You can be right about the future of artificial intelligence and lose everything anyway. Use enough debt and the market can force you to sell long before your thesis has time to play out.
Published on August 1, 2026 · by Alex · Hedge funds · Leverage · Artificial intelligence
Situational Awareness LP launched in late 2024 with $225 million and grew to manage around $45 billion. Its founder, Leopold Aschenbrenner, was not yet twenty-five and had come from OpenAI, which he left in 2024 in circumstances the company described as improper disclosure of information and which he has always disputed.
The thesis was simple and turned out to be right: artificial intelligence would grow for years, so you had to be in chips, memory and data centres. It compounded more than 1,000% since launch and 439% net in the first half of 2026 alone.
In July it lost 67% and ended up selling its entire public portfolio. This article is about why being right was not enough.
The fund ran roughly four times leverage on its public positions. While the market rose, that multiplied the gains — hence the spectacular figures.
The trouble with leverage is that it does not distinguish direction. A single shift in sentiment was enough for the same mechanism to run in reverse, and in July the sector moved violently.
| Reference | Decline |
|---|---|
| Philadelphia Semiconductor Index | −28.6% from the June 22 peak |
| Morgan Stanley Momentum TMT Index | −53.5% |
| The fund, over the month | −67% |
With losses mounting, the fund wrote to investors describing the selloff as among the most attractive opportunities since early 2025, and inviting fresh capital commitments before August 1.
The capital did not arrive in time. Six days later, on July 30, Goldman Sachs, JPMorgan and Bank of America issued margin calls. With no way to meet them, selling stopped being a decision.
Precision matters here, because this involves a named company and facts nobody has proven.
What was reported is this: days before the collapse, Citadel Securities — the group's market-making arm — publicly raised the possibility that the Federal Reserve might surprise with a rate hike. That scenario was not priced in, and AI-linked stocks fell.
Days later, after the forced liquidation, Citadel — the hedge fund — bought the entire public portfolio, by then at steep discounts. And in the sessions that followed the sector rebounded hard; South Korea's Kospi rose as much as 18%, recovering nearly all of the week's decline.
The sequence is striking and has been widely discussed in the financial press. None of it demonstrates a coordinated strategy or market manipulation. These are two distinct entities within the same group, a public opinion on monetary policy is perfectly legitimate, and buying assets in a forced liquidation is routine and legal. What remains is an uncomfortable sequence, not an accusation.
What matters in this story is not who bought. It is that a correct thesis and too much debt are incompatible.
Artificial intelligence probably will grow for years. That fund was probably right. But leverage introduces a condition the thesis does not account for: it is not enough to get the destination right, you have to survive the journey. And with four times borrowed capital, the margin for being wrong about timing is effectively zero.
The fund sold its public book but kept its private positions — among them a roughly $5 billion stake in Anthropic, taken at a valuation near $60 billion. That part does not get liquidated on a bad day because it does not trade, which is both its protection and its limit.
And the question that matters for the rest of the market remains: the sector has now seen how an over-leveraged fund ends. What it does not know is how many more are running the same structure.
The events described come from financial media reporting in late July 2026. The fund's figures come from those same sources.
It launched in late 2024 with $225 million, grew to manage around $45 billion, and returned more than 1,000% since inception, including 439% net in the first half of 2026 alone. In July it lost 67% and ended up selling its entire public portfolio.
Because it ran leverage of close to four times its capital in the public positions. Leverage does not care about direction: it multiplied gains while the market rose and worked just as fast in reverse when the sector moved violently in July.
The moment selling stops being a decision. On July 30, Goldman Sachs, JPMorgan and Bank of America demanded additional collateral; unable to post it, the portfolio was sold out of obligation. The letter to investors asking for fresh capital was dated July 24 — it did not arrive in time.
This article is the written version of the Saturday analysis.
Watch the video on TikTok · Follow the channel
