One of the AI boom’s biggest trades became a forced unwind

Leopold Aschenbrenner built one of the AI boom’s most spectacular trades on the belief that increasingly capable AI would require enormous spending on chips, memory, data centers and power. But after AI-linked stocks fell, his fund sought new capital and sold its public-stock portfolio to Citadel, Axios reported on July 30. Accounts of FT and WSJ reporting said Citadel took most, not all, of the roughly $16 billion public book.

The reversal was enormous. A post relaying Wall Street Journal reporting said Situational Awareness lost 67% in July but remained up 80% for the year. The crash followed extraordinary gains rather than erasing the entire run.

Two different figures describe the fund’s scale. It had recently reported about $20 billion under management. CNBC, citing people familiar with the matter, said assets peaked at about $45 billion earlier in July and cited reports of leverage as high as 400%.

The most consequential details come largely from unnamed sources. Public filings confirm selected holdings, but not the fund’s complete portfolio, financing or sale terms.

A long forecast met short-term financing

A May regulatory filing offers direct evidence of the infrastructure bet: Situational Awareness owned 12,410,060 shares of Nebius Group as of May 19, equal to 5.6% of its Class A shares.

An account of Financial Times reporting said the fund had borrowed against billions of dollars in AI-linked stocks. As prices dropped, lenders’ collateral lost value, creating pressure to raise money or sell. The fund was also hurt when short bets against software companies such as Adobe moved against it.

The result shows the risk of financing a long-term forecast with borrowing that lenders can reassess daily. A fund may be forced to sell before its underlying thesis has time to prove right or wrong.

Aschenbrenner’s 2024 essay argued that it was “strikingly plausible that by 2027, models will be able to do the work of an AI researcher/engineer.” Axios noted at the time that the wider expert consensus did not share his view that AGI would inevitably emerge from today’s generative-AI models.

Situational Awareness reportedly retained Anthropic and other private stakes. A firm spokesman told CNBC on Thursday that reports it was marketing its Anthropic stake were not accurate, while accounts of a proposed deal relied on unnamed sources. The reported $3.5 billion transaction was abandoned.

In plain terms, Situational Awareness made a bet that could take years to pay off using financing that could unravel in days.