A Leverage Parable Dressed Up as an 80 Percent Return

Leopold Aschenbrenner's fund closed out the year up roughly 80 percent. That number will travel further than the story behind it, so the story deserves the first look.

Aschenbrenner, the former OpenAI researcher who built his public profile on a widely circulated essay about the pace of AI progress, has been running a fund called Situational Awareness, concentrated in AI and semiconductor names. Through July, the fund's book grew to roughly $45 billion, carried on 4x leverage.

Four times leverage has one specific property worth translating into plain terms. A 25 percent decline in the underlying book erases the entire capital base behind it. That kind of drawdown shows up in a volatile sector during an ordinary bad quarter, sometimes a bad month, not once a decade. Chip stocks sold off through late July, and the math did what the math does.

On July 30, before the market opened, the fund liquidated its entire public book, longs and shorts together, to Ken Griffin's Citadel in a single block trade. A firm unwinding both sides of a book at once, in one trade, before the opening bell, is a firm that has run out of room to maneuver. Margin calls don't ask for a graceful exit. They ask for whatever price the buyer on the other side of the phone is willing to give you, and Citadel is very good at being that buyer.

Strip out everything that happened next, and this is a clean case study in what leverage does to a portfolio during a drawdown. A public, liquid book, financed at 4x, came within a routine correction of a fund-ending event, and was sold at a forced discount to a rival before regular trading hours. That's the entire mechanism. Nothing about it is unusual. That's exactly the point.

What saved the year was a separate decision entirely. Aschenbrenner held a concentrated, illiquid stake in Anthropic that was up 620 percent. That single position was large enough to absorb the full loss from the forced liquidation and still leave the fund net positive for the year. Being early and heavily weighted into a position that appreciated more than sixfold is a real call. It deserves credit on its own terms, separate from anything that happened to the leveraged book sitting next to it.

Those are two different investments with two different risk profiles, and reporting them as a single annual return figure flattens a distinction that matters. One position rewarded conviction and patience. The other trade very nearly failed for the most ordinary reason a leveraged trade fails: the underlying assets moved against it by less than the leverage ratio required to survive.

David Sacks and Chamath Palihapitiya discussed the episode on All-In as a parable about leverage and concentration risk in the AI trade, and that's the correct lesson to draw from it. In the same breath, both of them are also carrying their own leveraged exposure to the AI trade. A parable is a comfortable frame for a story that could just as easily be a preview.

I'd extend the lesson one step further than the parable usually goes. The instrument that nearly failed here is ordinary. Leverage is a multiplier: it takes whatever the underlying asset does next and scales it, in whichever direction that turns out to be. At 4x, an investor doesn't need a crash. An ordinary correction in a sector that has already produced one 10 percent single-day semiconductor selloff this year, on a model release rather than a genuine fundamental shock, is enough on its own.

None of this argues against conviction investing in AI. The Anthropic position is the proof that concentrated, high-conviction bets can pay off at a scale that changes a fund's entire year. What it argues against is treating a leveraged trading book and a long-term concentrated position as the same kind of risk, just because they sit inside the same fund and get reported on the same line.

The next fund that gets margin-called on an AI trade probably won't have a position that happened to be up 620 percent sitting next to it to cover the gap. When that happens, the number that survives the year will look a lot less like an 80 percent return, and a lot more like what actually occurred here before the market opened on July 30: a forced sale, at whatever price the buyer offered, with no room left to negotiate.