Citadel Buys Situational Awareness Positions After AI Selloff

Key Takeaways

• Situational Awareness lost a significant share of its public equities portfolio in July, prompting a major restructuring of its holdings.
• Citadel acquired part of the portfolio as Leopold Aschenbrenner’s fund moved to close most positions.
• The unwind exposes the liquidity and crowding risks behind concentrated bets on AI infrastructure.

Situational Awareness is liquidating most of its public equity positions following a severe reversal in artificial intelligence stocks, turning one of the market’s highest-profile AI wagers into a broader test of thematic investing. The hedge fund, founded by 25-year-old Leopold Aschenbrenner, lost a significant share of its public equities portfolio during July, according to Reuters reporting carried by CTech.

Citadel moved quickly once speculation spread across Wall Street that a large fund was under pressure. Citadel's founder assembled senior executives to examine Situational Awareness’ holdings, liquidity, and risk exposure before approaching Aschenbrenner. Shortly after, Citadel purchased a portion of the portfolio.

The precise terms and composition of that transaction were not disclosed. Citadel’s involvement, however, gives Situational Awareness a large counterparty capable of absorbing positions while the fund reduces exposure. It also potentially allows Citadel to acquire assets at prices shaped by a forced seller rather than ordinary market conditions.

The fund's founder addressed investors directly, acknowledging the losses in a recent letter.

Rather than holding a loose collection of technology stocks, the Situational Awareness portfolio represented a tightly connected macro thesis. As AI models became more capable, the fund argued, developers would require far more GPUs, memory, electricity, and data-center capacity. Companies supplying those resources could therefore capture an unusually large share of AI-related investment.

That logic shaped holdings that included SK Hynix, SanDisk, Micron, and CoreWeave. Several major positions declined sharply during the July selloff, according to coverage from 36kr. When correlated holdings fall together, diversification by ticker provides limited protection because the underlying economic exposure remains similar.

The long-term demand argument has not disappeared. McKinsey estimates that generative AI could contribute between $2.6 trillion and $4.4 trillion in annual economic value across industries. IDC forecasts that global spending on AI software, hardware, and services will rise from $184 billion in 2024 to $509.2 billion in 2027.

But growth in end-market spending does not ensure that every infrastructure supplier, or every investor entering at a high valuation, earns the expected return. Timing matters. So do financing conditions, capacity additions, customer concentration, and the market’s assumptions about future margins.

Crowding can turn a routine correction into a rapid unwind when multiple funds identify the same suppliers as obvious beneficiaries. PitchBook found that dedicated AI and machine-learning funds expanded assets under management by more than 50% year over year, while Preqin has tracked double-digit inflows into systematic and macro strategies tied to AI and data-center themes. Hedge funds collectively manage roughly $4.1 trillion, giving crowded positioning the potential to amplify price moves.

The episode also challenges the distinction between being directionally right and structuring a durable portfolio. AI may still reshape the global economy, while a leveraged or concentrated trade built around that outcome can fail in the meantime. Liquidity, position sizing, and correlations become especially important when public equities are being used to express a multi-year industrial forecast.

Aschenbrenner developed that forecast after publishing a series of essays in 2024 and leaving OpenAI. He had joined OpenAI after graduating summa cum laude from Columbia University at 19. OpenAI dismissed him in 2024 for allegedly mishandling confidential information. Aschenbrenner disputed that account, stating he shared a non-confidential design document with outside researchers and linking his departure to security concerns he had raised.

For Citadel, the deal follows an established playbook. The hedge fund acquired exposure during the Amaranth Advisors collapse in 2006 and Sowood Capital’s losses during the 2007 financial crisis. In 2021, Citadel and Point72 also invested $2.75 billion in Melvin Capital after its GameStop-related losses. Citadel's co-chief investment officer and other senior leaders reportedly spent hours reviewing the latest portfolio before an agreement was reached.

The immediate issue is now execution rather than ideology. Situational Awareness has to complete its unwind, communicate with investors, and determine whether its strategy can survive in a less concentrated form. As LinkedIn News noted in its coverage, the setback has forced a prominent AI investor to offload stocks after the fund took a major hit. For institutional investors, the lesson is blunt: a compelling technology forecast can still become a fragile trade when too much capital depends on the exact same outcome.