How a 24-Year-Old Blew Up a $45 Billion AI Fund

Leopold Aschenbrenner's Situational Awareness LP, a hedge fund he grew from $225 million to about $45 billion, lost most of its value in about four weeks in July 2026 after a 439% net return in the first half of the year, due to a concentrated bet on AI infrastructure and roughly four-times leverage. Ken Griffin's Citadel bought the fund's public portfolio in a matter of hours. Aschenbrenner, a 24-year-old former OpenAI researcher, had founded the fund in 2024 to bet on his thesis that AI infrastructure companies would win.

In the summer of 2026 a 24-year-old with no prior track record running money watched a hedge fund he had grown from $225 million to roughly $45 billion lose most of its value in about four weeks. Leopold Aschenbrenner’s Situational Awareness LP had posted a 439% net return in the first half of the year on a single, ferociously concentrated idea: that the companies building artificial intelligence infrastructure would keep winning. When that trade cracked in July, the fund’s use of roughly four-times leverage turned an ordinary drawdown into an existential one, and Ken Griffin’s Citadel bought the wreckage of its public portfolio in a matter of hours. The mainstream story is a morality tale about a wunderkind who flew too close to the sun. The more useful story is about leverage, concentration, and who is standing ready to profit when a levered bet unwinds. Get your headlines checked for spin. This is exactly the kind of story most mainstream coverage waves through. The Rubbish Talk app runs the day’s news through our Rubbish-meter and scores each headline for spin, from 1 the plain, unspun fact to 10 pure rubbish , so you can see what is being downplayed before it moves the market. Download it on the App Store: Rubbish Talk News https://apps.apple.com/us/app/rubbish-talk-news/id6797190131 . The boy who wrote the AI manifesto Leopold Aschenbrenner was born in Germany in 2001 or 2002, the son of two physicians, and attended the John F. Kennedy School in Berlin, a bilingual German-American public school. He enrolled at Columbia University and graduated in 2021, at 19, as valedictorian, with a degree in economics and mathematics-statistics. While there he co-founded Columbia’s effective altruism chapter, and he later did research connected to Oxford’s Global Priorities Institute. This is not the biography of a career trader. It is the biography of a very young, very credentialed true believer. His path into AI ran through the effective-altruism world that also produced Sam Bankman-Fried. Aschenbrenner joined the FTX Future Fund in early 2022 and left before FTX’s November 2022 bankruptcy. In 2023 he joined OpenAI’s Superalignment team, the group led by Jan Leike and Ilya Sutskever that was supposed to figure out how to control AI systems smarter than humans. In April 2024 OpenAI fired him, citing an alleged information leak that Aschenbrenner has characterised as a benign brainstorming document. He contends the real trigger was a security memo he had sent the board. One month after his dismissal, the Superalignment team dissolved entirely. Two months after that, in June 2024, Aschenbrenner published “Situational Awareness: The Decade Ahead,” a 165-page essay arguing that artificial general intelligence could arrive around 2027 and that superintelligence would follow. The essay went viral in tech and policy circles. He then did the thing that turned an argument into a wager: he opened a hedge fund of the same name to bet real money on the thesis. Backers reportedly included Stripe’s Patrick and John Collison, former GitHub chief Nat Friedman, investor Daniel Gross, and, unusually, the quantitative trading firm Jane Street, which rarely backs outside managers. One idea, maximum conviction The strategy was the essay, expressed as a portfolio. Aschenbrenner went long the companies he believed would capture the economics of the AI build-out chipmakers, memory producers, data-center power and compute providers and short the software businesses and incumbents he expected AI to disrupt. It was a coherent, aggressive macro view executed with very little diversification. Public filings give a sense of just how concentrated. As of its second-quarter 2026 13F, the fund’s roughly $10.6 billion U.S.-listed long book was dominated by a single position: Micron Technology, at about 46% of the disclosed portfolio. Behind Micron sat Bloom Energy at roughly 13.7%, Taiwan Semiconductor at 10.7%, and a cluster of smaller compute-and-power and crypto-mining names: Core Scientific, Applied Digital, STMicroelectronics, Iris Energy, Sharon AI, Riot Platforms and CleanSpark. Reporting on the blow-up added positions that a U.S. 13F would not capture, including a large stake in South Korea’s SK Hynix and holdings in the neocloud providers CoreWeave and Nebius, plus shorts against software names such as Adobe. The through-line was singular: everything was a bet on the same AI-infrastructure trade, financed and re-financed to press the advantage while it worked. And through the first half of 2026, it worked spectacularly, returning 439% net after fees according to an investor letter dated July 24. The trade that ate itself The detail that turned a good year into a wipeout was leverage. Situational Awareness ran its public equity book at roughly four-times leverage, borrowing against its positions through three prime brokers: Goldman Sachs, JPMorgan Chase, and Bank of America. That is the mechanism that both manufactured the 439% gain and guaranteed the fund could not survive a sharp reversal. The arithmetic is unforgiving. At four-times leverage, an investor controls four dollars of exposure for every dollar of real equity. A 25% fall in the underlying positions is therefore enough to erase the entire equity cushion. The AI-infrastructure stocks the fund was long did not fall 25%. In the space of about a month, several fell far more: SK Hynix dropped roughly 47%, Nebius around 43%, and CoreWeave about 36%, with Bloom Energy and other names in the same 35-to-47% range. Positions of that size, held with borrowed money, do not simply lose value; they trigger margin calls, demands from the prime brokers for more collateral against loans that are now underwater. Aswath Damodaran, the New York University valuation professor, framed the lesson bluntly in his own analysis of the collapse: a maximal-conviction macro bet on AI, combined with maximal leverage, is a time bomb. Without the borrowing, he argued, the fund would have taken a bad month and lived. With it, a bad month was fatal. Four weeks in July The unwind had a clear arc. The fund peaked around June 19, 2026. The pressure began building on July 10, when a large initial public offering of SK Hynix stock coincided with leveraged holders being forced to sell, and the stock rolled over. By mid-July, credit markets were flashing warnings of their own, with the cost of insuring the debt of AI-infrastructure borrowers rising sharply. By late July the fund faced a squeeze from every direction at once: long positions were falling, short positions such as Adobe were moving against it, prime brokers were demanding more collateral, and no fresh capital was arriving to plug the hole. Public investors had begun to doubt the core premise, worrying aloud that enormous AI capital spending was not translating into near-term revenue. Aschenbrenner tried to avoid a forced sale. He invited existing investors to add capital, setting a deadline of August 1, and shopped the portfolio to other firms. Both Millennium Management and Jane Street looked at the book and declined the concentrated risk. In an investor letter, Aschenbrenner framed the episode in the language of survival. “We took the steps that were necessary to fight another day. But our fund must always be structured such that we can take a loss and fight another day.” Leopold Aschenbrenner, Situational Awareness investor letter, as reported by Bloomberg August 2026 By the numbers, the damage was severe: the fund’s public equity holdings lost more than two-thirds of their value, and overall principal losses exceeded 43%. Assets that had touched roughly $45 billion at the peak fell to around $10 billion. Somewhere near $35 billion of notional value had evaporated in about four weeks. Citadel rides to the rescue, again On July 30, 2026, Ken Griffin’s Citadel bought most of Situational Awareness’s public equities portfolio, a book reported at around $16 billion, at a discount of roughly 10%. The deal reportedly closed within 24 hours of Griffin personally speaking with Aschenbrenner, after Griffin pulled in his senior leadership, including co-chief investment officer Pablo Salame and chief operating officer Gerald Beeson, for an overnight review. Structuring it as a single-buyer purchase rather than a piecemeal liquidation was deliberate: it kept other traders from front-running the unwind of individual positions, which would have driven prices down further. Several of the stocks involved, including CoreWeave and Bloom Energy, rallied once news of the transaction removed the overhang of forced selling. This is a playbook Citadel has run before, and the pattern is the real story. In 2006 it absorbed the natural-gas positions of Amaranth Advisors after that fund lost about $6.4 billion. In 2007 it took over the portfolio of Sowood Capital during the early tremors of the financial crisis; Sowood’s founder, Jeffrey Larson, said afterwards that Citadel had offered the only workable exit. “Citadel offered the only immediate and comprehensive solution.” Jeffrey Larson, founder of Sowood Capital, on Citadel’s 2007 purchase of his fund’s portfolio as reported In 2021, Citadel and Steven Cohen’s Point72 injected $2.75 billion into Melvin Capital during the GameStop short squeeze; Melvin shut down in 2022, and Citadel reportedly avoided a loss on the arrangement. Each time, the framing in the press is “rescue.” The mechanics are closer to opportunistic distressed buying by the one player with the balance sheet, the risk systems, and the speed to price a complex book overnight. A struggling fund gets an exit and avoids a disorderly collapse; Citadel gets assets at a discount during a moment of maximum seller weakness. Both things are true at once, which is precisely why the strategy works so well for Griffin. One manager on the receiving end of a Citadel rescue call once likened the experience, as recounted in coverage of the Situational Awareness deal, to seeing the Valkyries coming and hearing the Grim Reaper at the door. What he kept, and what he is doing now The most important fact about the collapse is what did not collapse. Situational Awareness held a private stake in Anthropic, reported at around $5 billion and built starting at roughly a $60 billion company valuation in early 2025, along with private positions in AI infrastructure names such as Fluidstack and MatX. Those holdings survived the margin crisis intact for a simple structural reason: private positions have no daily market price, so prime brokers cannot margin-call them. The public book, marked to market every day, was the part that could be seized. The private book, illiquid and unpriced, was untouchable. That distinction may prove enormously valuable. Anthropic is reportedly targeting an IPO around October 2026 at a headline valuation near $965 billion, following a large funding round, with some accounts speculating an even higher figure. A stake acquired near a $60 billion valuation would, on paper, be worth many times its cost at those levels. In other words, the leveraged public trade that nearly destroyed the fund was arguably the less valuable half of the portfolio all along. Aschenbrenner appears to understand this. Within days of the fire sale, on August 5, 2026, Situational Awareness closed a $400 million private investment in an undisclosed company, following a $100 million commitment the prior month, and he told investors the environment offered some of the most attractive opportunities since early 2025. There was no new fund and no formal wind-down. He is, in his own framing, living to fight another day, and pivoting from the public markets that nearly killed him toward the private bets that survived. What everyone should actually take from this The tempting lesson is the personal one: an overconfident 24-year-old with a viral essay and no trading scars got humbled. That is real, but it is not the point. The point is that the machinery around him did exactly what it was built to do. Prime brokers extended four-times leverage against a wildly concentrated book because the fees and the returns were good while the trade worked, then called the collateral the instant it did not. The largest hedge fund in the world stood ready to buy the forced seller’s assets at a discount within a day. None of that is a malfunction. It is the system operating as designed, and it will operate the same way for the next fund that mistakes a bull market for genius. There is also a signal here for anyone watching the broader AI trade. Situational Awareness was a pure, levered expression of the “AI infrastructure always wins” thesis, and it was the first high-profile casualty of the market beginning to ask whether the capital spending will ever earn its return. The stocks that broke it, the memory makers and neoclouds and power providers, are the same names propping up the major indices. The fund’s implosion did not require a market crash to happen; it happened in an otherwise calm tape, felled by its own leverage and concentration. That should be the uncomfortable takeaway. If a modest wobble in AI-infrastructure stocks can vaporise $35 billion at one fund, the question is not whether Aschenbrenner was reckless. It is how many other bets are structured the same way, and who will be there to buy the wreckage next time. Sources: Leopold Aschenbrenner: biography and career https://en.wikipedia.org/wiki/Leopold Aschenbrenner Citadel buys Situational Awareness portfolio, 4x leverage ends AI fund’s run TechTimes https://www.techtimes.com/articles/322285/20260730/citadel-buys-situational-awareness-portfolio-4x-leverage-ends-ai-funds-1000-run.htm The Situational Awareness Fund Blow-up Aswath Damodaran https://aswathdamodaran.substack.com/p/the-situational-awareness-fund-blow AI hedge fund Situational Awareness may have sold its public portfolio but still has its Anthropic shares TechCrunch https://techcrunch.com/2026/07/30/ai-hedge-fund-situational-awareness-may-have-sold-its-public-portfolio-but-it-still-has-its-anthropic-shares/ Aschenbrenner makes quick return after near-collapse of Situational Awareness Yahoo Finance https://finance.yahoo.com/markets/stocks/articles/aschenbrenner-makes-quick-return-near-143931633.html Situational Awareness makes $400M private bet Quartz https://qz.com/situational-awareness-aschenbrenner-400-million-private-investment-080626 Billionaire Ken Griffin just saved Situational Awareness Yahoo Finance https://finance.yahoo.com/markets/stocks/articles/billionaire-ken-griffin-just-saved-200506637.html Citadel’s crisis investing playbook back in focus after AI hedge fund rescue Hedgeweek https://www.hedgeweek.com/citadels-crisis-investing-playbook-back-in-focus-after-ai-hedge-fund-rescue/ With Situational Awareness AI deal, Citadel’s Griffin rides to the rescue again Reuters via WTVB https://wtvbam.com/2026/07/31/with-situational-awareness-ai-deal-citadels-griffin-rides-to-the-rescue-again/ Can the Situational Awareness hedge fund raise capital after its 439% H1 gain? Disruption Banking https://www.disruptionbanking.com/2026/07/30/can-the-situational-awareness-hedge-fund-raise-capital-after-its-439-h1-gain/ Situational Awareness LP portfolio holdings StockCircle 13F https://stockcircle.com/portfolio/situational-awareness Situational Awareness LP portfolio Quiver Quantitative 13F https://www.quiverquant.com/institutions/Situational%20Awareness%20LP/ Anthropic targets IPO by October 2026 after $965B valuation Crypto Briefing https://cryptobriefing.com/anthropic-targets-ipo-by-october-2026-after-965b-valuation/