watching, per Barclays head of US equity strategy Venu Krishna: financing uncertainty, corporate capital expenditure expansion, and pressure on big tech free cash flow. None of those resolve because one forced seller left the building.
Frequently asked questions
What is the Situational Awareness hedge fund?
Situational Awareness LP is an AI-focused hedge fund founded in late 2024 by Leopold Aschenbrenner, a former OpenAI researcher, and named after his 2024 essay of the same name. It invested in the physical infrastructure behind AI, including chips, memory, data centers, power, and bitcoin miners, while betting against software companies it viewed as vulnerable to AI disruption.
How much did Situational Awareness lose?
The fund's assets fell from roughly $45 billion at the start of July 2026 to about $10 billion after the forced sale, per Bloomberg. In performance terms, Aschenbrenner reported unaudited returns of negative 67% for July and positive 80% for the year to date, following a 439% net return in the first half of 2026.
Did Situational Awareness shut down?
No. It sold its entire public equity portfolio to Citadel but retains its private holdings, including a stake in Anthropic valued at roughly $5 billion, and continues to operate as a private investment vehicle. Aschenbrenner told investors the fund was not liquidated.
How much leverage was Situational Awareness using?
Reporting put it at up to four times, with CNBC describing some positions as carrying as much as 400%, meaning about $4 of market exposure per $1 of capital. At four times leverage a 25% decline wipes out the entire equity cushion, so the fact that the fund finished July down 67% rather than down 100% suggests the book as a whole was levered below that peak ratio.
Did Citadel cause the collapse?
No. Citadel Securities, a separate firm from the Citadel hedge fund, published a note on July 27 expecting a Fed rate hike, but other forecasters made the same call, the selloff had independent drivers including oil prices and labor data, and the Fed ultimately voted 9 to 3 to hold rates steady. Citadel has a long documented history of buying distressed portfolios from forced sellers, including Amaranth in 2006 and Sowood in 2007.
Was Aschenbrenner's AI thesis wrong?
The evidence says no. Many of the fund's core positions were up well over 100% on average even after the July selloff, and AI demand signals from Microsoft, Nvidia, and AMD strengthened during the drawdown. Analysts covering the collapse have consistently pointed to borrowing rather than stock selection as the cause.
What is a margin call?
A margin call is a demand from your broker for more collateral when the equity in a margin account falls below a required minimum. If you cannot meet it, the broker can sell your positions without notifying you first and does not have to let you choose which ones. We cover the full mechanics in How Leverage Works in the Stock Market.
Interested in learning more?
A few related reads from Online Queso:
-
How Leverage Works in the Stock Market: The Real Math on Gains and Losses
-
The AI Jobs Wipeout That Wasn't: Why Big Tech Flipped the Story
-
StubHub CEO and Mass Scalping: What SEC Filings Reveal About Eric Baker and Andro Capital
-
How Tupperware Built an Iconic Product That Led to Its Business Collapse
Final Thoughts
I am not a hedge fund manager and none of this is investment advice. What I am is an operator who has spent a decade watching people confuse being right with being able to survive long enough to collect on it.
The thing that sticks with me is the July 24 letter. He told his investors the selloff was one of the best buying opportunities in over a year and asked them to send more money by August 1. He was correct. Chip stocks bounced hard within a week. He just was not holding the positions anymore when it happened, because six days after he wrote that letter, three banks decided the answer was no.
I move sports cards, and that market has ugly stretches like everything else. When a card I hold drops 40%, I put it back in the box and look at it in two years. Nobody calls me. Nobody liquidates my shoebox before the opening bell to protect their own exposure. That ability to simply wait is worth a lot more than most people account for, and the moment you borrow against a position you have handed it away.
Leopold Aschenbrenner wrote the document that convinced an entire market to buy AI infrastructure, then got carried out of the trade he invented four weeks before the rebound. The essay was not wrong. The portfolio was not wrong. The financing was.
