AI hedge fund Situational Awareness may have sold its public portfolio, but it still has its Anthropic shares

Leopold Aschenbrenner built a $45 billion hedge fund on a single conviction: that the AI boom would run on silicon, memory, and electricity before it ever produced a chatbot. For six months the market agreed. Then it didn't. The selloff in SK Hynix, Sandisk, Bloom Energy, and Nebius Group wasn't a correction — it was a verdict. Public investors decided that the infrastructure layer of the AI stack is a capital pit with no near-term revenue floor. Aschenbrenner's leverage turned that verdict into a margin call.

Citadel didn't buy a portfolio; it bought a liquidation. Ken Griffin's shop has made a specialty of picking up the pieces when leveraged players unwind. The price Citadel paid effectively sets the market's new valuation on the entire AI infrastructure trade. That valuation is roughly 75% below the peak. Situational Awareness now sits on $10 billion, mostly in illiquid private positions. The public book is gone.

Aschenbrenner's July 24 letter to investors read like a man who still believes his own thesis. He called the crash one of the best buying opportunities since early 2023 and asked for fresh capital by August 1. The commitments didn't come. Jane Street, the Collison brothers, Daniel Gross, Nat Friedman — the smart money that seeded the fund — didn't ride to the rescue. Their silence is louder than any redemption notice.

The fund's origin story mattered. Aschenbrenner, valedictorian at 19, OpenAI superalignment team at 22, fired at 23 for leaking internal work. His former boss Ilya Sutskever left to start Safe Superintelligence. Fellow researcher Jan Leike jumped to Anthropic. Aschenbrenner launched Situational Awareness. The narrative was clean: the kid who saw the scaling laws before anyone else, now betting his own fund on them. Narratives attract capital. They also attract scrutiny when the trade turns.

The Anthropic stake is the tell. It's the only position Aschenbrenner didn't have to sell because he couldn't — it's private, unmarked, and tied to the same scaling bet that just got torched in public markets. If Anthropic's next model release flops, or if its compute economics don't improve, that stake becomes a mark-to-market nightmare with no Citadel to bail it out. If it hits, Aschenbrenner looks prescient again. Either way, the fund is now a binary bet on a single private company.

Hedge funds are supposed to hedge. Situational Awareness did the opposite: it levered a concentrated, thematic long book and rode the momentum until momentum reversed. The 439% return through June wasn't alpha; it was beta on steroids. The collapse isn't bad luck — it's the inevitable math of leverage applied to a thesis that the public market just rejected.

Citadel's purchase also signals something about the AI infrastructure trade: sophisticated distressed players see value at these levels. But they see it with their own risk controls, their own leverage limits, and their own exit plans. They didn't buy the fund; they bought the assets. Aschenbrenner keeps the Anthropic ticket and the management fee on a shrunken base.

The broader lesson is blunt. The market's patience for AI capex stories has a half-life measured in quarters, not years. Semiconductor cycles, memory pricing, energy project timelines — these move on industrial clocks. AI hype moves on product demos. When the two desynchronize, the leveraged get carried out. Aschenbrenner's fund is the first high-profile casualty. It won't be the last.

What remains is a 25-year-old with a private stake in the most important AI lab outside OpenAI, a track record that now shows both 439% up and 75% down, and a thesis he still hasn't abandoned. The public portfolio is Citadel's problem now. The Anthropic stake is Aschenbrenner's alone.