Leopold Aschenbrenner Built a $45 Billion AI Hedge Fund. Then It Collapsed in Days.

There is a certain type of Silicon Valley story that seems to write itself. A young genius leaves a prestigious school in a contentious situation. Puts out a manifesto. Hundreds of millions of dollars are raised. Becomes the person that everyone either looks up to or looks down on. Then it turns into a lesson that everyone talks about at cocktail parties after one bad week.

That story is Leopold Aschenbrenner’s story, but it moved faster than most and has so many numbers that it’s hard to keep them all in your head.

Aschenbrenner was born in Germany to parents who were doctors. Most people said he was a strange kid. In German schools, he skipped grades and graduated from high school at age 15. He then went to Columbia University and graduated as valedictorian at age 19. Sofia Montrone, who was his salutatorian, later told CNBC that the only time she had met him was on Zoom before their 2021 graduation. She was honest and kind of charming when she said, “He was just some guy.” It turned out that he wasn’t a guy. Someone like him would write an undergrad paper called “Existential Risk and Growth” and really mean every word of it.

After Columbia, he moved around in the Effective Altruism community and worked for a short time under Sam Bankman-Fried at the FTX-affiliated Future Fund in a penthouse in the Bahamas. In 2023, he joined OpenAI’s Superalignment team. The last job didn’t go well. A hacker got into OpenAI’s internal systems, so Aschenbrenner sent a memo to the board warning that the company’s security wasn’t strong enough to stop foreign espionage. In 2024, OpenAI fired him because they thought he had improperly shared private information. He completely disagrees with the way he was described, saying that he was just trying to do the right thing. Scott Aaronson, a computer scientist who had worked with him, told CNBC that the firing seemed like it was too harsh.

Aschenbrenner wrote and published a 165-page essay that would shape the next part of his life a few weeks after he left. The essay, which was only called “Situational Awareness,” described how artificial general intelligence could be used within the next few years and argued that governments, markets, and institutions were greatly overestimating how soon this could happen. Some people thought it was visionary. Some people said it was too sure of itself. People read it either way. People then gave him cash.

By July 2024, he was said to have raised $225 million in seed money from investors Daniel Gross, founders of Stripe Patrick and John Collison, and Nat Friedman, who used to be CEO of GitHub. As you might have guessed, the fund was called Situational Awareness. It made a clear and strong case: the coming AI boom would need a huge amount of chips, memory, data centers, and power infrastructure. So, Aschenbrenner bet, and for a while, it looked like he was right. Before it went down this month, the fund was said to have made more than 1,000% since it started.

It turned out that the problems were built in from the start. Reports say that the fund used up to 400% of its own money in loans to boost those returns. In a market that is going up, that kind of leverage is not inherently risky. On the other hand, it is very sensitive to any long-term downturn. In July 2026, AI infrastructure stocks like SK Hynix, CoreWeave, Nebius, and SanDisk started falling sharply. This made the fund’s losses grow in ways that mirrors couldn’t fix. The direction of short positions against software companies like Adobe also went in the wrong direction. Then came the margin calls. Goldman Sachs, JPMorgan, and Bank of America all sent prime brokers to help handle the mess.

Situational Awareness’s assets dropped from $45 billion to about $10 billion by the end of July. Ken Griffin’s Citadel bought all of its public equity books. This was one of the biggest emergency stock transfers Wall Street has seen in years. It was said that the deal was completed in less than 24 hours. A Wall Street coach named Jerry Diao said it best: “A lot of people saw this blow-up as a matter of not if, but when.”

Take a moment to think about what really makes this story hard to understand. That long-term idea that Aschenbrenner had about AI might still be true. Citadel’s choice to buy those same assets shows that at least one of the smartest companies in finance sees value in them. The collapse isn’t always a bad sign for the idea; it’s usually a bad sign for the position size. Being right about the direction of a ten-year technological shift doesn’t mean you can handle a 400% correction that lasts three weeks.

Experience is another thing to think about. Right from the start, critics pointed out that Aschenbrenner had never managed money professionally before he started Situational Awareness. His time at FTX before, even though it was short and had nothing to do with the company’s fraud, made things look bad. People who supported him said that his knowledge of AI was unique and real, and that he understood technology better than most fund managers. It looks like both are true. Even if you are very sure of what you believe, you might not be ready for how risk management works at a large scale.

As of this writing, Aschenbrenner is still said to be bullish. He told investors in a letter sent at the end of July that the correction was the best time to buy in over a year. He’s letting people make new capital contributions before August. The fund is still there, but it’s smaller and hurt. At this point, it doesn’t look like investors will take him up on his offer.

Another fact that the financial press has mentioned almost as an aside is that the fund went down during Aschenbrenner’s wedding weekend. He is going to marry Avital Balwit, who is the chief of staff to Anthropic CEO Dario Amodei. There’s something almost novelistic about the fact that the wedding and the margin calls happened in the same 72 hours. This is the kind of detail that makes you wonder what will happen next in this story and whether Aschenbrenner’s predictions, even though they’ve been wrong this month, will turn out to be correct in the end.

Quick Reference Table