How the Best-Performing Hedge Fund in the World Became a Fire Sale

What the collapse of Situational Awareness tells us about leverage, conviction, and the difference between being right and surviving long enough to prove it.

Source: Josh Edelson

On the Morning of July 30th, 2026 Ken Griffin's Citadel absorbed the entire public equities portfolio of Situational Awareness LP in a single block trade before markets opened. The portfolio was valued at approximately $16 billion, and Citadel picked it up at a roughly 10% discount. The fund that had been up over 1,000% since inception––Wall Street's most-watched AI trade––was a forced seller by breakfast.

Who Is Leopold Aschenbrenner

Leopold Aschenbrenner founded Situational Awareness after being fired from OpenAI in 2024 over an alleged information leak that he disputes. Before that, he had written a 165-page essay, also called Situational Awareness, that became required reading in Silicon Valley, arguing that artificial general intelligence was closer than most people believed and that the physical infrastructure required to build it would be the defining investment of the decade.

He put his money where his manifesto was. The fund launched in late 2024 with roughly $225 million and grew to $45 billion at its peak by early July 2026. Early backers included Stripe co-founders Patrick and John Collison and former GitHub CEO Nat Friedman. Through June 2026, the fund was up 439% on a net basis for the first half of the year alone, and more than 1,000% since inception.

By any measure, it was one of the greatest runs in hedge fund history.

How It Unraveled

The fund's strategy was straightforward: own the companies building AI infrastructure (chips, data centers, power) and bet against the software companies most vulnerable to AI disruption. Top long positions included SK Hynix, Nebius, Micron, and CoreWeave. The short book targeted names like Adobe.

What made the strategy fragile was the leverage sitting underneath it. The fund was running approximately four times leverage; meaning for every dollar of equity, it had borrowed three more. When AI hardware stocks fell sharply in July, two things happened simultaneously. The long positions lost value, weakening the collateral behind the fund's borrowing. At the same time, the short positions moved the wrong direction as software shares rose. Many of the fund's top AI longs fell 40 to 50% in the month while its software shorts climbed against it.

By Thursday morning, Goldman Sachs, JPMorgan Chase, and Bank of America, the fund's three prime brokers, had issued margin calls on the leveraged portfolio, and Aschenbrenner was out of options. Six days earlier, he had sent investors a letter describing July's selloff as some of the most attractive buying opportunities since early 2025, inviting them to commit fresh capital by August 1st. That capital never arrived. What remains after the Citadel transaction is a $5 billion stake in Anthropic and a firm that will operate going forward as a private investment vehicle.

The portfolio ultimately fell 67% in July. Aschenbrenner later admitted that the fund had come "closer to permanent capital impairment than is acceptable to us," comparing the feedback loop around its positions to a bank run.

The Market's Response

The immediate aftermath was telling. The same AI infrastructure stocks that had been collapsing for weeks surged 15 to 25% within hours of the liquidation news. Situational Awareness had been the most visible forced seller in the market, and once that pressure lifted, buyers stepped back in.

But that rebound does not necessarily prove Aschenbrenner's AI thesis was right. If the fund's own forced selling helped push its holdings downward, their recovery after the seller disappeared tells us partly about market mechanics, not just fundamentals.

So July tells us less about Aschenbrenner's thesis than it first appears. The fund became large enough that its liquidation influenced the prices being used to judge whether its investments were correct. In other words, Situational Awareness did not simply receive the market's verdict; it became part of the verdict.

What Citadel did next makes that even clearer. By August 21, it had already shed more than 80% of the aggregate risk acquired from Situational Awareness, completing nearly 100 block trades worth more than $4 billion. Citadel wasn't necessarily betting that Aschenbrenner was right. It may simply have seen a distressed seller and a 10% discount.

The Oldest Lesson on Wall Street

Financial history has seen this pattern before. Long-Term Capital Management collapsed in 1998 with positions that were fundamentally sound but leveraged beyond what a bad month could withstand. Archegos blew up in 2021 the same way. In each case, the underlying idea wasn't the problem––the structure around it was.

Situational Awareness adds one important wrinkle. Its collapse did not prove Aschenbrenner right or wrong about AI infrastructure. It proved that a thesis measured in years had been placed inside a financing structure measured in days. At roughly four times leverage, the fund did not merely need to be right eventually; it had to survive every move along the way.

The prime brokers were part of that structure too. Bank of America CEO Brian Moynihan, whose bank financed Situational Awareness, later called episodes like this "warning shots" and said lenders typically respond by tightening underwriting standards. The leverage was Aschenbrenner's choice, but the terms on which Wall Street supplied it helped determine how long he could wait for his thesis to play out.

The Anthropic stake is what's left, and it survived precisely because it was private. No margin call mechanism, no prime broker, no forced sale. Anthropic raised at a $965 billion post-money valuation in May 2026, making that $5 billion holding potentially the most valuable asset the fund ever owned. The long-term conviction is still intact. The structure built around it wasn't.

Situational Awareness may eventually be proved right about AI infrastructure. But July did not answer that question––leverage prevented the fund from staying in the experiment long enough to find out.

At four times leverage, eventually is not a time horizon.