Former OpenAI researcher Leopold Aschenbrenner’s hedge fund, Situational Awareness, has reportedly sold most of its public equity portfolio to Citadel after a sharp drawdown in AI infrastructure stocks. The reported transaction, first surfaced by The Wall Street Journal and picked up by TechCrunch and other outlets, marks a sudden reversal for one of the market’s most closely watched AI-themed funds.

What makes the episode notable for the AI industry is not only the scale of the unwind, but what remains in place. According to Bloomberg, as cited by TechCrunch, Situational Awareness still holds private positions including a large stake in Anthropic, reportedly valued at about $5 billion. That means the fund’s public-market AI infrastructure thesis has been cut back, but its exposure to private frontier-model and compute-adjacent assets remains materially intact.

A fast rise built on the AI infrastructure trade

Situational Awareness was launched in 2024 by Aschenbrenner, who became well known in AI circles for arguing that continued model scaling would require a massive expansion in chips, memory, power, and data center capacity. That thesis matched a broader market narrative that demand for AI systems would flow not just to model developers, but to the suppliers building the stack underneath them.

TechCrunch reports that the fund at one point delivered a 439% return for the year through June, citing the Financial Times. CNBC, also cited by TechCrunch, reported that assets under management reached as much as $45 billion at the peak before the recent decline. Those figures, while widely repeated, come through media reports rather than direct public filings in the evidence provided here, so they should be read as reported estimates rather than independently verified numbers.

The strategy appears to have been concentrated in public names tied to the AI buildout. TechCrunch identified positions including SK Hynix, Sandisk, Bloom Energy, and Nebius Group. Over the past month, according to that report, those stocks fell by more than 30% as investors reassessed whether heavy AI capital spending would produce near-term revenue quickly enough to justify elevated valuations.

That reassessment hit leveraged investors especially hard. Hedge funds often use borrowed capital to amplify conviction trades, and the same leverage that accelerates gains can force rapid selling when prices move the other way. In this case, the broad AI infrastructure pullback appears to have turned a high-conviction theme into a liquidity event.

Why Citadel stepped in

The central news event in this cluster is that Citadel, Ken Griffin’s hedge fund, reportedly bought the bulk of Situational Awareness’ public holdings after the selloff. Citybiz and TechCrunch both point to that outcome, with TechCrunch attributing the original reporting to The Wall Street Journal.

The logic behind the deal is straightforward even if the precise terms have not been disclosed in the source material. Citadel has a reputation for buying assets from stressed or forced sellers, especially when it believes the underlying positions may recover over a longer time frame than the seller can afford to wait. TechCrunch notes that Citadel already held some similar AI infrastructure positions before this purchase, suggesting that Griffin’s firm may share part of the long-term sector view while having a different risk tolerance and balance-sheet profile.

Bloomberg, as cited by TechCrunch, reported that after the transaction Situational Awareness’ assets fell to roughly $10 billion, down from around $20 billion in recent months according to an earlier Wall Street Journal report. Again, those numbers are secondhand via financial media, not direct statements in the evidence here. Still, the broad picture is consistent across outlets: a large reduction in the fund’s public-market exposure following a sharp decline.

For AI market observers, the important distinction is that this was not a blanket exit from AI. It was, based on the available reporting, a retreat from public equities exposed to the AI infrastructure cycle under conditions where leverage and timing became critical.

The Anthropic stake changes the story

The reason this is more than a hedge-fund blowup story is Anthropic. TechCrunch reports that Situational Awareness did not sell its private-company investments and still owns an Anthropic stake valued at around $5 billion, according to Bloomberg.

That matters because private AI assets and public AI infrastructure stocks can follow very different clocks. Public names such as SK Hynix, Bloom Energy, and Nebius Group are repriced daily as investors react to capex fears, earnings visibility, and macro pressure. A private stake in Anthropic, by contrast, is tied more to negotiated funding rounds, strategic demand for model providers, and a possible liquidity event such as an IPO.

TechCrunch says Anthropic was last valued at $96.5 billion in a Series H round in May and is expected to go public as soon as October, potentially at a higher valuation. Those are reported expectations, not confirmed company guidance in the evidence presented here. But if an IPO or secondary sale were to happen on favorable terms, it could materially offset losses from the public-market unwind.

Situational Awareness also reportedly retains other private positions, including MatX and Fluidstack. TechCrunch notes that Fluidstack was reportedly in talks in April to raise capital at an $18 billion valuation. For fund investors, those holdings likely now matter more than the reduced basket of listed stocks.

Evidence, claims, and what is still unconfirmed

This story rests mainly on media reporting rather than direct disclosures from Situational Awareness, Citadel, or Anthropic in the materials provided. The strongest factual claim across the cluster is that Citadel bought most of Situational Awareness’ public portfolio after the selloff; that appears in multiple reports, but the full transaction details, pricing, and exact list of transferred securities are not included here.

Several of the headline figures also require caution. The reported 439% return through June comes from the Financial Times as cited by TechCrunch. The peak assets-under-management figure of $45 billion comes from CNBC as cited by TechCrunch. The current Anthropic stake value of about $5 billion comes from Bloomberg as cited by TechCrunch. The expectation of an Anthropic IPO as soon as October is likewise a media report, not a confirmed timetable in the evidence.

There is also important context around Aschenbrenner himself. TechCrunch recounts that he joined OpenAI’s superalignment team in 2023 and left the company after OpenAI said he had improperly disclosed internal information. That history matters because it shaped his profile in AI circles and likely helped draw attention to his fund’s thesis, but it is separate from the current portfolio event.

The backers cited by TechCrunch — Jane Street, Patrick Collison, John Collison, Daniel Gross, and Nat Friedman — illustrate how closely this fund sat to both Wall Street and Silicon Valley networks. However, the presence of well-known backers should not be read as validation of current marks or future recovery.

What this means for AI builders and enterprise buyers

For builders and enterprise buyers, the selloff is a reminder that the AI stack is not one homogenous trade. Model companies such as Anthropic, infrastructure enablers such as SK Hynix, and service providers such as Fluidstack may all benefit from AI demand, but they carry different revenue timelines, capital needs, and valuation risks.

The public-market pressure described here reflects a growing investor concern: huge spending on GPUs, memory, power, and data centers does not automatically convert into short-term profits across the ecosystem. For enterprise AI teams, that matters because vendor roadmaps may increasingly be shaped by financing constraints, not just technical ambition. A company building around AI infrastructure may have strong product logic but still face harsher capital markets if revenue lags spending.

For founders, the split between what was sold and what was retained is also revealing. In a stressed market, liquid public names can be the first assets to go, while scarce private positions in companies like Anthropic can remain the strategic core. That does not mean private marks are safer in any absolute sense; it means they are priced differently and may offer more optionality if a financing or IPO window opens.

For enterprise AI procurement teams, the practical lesson is to watch counterparties closely. If your stack depends on providers exposed to the AI infrastructure cycle, volatility in public markets can affect pricing, support, expansion plans, and reliability even when the underlying technology demand remains real.

What to watch next

The first signal to watch is whether Situational Awareness stabilizes around its remaining private assets or seeks additional restructuring. TechCrunch reports that Aschenbrenner told investors in a July 24 letter, seen by the Financial Times, that the selloff looked like a buying opportunity and that he invited fresh capital starting August 1. Bloomberg, according to TechCrunch, reported that the effort did not produce the commitments he wanted. Whether new investor support emerges will indicate how much confidence remains in the original thesis.

The second signal is whether Citadel adds to or trims the acquired exposure. If Citadel holds or expands those AI infrastructure positions, it would suggest that the firm sees the drawdown as a valuation reset rather than a structural break.

Third, any movement around an Anthropic liquidity event could reshape the narrative quickly. An IPO filing, a new funding round, or secondary-market activity would provide a clearer read on whether Situational Awareness’ retained private stake can cushion the public-market losses implied by the recent reports.

Finally, watch the broader pricing of AI infrastructure names such as SK Hynix, Bloom Energy, and Nebius Group. If those stocks recover, the episode may look like a leverage-driven unwind during a temporary air pocket. If they continue to weaken, investors may be signaling deeper doubts about the timing and economics of enterprise AI demand.

Creati.ai perspective

This story is less about one fund manager’s rise and stumble than about the emerging fault lines inside AI investing. The market spent the past year rewarding nearly every layer of the AI supply chain at once. The unwind at Situational Awareness suggests that investors are starting to separate long-term belief in AI from short-term confidence in how and when each layer gets paid.

For the industry, the key takeaway is that capital discipline is becoming as important as technical narrative. A strong thesis around compute, memory, power, and model demand can still be directionally right and yet fail in public markets if leverage, timing, and liquidity do not line up. The retention of the Anthropic stake underlines that frontier-model exposure is still seen by many investors as a different class of bet from the broader AI infrastructure basket. That distinction is likely to matter more, not less, as the next phase of enterprise AI spending gets tested against real revenue.