NextFin News - Ken Griffin's Citadel has offloaded more than 80% of the aggregate risk in the roughly $16 billion stock portfolio it bought from the collapsed hedge fund Situational Awareness, completing one of the fastest controlled de-risking exercises in recent hedge-fund history. The unwind, finished within about three weeks of the late-July acquisition, stands in deliberate contrast to the forced liquidation that brought down Leopold Aschenbrenner's fund — and it underscores a rule that leveraged markets teach at regular intervals: the price of the same assets depends less on what they are worth than on who holds them, and how they were financed.
The Rescue and the Unwind
Citadel said on August 21:
"To date, working closely with our trading partners, we have successfully shed more than 80% of the aggregate risk from the original portfolio."The firm sold more than $4 billion of the positions. The original portfolio — the public equity book of Situational Awareness — was valued at approximately $16 billion when Citadel absorbed it at a discount of around 10% in the final days of July.
The deal was a rescue of sorts. Situational Awareness, the AI-focused hedge fund founded in early 2025 by former OpenAI researcher Leopold Aschenbrenner, had accumulated well over $20 billion in assets under management and posted a 439% net return through the first half of 2026. Then July happened. Concentrated, leveraged bets on AI infrastructure names — memory chips, data centers, power equipment — reversed sharply. The fund lost 67% in a single month. Prime brokers including Goldman Sachs, JPMorgan Chase, and Bank of America pressed for collateral. Aschenbrenner sold the bulk of his public holdings to Citadel and retained a multibillion-dollar private stake in Anthropic.
A 13F filing as of June 30 showed just how concentrated the bet was. SanDisk and Micron Technology together accounted for more than 56% of the fund's disclosed U.S. long holdings — $5.7 billion and $5.6 billion respectively. The next three positions were Bloom Energy (9.38%), Taiwan Semiconductor (6.25%), and Nebius Group (6.09%). The fund also held $745 million of CoreWeave. In July, SanDisk fell nearly 47%, Micron dropped roughly 29%, Bloom Energy slid 32%, Nebius fell 31%, and TSM lost 15%.
The strategy itself was not subtle: own the picks and shovels of the AI buildout, and short the software companies Aschenbrenner believed AI would disrupt. It worked spectacularly until it didn't.
Why Citadel Could Sell Where Aschenbrenner Could Not
The difference between the two outcomes is not stock selection. It is balance-sheet structure and the option to sell slowly.
Situational Awareness ran a concentrated, leveraged book inside a fund that had grown from zero to roughly $45 billion in about two years. When the AI trade turned, the fund faced margin calls — demands for more collateral that forced selling into a falling market. Forced selling is the most expensive kind: it is price-insensitive, it telegraphs urgency to the other side of the trade, and it compounds the very losses it is meant to contain.
Citadel entered from the opposite position. It had the cash, the credit lines, and — critically — the choice of when to sell. The firm's senior team, including co-chief investment officer Pablo Salame, chief operating officer Gerald Beeson, head of equity quantitative research Perry Vais, and chief legal officer Shawn Fagan, pulled an all-nighter with Griffin to analyze the book before agreeing to take it. That preparation let Citadel buy at a roughly 10% discount and then unwind on its own timetable. Shedding 80% of aggregate risk in roughly three weeks on a $16 billion package is not a distressed fire sale; it is a controlled exit executed from strength.
The lesson is structural, not cyclical. Leverage does not change the fundamental value of an asset, but it changes the holder's ability to wait for that value to be recognized. A leveraged owner is a price-taker at the worst possible moment. An owner with deep liquidity and no forced seller behind it is a price-negotiator.
A Twenty-Year-Old Pattern Repeats Itself
The Situational deal fits a pattern two decades in the making. In September 2006, Amaranth Advisors lost nearly $6 billion — about 65% of its assets — betting on natural gas prices. Citadel and JPMorgan Chase acquired Amaranth's energy portfolio at a $2.15 billion discount to the previous day's mark-to-market value. Twenty years later, Citadel and JPMorgan appear again on the same side of the same trade: the well-capitalized buyer absorbing a collapsed fund's book at a haircut while the original manager exits under duress.
That recurrence is the point. The specific asset changes — natural gas in 2006, AI infrastructure in 2026 — but the mechanism does not. Crowded positioning plus leverage plus a catalyst produces forced selling; forced selling produces a discount; and the discount is captured by whoever has the balance sheet to stand still. Citadel's business model includes an implicit option on market stress: when other managers are forced to sell, Citadel is positioned to buy. The Situational unwind is the latest exercise of that option.
The Discount Was the Margin of Safety
Buying $16 billion of assets at a 10% discount gave Citadel an immediate $1.6 billion cushion before any market move. That cushion served two purposes. First, it absorbed the bid-ask spread and market impact of unwinding a large, crowded book. Second, it let Citadel sell into strength. Once the deal was announced on July 31, the fear of a disorderly liquidation eased and the very AI infrastructure names Citadel had just bought rebounded sharply — Nebius and Micron recovered in the final days of July after the liquidation scare passed. The discount plus the rebound is what turned the trade.
Citadel's July performance shows the result. The flagship Wellington multistrategy fund returned 5.9% in July — its best month since 2022 — lifting its year-to-date gain to 12%. The Tactical Trading fund gained 11.1% for its best month ever and is up 27% for the year. The stand-alone equities fund advanced 14.2%, also a record month, and is up 27% year-to-date. Through June, Wellington had been up 5.7%, trailing rivals such as Point72 and Millennium. After the Situational deal, Griffin's firm moved to the top of the multistrategy leaderboard while many peers lost money in a chaotic July. Citadel managed about $71 billion in assets as of July 1 and declined to comment on the trade.
The Second-Order Effect: The Cost of Leverage Just Went Up
The first-order lesson of this episode is obvious: leverage killed a brilliant trade. The second-order question is what the market will conclude from it, and whether that conclusion is correct.
One reading is that the AI infrastructure trade is broken — that the memory, power, and data-center names were simply overvalued, and that the July rout was the beginning of a longer repricing. That reading is available, but it is not the one the Citadel unwind supports. Citadel did not hold the positions because it fell in love with SanDisk or Micron. It held them because it bought a package at a discount and needed to monetize it. The selling was inventory turnover, not a fundamental downgrade.
The more important second-order effect runs through the lender side. Bank of America CEO Brian Moynihan, whose bank was one of Situational's prime brokers, framed the episode plainly in an interview:
"These are all warning shots. Valuations get out, leverage in the system gets there. You have to be careful."The warning is not about AI. It is about what happens when a crowded, leveraged trade unwinds in a market that has grown used to smooth exits. Prime brokers that underwrote the leverage are now the ones managing the fallout. Expect tighter margin terms, lower concentration limits, and more frequent collateral calls for funds running similar crowded AI trades. The cost of leverage for the next Aschenbrenner just went up.
There is also a market-structure consequence worth naming. The July chip selloff alone wiped more than $1 trillion from the market value of the most valuable semiconductor stocks in a matter of days, with Nvidia, SK Hynix, Samsung Electronics, Micron, AMD, and TSMC each losing more than $100 billion. In a market that size, a single $45 billion fund with more than half its book in two names can move prices against itself simply by needing to exit. That is the definition of fragility: your own size becomes your counterparty.
Cyclical Drawdown, Structural Fragility
The AI drawdown itself is best read as cyclical. Valuations stretched, positioning crowded, and a catalyst — signs of progress in China's chipmaking, worries about the sustainability of AI spending — triggered a rotation. Cyclical drawdowns mean-revert when the underlying demand thesis is intact, and global AI capital expenditure has not been disproven by one bad month.
But the fragility exposed in Situational Awareness is structural. A fund that grows to $45 billion in two years, concentrates more than half its book in two memory names, and runs it on borrowed money has built a structure that cannot survive a normal volatility shock. That is not a market cycle; it is a design flaw. The same market that rewarded Aschenbrenner's 439% first half will always contain the possibility of a 67% month if the leverage stays. The two are the same phenomenon viewed from opposite ends.
The Counter-Thesis, and What Would Falsify This One
The strongest argument against reading this as a leverage story is that the AI trade genuinely broke. The bears point to the trillion-dollar chip wipeout, to China's advancing chip capabilities, and to the possibility that AI infrastructure spending has been pulled forward faster than the revenue to justify it. If that view is right, Citadel did not execute a clever risk-managed unwind — it simply got lucky exiting before a deeper fundamental repricing.
That case is real, but it does not survive the timing. Citadel shed 80% of the risk within three weeks of buying the book. A firm executing a fundamental exit does not move that fast on a $16 billion package; it re-evaluates each name, forms a view, and holds or trims accordingly. Citadel's pace is the pace of a risk manager closing out inventory, not of an investor changing its mind about AI. The falsifying signal would be evidence that Citadel retained the weakest names and sold the strongest — that its remaining 20% is concentrated in the positions with the poorest fundamentals. If the residual book skews toward the names with deteriorating earnings or order books rather than toward illiquid positions that simply took longer to exit, the "lucky exit" thesis gains force.
What Comes Next
The immediate beneficiaries of this episode are clear: Citadel, which converted a distressed purchase into a record month, and the surviving holders of AI infrastructure names, who got a reprieve from forced selling once the liquidation risk was absorbed. The exposed are the funds still running crowded, leveraged versions of the same trade with prime brokers who have now seen what can go wrong.
Short term, the market should treat Citadel's selling as mechanically driven and largely complete — the 80% figure means the overhang is mostly gone. Medium term, the question is whether the AI infrastructure names can grow into their valuations on actual earnings, not just capital-expenditure announcements. Long term, the structural lesson is that the next generation of AI-native funds will be financed on stricter terms, and that concentration without commensurate liquidity is a bet against volatility that eventually collects.
Watch three signals. First, the composition of Citadel's remaining 20% — whether it is the hard-to-sell or the weak-fundamental names. Second, prime-broker margin terms for concentrated AI funds, which Moynihan's warning suggests are about to tighten. Third, whether AI capital-expenditure announcements continue to convert into revenue at the memory and power companies that anchored the trade.
The market paid Leopold Aschenbrenner $45 billion to be right about AI. It then charged him 67% for being wrong about leverage. Citadel's lesson is cheaper and sharper: in a crisis, liquidity is not an attribute of the asset. It is an attribute of the balance sheet.
Explore more exclusive insights at nextfin.ai.
