At the end of June, a two-year-old hedge fund had gained 439% in six months and grown to more than $20 billion. Four weeks later, it was selling the bulk of a public stock portfolio valued at about $16 billion.

The immediate cause was a brutal reversal in the artificial intelligence trade. Situational Awareness, founded by former OpenAI researcher Leopold Aschenbrenner, lost about 67% in July as shares across its concentrated portfolio of AI infrastructure companies tumbled.

But falling stocks alone do not explain why the retreat happened so quickly. The missing link is leverage.

Situational Awareness had borrowed from Wall Street banks to make its positions larger. That amplified the gains while AI stocks were rising. When they fell, it reduced the value of the collateral supporting those loans and left the fund under pressure to find fresh capital or sell assets.

Its decade-long investment thesis had suddenly acquired an overnight deadline.

After urgent talks with potential investors late on Wednesday, 29 July, the fund agreed within 24 hours to sell most of its public stock portfolio to Ken Griffin’s Citadel.

Prime brokers including Goldman Sachs, JPMorgan Chase, Bank of America and Citigroup helped arrange the transaction. Citadel acquired the portion financed with money borrowed from the brokers, according to Reuters.

The Financial Times described it as one of the largest rushed stock transactions in Wall Street history. For Citadel, the turmoil created an opportunity to acquire the holdings from a seller that could no longer afford to wait.

“We let you down this month,” Aschenbrenner wrote in an investor letter seen by The Wall Street Journal.

Despite the July loss, Situational Awareness remained up about 80% for the year. That is what makes the episode more instructive than the ordinary collapse of a bad investment.

The fund may yet prove right about the enormous demand for chips, power and data centers created by AI. Its mistake was financing that long-term conviction with borrowing that could not survive a short-term reversal.

A 10-year thesis meets an overnight deadline

Aschenbrenner became prominent after publishing “Situational Awareness,” a widely read series of essays arguing that artificial general intelligence could arrive as early as 2027.

His central investment idea followed naturally. If AI capabilities were about to improve rapidly, technology companies would require far more computing power, memory, data centers and electricity than markets anticipated.

Situational Awareness invested across that physical supply chain. Its publicly disclosed positions included power provider Bloom Energy, data-storage company Sandisk, cloud operator CoreWeave and bitcoin miners that could repurpose their electricity connections and sites for AI computing.

A regulatory filing covering the end of March reported $13.7 billion of US-listed securities and options exposure, up from $5.5 billion three months earlier. The filing also showed large put positions tied to Nvidia, Broadcom, AMD, Oracle and a semiconductor exchange-traded fund.

Such filings offer only a partial and delayed picture. They do not reveal every short position, swap, foreign holding or financing arrangement. Still, they showed how much of the fund revolved around a single economic transformation.

That concentration produced exceptional returns while AI-related shares rose. It also meant that positions that looked different on paper could fall together when investors began questioning valuations, debt-funded data-center construction and the likely returns from AI spending.

Leverage changed the clock

A long-term investor can wait through a steep fall if there is enough cash and no lender demanding additional collateral. A leveraged fund may not have that choice.

Situational Awareness used money borrowed from prime brokers to increase the size of its positions. Leverage magnified the gains that attracted investors, including trading firm Jane Street, Stripe founders Patrick and John Collison, and technology investors Nat Friedman and Daniel Gross. It also reduced the fund’s ability to withstand a sudden reversal.

Reuters reported that the fund was under pressure either to raise capital or sell its public portfolio. It chose the sale. Citadel acquired the portion of the portfolio financed with broker leverage.

It remains unclear whether Situational Awareness received formal margin calls before the transaction. It would be inaccurate to state as fact that margin calls caused the sale. Economically, however, the fund had lost control of the timetable. Its need for capital had become more urgent than its ability to wait for prices to recover.

Aschenbrenner compared the pressure to a “bank run” and said traders had moved against stocks publicly associated with the fund. Public knowledge of concentrated positions can certainly make an unwind more difficult. Other investors may sell first, short the same securities or wait for a distressed seller to accept lower prices.

But that possibility is itself part of portfolio risk. A fund cannot assume that crowded holdings will remain liquid precisely when everyone wants to leave.

Citadel bought time

Citadel approached the same portfolio from the opposite position. It had capital, diversified operations and no immediate need to sell.

The terms of the transaction were not disclosed, although the Financial Times reported that the holdings were sold at a discount. Citadel did not need to endorse every part of Aschenbrenner’s AI thesis. It needed to determine whether the assets were worth more than the price demanded by a seller running short of options.

This is what forced sales do. They transfer assets from an investor constrained by time and financing to one able to wait.

Selling the portfolio in a negotiated block also probably reduced the danger of a disorderly liquidation in the open market. That mattered because Situational Awareness was not alone. Global hedge funds were suffering their largest monthly drawdown on record, while Asia-focused fundamental long-short funds had lost an average of 18.6% through 28 July, Reuters reported.

When leveraged funds sell simultaneously, falling prices generate further losses, which prompt lenders to demand more collateral and force still more selling. The original question of what an asset is worth can temporarily become secondary to who needs cash first.

The AI thesis survives

Situational Awareness has not disappeared. Reuters reported that it retained a portfolio of roughly $10 billion after the Citadel transaction, including private investments such as its Anthropic stake. The fund did not sell that holding.

Aschenbrenner also told investors that the firm would continue investing in public equities but would stop borrowing from banks to magnify its positions.

That makes the episode a forced reset rather than a conventional hedge-fund collapse. Whether the fund rebuilds will depend partly on its remaining investments and partly on whether investors trust its new approach to risk.

The wider lesson is narrower than declaring that the AI boom is over or that Aschenbrenner’s thesis has been disproved. Situational Awareness may ultimately be right that AI will require a historic expansion of computing and energy infrastructure.

Its mistake was allowing the financing of that conviction to become less durable than the thesis itself.

An investor can correctly identify the defining technology of a decade and still lose the ability to own it. In markets, being right about the destination is not enough. The portfolio must survive the journey.