Aschenbrenner's Fund Was Up 439% and Closed in Four Days
Thursday, July 3010 Min Read
Situational Awareness closed the first half of the year up 439% after fees. Four days later every listed position it owned — long and short alike — was sold to Citadel in a single block. What broke was not the thesis. It was the way the thesis had been financed.
439%
First-half 2026 Return
$45B
Peak Assets in Early July
~4x
Position Size Versus Own Capital
96 Hours
From Investor Letter to Forced Sale
In This Piece7 Sections
On Wall Street, the clearest sign a fund is finished is its portfolio changing hands in one block. Nobody announces it; one morning, before the bell, everything moves to one buyer.
That is what happened on July 30. Every listed position held by Leopold Aschenbrenner's Situational Awareness LP — the longs and the shorts — went to Ken Griffin's Citadel.
Six months earlier, the same fund had closed the first half of the year up 439% after fees.
From Manifesto to Fund: The Birth of a Thesis
Leopold left OpenAI in 2024 and published "Situational Awareness," an essay resting on one claim. Artificial general intelligence is far closer than consensus believes, and its arrival will set off the largest wave of physical infrastructure investment in history.
Its power was turning an abstract forecast into things you could actually buy. If AI accelerates you need chips, chips need memory, both need data centers, and data centers need power. Not a think piece — a shopping list.
The fund launched in late 2024 with roughly $225 million, backed by Stripe founders Patrick and John Collison, Nat Friedman, Daniel Gross and — unusually — Jane Street.
Inside two years, assets passed $20 billion; a source told CNBC the peak was $45 billion by early July. Leopold was 25.
Two Fragilities: Concentration and Leverage
A 439% return does not happen by accident. Reaching that number takes not just being right but pressing very hard on what you are right about.
First, almost everything sat in a handful of names. The top five were more than three-quarters of the disclosed portfolio: SK Hynix, Nebius, CoreWeave, Micron, SanDisk and Bloom Energy.
Second, he did it with borrowed money. The fund carried gross exposure of roughly four times its capital: $4 of stock for every $1 of equity.
A third layer: the fund was short software — the "long the chip, short the software" trade. The idea was that AI would compress software earnings while blowing out infrastructure demand. It was the most talked-about trade of 2026.
The real issue is not the size of the loss but who owns the decision to sell. Unlevered you choose; on borrowed money the broker does, at the worst price.
July: The Month the Thesis Was Tested
In July, AI infrastructure stocks sold off, and it stopped looking like an ordinary correction.
The decline did not hit the whole market; it hit exactly where the fund stood. Core positions lost between 27% and 54%.
In the same week, shorts such as Adobe rose. Both sides lost at once.
At four times leverage, a 30%-plus decline in the core holdings consumes the equity. Goldman Sachs, JPMorgan Chase and Bank of America issued margin calls.
The Fund That Ended in Four Days
What followed, per the Financial Times and CNBC, ran as these collapses do: slowly, then all at once.
Timeline
- July 24Leopold writes to investors. He acknowledges that the fund has been hit by the selloff but argues this is one of the best buying opportunities since the start of 2025. A note at the end of the letter: new capital will be accepted on August 1.
- MidweekThe fund approaches existing investors and lenders for fresh money. Some are offered the chance to buy positions directly out of the portfolio.
- July 29Bank of America, Goldman Sachs and JPMorgan begin shopping the fund's entire book to potential buyers.
- July 30, pre-marketEverything is sold in one block. The buyer is Citadel.
The third line is where the story ends. Once a book is known to be shopped by its banks, the price moves against it: everyone knows those shares have to be sold.
The price was not disclosed. Figures like 40 cents on the dollar circulated online without a verified source.
Why the Buyer Was Citadel
Griffin's firm has played this role many times; standing opposite a forced seller is a Citadel model.
The gap between the fundamental value of the positions and the seller's time pressure is the buyer's profit.
Two readings, both defensible:
| Reading | Claim | Weak Point |
|---|---|---|
| Bearish | AI infrastructure is in such bad shape that an entire fund was wiped out | What killed the fund was not the stocks, it was the leverage wrapped around them |
| Bullish | Citadel thought these shares were worth owning, so the problem was never the shares | Citadel bought at a discount, and "cheap" is not the same as "valuable" |
The only certainty sits between the two: the largest forced seller in the market is now gone.
What Is Left
Situational Awareness is not shutting down. Its private holdings remain, the largest a stake in Anthropic the Financial Times values at roughly $5 billion.
Not a coincidence; the reason is technical:
Prime brokers calculate collateral against assets that carry a price every day. A private company has no daily mark, and therefore never enters the margin calculation.
The borrowed money killed precisely the listed book. Nothing else entered the calculation.
What is left is no longer a levered equity fund but an investment firm holding mostly Anthropic stock.
A spokesperson denied reports that the Anthropic stake was up for sale; CNBC had reported the fund was trying to sell it. The sources conflict.
The Lesson: You Do Not Choose When You Sell
Leopold's idea was not disproven this week; whether he is right about AI infrastructure demand the next several years will answer. What collapsed was not the thesis but the way it was financed.
The second lesson applies even to those who never borrow: AI stocks are very different companies, but the market increasingly trades them as a single stock.
When a memory maker, a server rental company and an energy firm all fall the same day, the same direction, by similar magnitudes, your portfolio is not diversified. You bought six companies; you own one sentence.
Leopold wrote the most widely read investment text of the AI era. He also sat the most expensive exam that idea has faced.
This article is based on reporting by the Financial Times, CNBC, The Wall Street Journal and Bloomberg as of July 30, 2026. Much of that reporting relies on unnamed sources, and no party has issued an official statement. Figures such as the fund's assets under management and the transaction price differ between sources.
Correction (September 13, 2026): prose tightened for rhythm; figures and sources unchanged.
Companies in This Piece
Since The Event