Leopold Aschenbrenner, the former OpenAI researcher who left the Superalignment team in 2024 amid a leak dispute he denies, ran a hedge fund called Situational Awareness with about $20 billion in assets, according to investor James Wang, writing at Weighty Thoughts. In July, the fund’s losses grew severe enough that Citadel purchased its stock portfolio, a detail Wang draws from Wall Street Journal reporting on the episode.
Wang writes that Aschenbrenner was reportedly trading with roughly four times leverage in AI-adjacent public stocks through July, including losses in neocloud, memory chip and datacenter power names. Those losses were compounded by short positions in software stocks tied to what Wang calls the “SaaSpocolypse” trade, which moved against the fund at the same time.
According to Wang, the fund was margin called and its holdings were ultimately liquidated to Citadel, a detail he says points to weak risk controls rather than a mistaken read on artificial intelligence. He does not disclose a specific dollar figure for the losses themselves, only that they were large enough to force that outcome.
This is the distinction Wang’s piece turns on, and it is worth stating plainly. A leveraged bet blowing up is a risk management outcome, and by itself it says little about whether the underlying thesis, that transformative AI is close and its shape already visible, was correct. A trader running four times leverage who gets margin called can still be right about a market’s direction and wrong only about its timing. That confusion is one of the oldest ways to lose money.
Wang, who describes his own background as a former hedge fund manager with an AI background, compares the episode to Long-Term Capital Management, the 1998 fund staffed with Nobel laureates and veteran bond traders that quadrupled investor money over four years before collapsing so completely that the Federal Reserve had to organize a Wall Street rescue. His point is that being the smartest people in a room offers no protection against catastrophic position sizing, and that finance history is full of credentialed experts who lost fortunes anyway.
Wang traces the failure back to Aschenbrenner’s own words. In the 2024 essay that made his name, Aschenbrenner wrote that AGI’s arrival felt “extremely visceral” to him, adding: “Sure, going all-in leveraged long Nvidia in early 2023 has been great and all, but the burdens of history are heavy. I would not choose this.” Wang argues that this same conviction, carried directly into a leveraged and time-bound bet, is the more instructive story than the losses themselves. The failure was not believing AI was moving fast. It was sizing a fund around exactly when.
Wang extends the critique past one fund, arguing that frontier AI lab culture more broadly carries an assumption that deep expertise in AI transfers cleanly to other fields, citing unrelated cases of lab-affiliated teams assuming they could out-perform specialists in areas like materials science. He is careful to add that he is not an AI skeptic and expects the technology to broadly benefit ordinary workers.
Wang’s framing holds up under scrutiny. Leverage and timing, not the AGI thesis, are what verifiably failed here, and collapsing the two into one story lets easy schadenfreude stand in for actual analysis of where AI is headed. But the episode still carries a real warning for anyone investing around an AI thesis: strong conviction about a technology’s direction says nothing about the size or timing of the bet that should follow from it, and the people best positioned to judge where a technology is going are not automatically the people best positioned to price it.
For any investor building a portfolio around AI conviction, the lesson is to separate the research from the risk management, because getting the second one wrong can wipe out a position even when the thesis behind it later proves correct.
James Wang detailed the fund’s collapse and its broader lessons for AI lab culture in a July 31, 2026 piece for his newsletter, Weighty Thoughts.
