Background
Situational Awareness LP, the hedge fund founded by former OpenAI superalignment team researcher Leopold Aschenbrenner (born 2001), collapsed in July amid a historic AI-sector selloff and margin-call cascade. On July 30, the fund sold most of its public equity portfolio to Ken Griffin's Citadel. The fund had gained 439% in the first half of the year, with assets swelling to roughly $45 billion — before nearly reaching zero within a month.
The Founder and the Strategy
Aschenbrenner, a German-born prodigy, entered Columbia University at 15 and graduated first in his class at 19. He joined OpenAI's superalignment team, was dismissed in 2023 over a security memo, and then published a 165-page essay predicting AGI would trigger a compute-infrastructure boom by 2027, winning backing from top investors. He founded the fund in September 2024 with about $225 million in seed capital from the Stripe founders, Nat Friedman, Daniel Gross, and Jane Street.
The strategy was simple but aggressive: high leverage (~4x) concentrated on AI infrastructure "picks and shovels" suppliers, long names including SK Hynix, Micron, Sandisk, Nebius, CoreWeave, and Bloom Energy, while shorting software stocks such as Adobe.
The July Collapse
- A deep global chip-stock correction in July hit the long book hard: Sandisk fell 56% and Nebius 48%.
- Software stocks rebounded on valuation repair (Adobe rose against the trend), so the short book also lost money.
- Combined with ~4x leverage, the fund suffered a double-sided loss.
- A major trigger came from South Korea: leveraged retail investors heavily concentrated in AI names faced mass margin calls, sparking cascading selling that dragged down global AI assets.
- In the early hours of July 30, the Federal Reserve's "hawkish pause" (holding rates at 3.50–3.75%, with three governors favoring a hike) was read as a prelude to renewed tightening; chip stocks were hammered late in the session, and rapid declines triggered broker margin calls from Bank of America, Goldman Sachs, and JPMorgan.
- On July 30, the fund sold its public equity portfolio to Citadel in a single deal worth over $10 billion.
- The more consensus the theme, the greater the crowding risk: when everyone bets the same way, a liquidity inflection becomes a stampede.
- Leverage magnifies mistakes, not just returns: 4x leverage turned a 439% annualized gain into a 67% single-month loss.
- Long/short is not a natural hedge: when the shorted software names rose on valuation repair, the "hedge" became a second wound.
The Ironic Timing
Just six days before the blowup (July 24), Aschenbrenner told investors in a letter that the volatility represented "the best buying window since early 2025" and invited additional capital by August 1. His July 30 letter to LPs admitted a ~67% loss for the month, writing "we let you down this month" and comparing the experience to a "bank run."
The fund then sold at what proved to be the bottom: the next day, chip stocks rebounded sharply — Sandisk +25.99%, CoreWeave +21.51%, Bloom Energy +26.49%.
What Survived
The fund retained its private holdings, the largest being roughly $5 billion in Anthropic equity (not sold). A slimmed-down team of about 20 continues operating as a private investment company with small-scale public stock trading. Elad Gil publicly said he had invested in the fund for the first time.
Takeaways: A Warning for Crowded AI Trades
The person who wrote the "defining case" for the AI bull market was himself forced out by leverage. His long-chips, short-software thesis was the year's most crowded trade, pushed to an extreme with borrowed money, turning one bad month into a crisis. Lessons for quantitative and discretionary strategies alike:
Sources
1. Bloomberg Businessweek, 7/31/2026 2. 21st Century Business Herald, 7/31/2026 3. Dealroom, 8/2026 (Citadel sale details) 4. Yahoo Finance HK, 7/31/2026