Situational Awareness was built around two bets. Long the physical layer of AI (chips, memory, data centres, power) and short enterprise software, on the reasoning that AI would eat the margins of companies selling seats and subscriptions. Both bets pointed the same way, which is what made the fund famous and what eventually killed it. On 30 July, Aschenbrenner sold the entire public book to Ken Griffin's Citadel in a single block trade, after Bank of America, Goldman Sachs and JPMorgan called for collateral he could not produce.
He was right for a long time first. An investor letter reviewed by the Financial Times reported net returns of 439% through June. Take the VanEck Semiconductor ETF as a proxy for the long leg and the SPDR S&P Software & Services ETF for the short one: between January 2025 and the end of July 2026, semis roughly doubled while software fell about 4%. The gap between them reached 189 points on 22 June.
Then the leverage: the fund ran at close to four times, which means the thesis had to be right continuously, not eventually. Between 23 and 31 July software rose 10% and semis fell nearly 7%: both legs moving against him inside six sessions. That is an ordinary bad week by historical standards. At 4x it triggers margin calls, and margin calls force selling, and forced selling moves prices further against you. The thesis may still be broadly correct, but it didn't matter.
Betting against software means surviving everyone else being wrong for as long as they choose to be wrong, and Aschenbrenner sized the position as if that period would be short. But of course the more interesting part is in the numbers: software finished the whole nineteen months roughly where it started, so the short never really worked at all. Every point of that 189-point gap came from the long side. And there's a decent case he underestimated what he was shorting. Enterprise software firms sell customer relationships, compliance, proprietary data, integrations, distribution, and switching costs that make replacement genuinely expensive, and they may end up among the bigger beneficiaries of AI by folding it into products nobody wants to remove.
Which leaves the third thing, the one that's easy to say afterwards and hard to do at the time: he never took enough off the table. After returns like that, concentration and leverage stop being strategy and start being exposure. Taking profits would have meant admitting the trade could reverse while it was still working, and nobody does that well. The fund reportedly still sits above where it started, and it still holds a stake in Anthropic worth around $5 billion, so this is not a wipeout.