No Kiddie Pool

Leopold Aschenbrenner, a 25-year-old former OpenAI employee and Columbia valedictorian, founded the hedge fund Situational Awareness after writing a widely read 2024 essay series arguing AI would grow to enormous scale. The fund returned approximately 1000 percent over its first two years, compared to roughly 200 percent for an unlevered version of the same portfolio, meaning the entire outperformance came from leverage rather than superior stock selection.

Aschenbrenner was heavily long memory chip companies and short software companies, and both positions moved against him simultaneously. Prime brokers issued margin calls and forced him to offload a large portion of his positions, with Citadel acquiring many of those public stock positions at a discount and likely recovering significant gains within approximately 12 hours as the trades rebounded. Assets under management fell to approximately 10 billion dollars, down from a peak that would have been worth roughly 80 billion dollars had the fund held through the volatility. Aschenbrenner still holds a significant Anthropic stake that was not subject to the forced selling.

Matt Levine identified the core structural failure as using short-term callable margin debt to fund long-term, highly volatile bets on AI markets. Established large hedge funds issue bonds to secure longer-term non-callable funding, which prevents forced selling during drawdowns. A fund with only a two-year track record receives less forbearance from prime brokers than an established one, and Situational Awareness compounded its fragility by continuously scaling up leverage as assets grew. Levine argued that Citadel's edge in this episode was superior management of funding risk rather than any superior insight into AI markets. He drew a parallel to Sam Bankman-Fried, who held approximately 8 percent of Anthropic funded with customer money from FTX, representing the most extreme form of short-term funding mismatch. That stake was sold during the FTX bankruptcy and would be worth approximately 100 billion dollars today. Aschenbrenner had worked at an FTX-related fund but borrowed from banks on normal prime brokerage terms rather than misusing customer funds, which Levine treated as a meaningful distinction.

Xoma sold itself to Lagann for 39 dollars per share in cash plus a contingent value right tied to ongoing litigation, with the CVR estimated to be worth approximately 5 dollars in expected value. The merger closed on July 14th. Xoma issued a press release stating the CVR record date was 5 p.m. on July 13th, the day before closing, but because stock trades settle on a T-plus-one basis, buyers on July 13th would not settle until July 14th, creating ambiguity about who was entitled to the CVR. NASDAQ told traders they needed to own the stock by Friday July 10th to receive it. The stock traded around 40 dollars on July 13th rather than 39 dollars, reflecting genuine disagreement among traders about their entitlement. Approximately two days after closing, Xoma issued a follow-up press release stating the original record date announcement was wrong and that the merger closing date governs. Traders who bought on July 10th and sold on July 13th expecting to retain the CVR appear to have kept it.

The underlying confusion in the Xoma situation was that the company applied dividend distribution mechanics, which require a formal record date and press release, to a merger CVR, which is automatically distributed to shareholders when the merger closes and requires no separate record date. Xoma had made a similar record date error on a prior biotech acquisition, and in that earlier case a press release simultaneously described a date as both the record date and the ex-date, which are contradictory concepts. Levine noted there is an established industry of former lawyers at hedge funds who specialize in finding and trading on exactly these kinds of document ambiguities, with analogous situations arising in credit markets over bond indenture language.

Morgan Stanley issued an 18-month structured note linked to SpaceX, which had already fallen below its IPO price roughly a month after going public. The note pays investors 140 cents on the dollar at maturity regardless of how much SpaceX rises, but if SpaceX falls more than 50 percent the investor loses proportionally, so a 60 percent decline on a 100-dollar investment returns only 40 dollars. Levine argued the actual function of the note from Morgan Stanley's perspective is to source very out-of-the-money puts on SpaceX from retail customers at favorable terms, because such puts are difficult to buy in normal markets. Sellers face catastrophic payouts in bad scenarios, and buyers face counterparty risk from sellers most likely to fail in exactly those conditions. Levine described structured notes generally as a mechanism by which banks start with options exposures they want to offload and then construct a retail narrative around them, with the prospectus disclosing the payoff structure but the practical question being whether any buyer reads it.