Translated into English by KANSEI · original in Italian · Click here to read the original
AgentsEconomy·August 25, 2026 · 08:58 AM

Leopold Aschenbrenner had written that the decade belonged to AI. Then he lost 67%

Situational Awareness managed more than thirty billion with leverage reaching four hundred percent. In July it lost sixty-seven percent in a month and sold off the portfolio to Citadel in twenty-four hours.

Charlie
Charlie
Tech & AI
Leopold Aschenbrenner. Photograph via Reuters Connect, processed by KANSEI.
Leopold Aschenbrenner. Photograph via Reuters Connect, processed by KANSEI.Illustration KANSEI
Share

Situational Awareness is a San Francisco hedge fund founded two years ago by Leopold Aschenbrenner, who is now twenty-four: he graduated from Columbia at nineteen, top of his class, then passed through FTX, Sam Bankman-Fried's cryptocurrency exchange that subsequently collapsed, and OpenAI, which fired him on charges of leaking confidential information. In June 2024 he published "Situational Awareness: The Decade Ahead", a sprawling essay that set two precise dates: human-level artificial intelligence by 2027, superintelligence by 2030. The fund was born from the reception that essay received in the Valley, and took its name from it.

Aschenbrenner's bet came down to two moves. On one side, buy the infrastructure providers: chipmakers, companies that build and lease data centers, companies that sell the power. Among the positions the fund later found itself forced to sell were SK Hynix and CoreWeave. On the other, short the traditional software companies — the ones a language model should render obsolete. It is one of Wall Street's most crowded trades of the past two years, and it has serious industrial logic: while nobody knows which model will win, everyone needs the same chips and the same megawatts.

At its peak the fund managed more than thirty billion dollars, and for every dollar of its investors' money it had borrowed up to four: leverage of four hundred percent, according to CNBC. By June it had risen more than a thousand percent since inception, according to the Wall Street Journal.

Then in July semiconductors fell. The long positions lost value, and at the same moment the traditional software stocks — the ones sold short — rose. Both legs of the bet broke at once. The banks called for more cash, the fund couldn't find it, and sold the bulk of its equity portfolio to Citadel, Ken Griffin's hedge fund, at a discount of roughly ten percent, in a deal closed in twenty-four hours. July ended at minus sixty-seven percent. Citadel then resold more than eighty percent of that portfolio in nearly a hundred block trades worth more than four billion, and its flagship fund gained six percent.

Lending to hedge funds has become one of Wall Street's most profitable businesses, and as long as prices go up nobody measures how long the rope they've handed out actually is. When prices fall, lenders need to get their money back fast, and getting it back fast means forcing a sale: it's that same forced selling that pushes the price down further, even for people who never had a cent in that fund. Here it stopped at one fund's portfolio, and Citadel was there to buy it.

On August 24 the Securities and Exchange Commission, the authority that oversees American financial markets, sent a subpoena — that is, an order to produce documents — to Bank of America, Citi, Goldman Sachs and JPMorgan Chase. The four banks had been financing the operations of Situational Awareness, the hedge fund that nearly blew up in July. The SEC requested the precise timing of those trades and communications between the fund and its lenders regarding the borrowed money, and ordered the banks to preserve all the material. The fund has not been accused of anything, and the inquiry is at its earliest stage.

Some argue that nothing particularly anomalous happened. The fund survived — something Long-Term Capital Management in 1998 and Amaranth in 2006 failed to do. It still holds stakes in private companies, among them Anthropic, the lab behind Claude, which according to the New York Times is aiming to go public at a valuation that could reach two trillion dollars. Aschenbrenner took "full responsibility", said the fund would continue to operate and would stop borrowing from banks, and in early August had already put four hundred million into a private company. Two weeks after nearly blowing up.

When evaluating a bet on artificial intelligence, it's worth asking two numbers: how much borrowed money is holding it up, and who lent it. Both are in a public document.

Read the whole issue

The editorial, the video and every article, on a single page.

Open the issue