If you think AI can reliably print money

What we know

Aschenbrenner's pitch was never subtle: AGI is coming fast, and the market hasn't priced it in. So his fund went heavy on AI-related equities, concentrated positions, and probably a lot of leverage. For a while that works — AI names were the whole show. But when a trade is crowded and levered, the path down is always steeper than the path up. A sharp drawdown in the sector, maybe a margin call or forced liquidation, and a $45B book turns into pocket change.

I don't have a number for what "most of it" means, and maybe the real figure is fuzzy. But the pattern is familiar: a smart guy with a strong worldview builds a giant position, the market moves against him, and the risk management was either too late or never existed.

The uncomfortable lesson

This is a case study in why "AI-driven fund" doesn't mean "AI makes good decisions." The model might be great at reading sentiment or scanning 10-Ks, but the portfolio construction, the sizing, the deleveraging under stress — that's still humans, and humans get stubborn when they're winning.

There's also a deeper irony here. Aschenbrenner keeps talking about AGI being an existential race, fast and dangerous. Yet he ran a fund like the singularity was guaranteed to arrive before any unpleasant volatility. Turns out the market doesn't care about your timelines.

Should anyone care?

If you're building an AI workflow for trading — even a small personal bot — the lesson is the same: backtest the drawdowns, not just the returns. Cap your leverage. And don't marry your thesis so hard that you can't sell when the tape says something else. A $45B blowup is just a scaled-up version of what happens to retail traders who refuse to cut a loser because "the AI said it's a buy."

I'm not dunking on the guy. Having the conviction to put billions behind a prediction takes guts. But the next time someone tells you AI hedge funds are the future, remember that the future arrived, and then it margin-called itself.