Leopold Aschenbrenner is 24 years old. Last year his fund managed as much as $45 billion. Last month it came apart.
The story is not that he was stupid. He was early and largely right about artificial intelligence, which is more than most professionals managed. The story is what he did with being right.
The rise
He came out of the FTX Future Fund, worked at OpenAI, and was fired in 2024. In 2021 he had put roughly $500,000 into Anthropic. He later sold that stake to Jane Street for about $1.3 billion.
That is an extraordinary result, and it is exactly the kind of result that persuades a young man he has a system rather than a hit. Money followed. His fund, named Situational Awareness after an essay he wrote, grew to as much as $45 billion.
The fall
The portfolio was concentrated and it was levered. Concentrated means a small number of positions. Levered means borrowed money is doing part of the work.
Together those two words describe every fund that has ever gone from celebrated to gone in a single quarter. When the AI names sold off, the borrowed portion of the position had to be repaid regardless of what he believed about 2030.
| The unwind | Figure |
|---|---|
| Peak assets managed | up to $45B |
| Anthropic stake sold to Citadel | $3.5B |
| Terms of that sale | at a discount |
| Regulator involvement | SEC inquiry |
Citadel bought his $3.5 billion Anthropic stake at a discount. Nobody sells the best thing they own at a discount by choice. That single line tells you the fund was meeting demands for cash, not making decisions.
The Securities and Exchange Commission is now investigating. That is an inquiry, not a finding, and it should be read as neither an accusation nor an exoneration.
The lesson that transfers
Almost nothing about a $45 billion hedge fund resembles a retirement account. One thing does.
Being right about a trend and wrong about the size of the bet still ends at zero. The market can be exactly as smart as you think it is and still take a year to agree with you, and borrowed money does not wait a year.
Aschenbrenner’s view of artificial intelligence may well be vindicated in the end. He simply will not be there for it, because the position was structured so that he had to be right on a schedule.
What we actually do
We hold the artificial-intelligence theme through companies with real customers and real cash, and we hold them at weights set by a checklist rather than by conviction.
We use no leverage in client portfolios. None. That is not caution for its own sake — it is the single rule that converts a bad year into an inconvenience instead of an ending.
The uncomfortable part of this story is that the man was not a fraud or a fool. He was a good analyst with a bad position size. That failure mode is available to anybody.