Updated July 30, 2026, 1:54 p.m. PDT
Citadel has bought most, and possibly all, of the public-stock portfolio held by Situational Awareness, the AI-focused hedge fund founded by former OpenAI researcher Leopold Aschenbrenner. The hurried transaction followed steep losses, attempts to raise fresh capital and discussions with prime brokers about meeting margin requirements.
The deal is well corroborated. Its exact perimeter is not.
CNBC’s updated report says Situational Awareness sold all of its public-stock holdings. Axios reported the same. The Wall Street Journal described Citadel’s purchase as the bulk of the portfolio, while follow-up reports from the Financial Times, Bloomberg and Reuters used similarly narrower language.
None of those reports establishes that Situational Awareness has entered legal liquidation, gone bankrupt or lost its full reported $20 billion to $24 billion of assets under management. The adviser remains listed as approved in the SEC’s Investment Adviser Public Disclosure database. Major-news reporting says the firm retained at least some private-company investments, including its stake in Anthropic. The FT says it intends to continue as a private investment firm.
The evidence supports a more precise conclusion. Situational Awareness suffered a severe financing and liquidity crisis, then transferred most or all of its liquid public-equity book to a larger buyer. What remains, how much was lost and what investors will ultimately recover are still unknown.
Even the scale of the fund is unsettled. Bloomberg reported that assets had fallen to about $10 billion, citing people familiar with the matter. CNBC’s updated account cited one source who said the fund had been as large as $45 billion at the start of July. Those source-based estimates are neither audited net asset values nor directly comparable to the $9.278 billion of regulatory AUM reported in the firm’s June Form ADV. They should not be subtracted from one another to calculate a loss.
Six days from confidence to crisis
The speed of the reversal makes this story unusual.
In an investor letter dated July 24, Aschenbrenner reportedly told clients that the fund had gained 439 percent net through June. The figure comes from the letter as described by the Financial Times; it is not independently auditable through public filings. Earlier reporting had put the fund’s return at roughly 270 percent through May and more than 1,000 percent since its late-2024 launch.
The July 24 letter acknowledged damage from the market sell-off, especially in Asia, but presented the decline as an opportunity. Investors were given a chance to add money on August 1, according to the FT. Days later, the firm was approaching existing investors and lenders for cash and offering assets to some limited partners. One source characterized the outreach as improvised rather than a conventional fundraising round.
CNBC then reported that Situational Awareness’s AI-infrastructure investments, including South Korean memory-chip maker SK Hynix, had fallen sharply. At the same time, software stocks the fund had sold short, including Adobe, moved against it. Bank of America, Goldman Sachs and JPMorgan were working with the firm as it tried to satisfy margin requirements or cut positions in an orderly way, according to CNBC.
Early reports said the fund was exploring sales of private-company stakes, including Anthropic. CNBC initially said it could not determine whether negotiated sales would satisfy the margin pressure or whether a wider liquidation was under way. In its 12:49 p.m. PDT update, however, CNBC quoted a Situational Awareness spokesperson saying reports that the Anthropic stake was being marketed were “not accurate.” The same update said the fund had exited all public-stock positions.
By then, Citadel had emerged as the buyer in a transaction assembled within roughly a day, according to the FT. The Journal reported that Millennium Management had also bid for the portfolio.
This was not a routine quarterly rebalance. It was a block transfer arranged while the seller needed cash and the usual option of patiently waiting for prices to recover was disappearing.
The fund turned an AI forecast into a trade
Aschenbrenner built Situational Awareness around the argument he set out in his 2024 essay series, Situational Awareness: The Decade Ahead. He expects increasingly capable AI systems to drive an enormous build-out of compute, memory, data centres and electricity generation. The investment implication was to own the physical bottlenecks.
The fund’s public filings show that logic in practice. Its March 31 portfolio included Bloom Energy, CoreWeave, Sandisk, IREN, Core Scientific, Applied Digital, CleanSpark and Riot Platforms. These companies span on-site power, memory, cloud compute and former bitcoin-mining infrastructure being repositioned for AI workloads.
The same filing disclosed large put positions tied to Nvidia, Broadcom, AMD, Oracle, Taiwan Semiconductor Manufacturing Company and the VanEck Semiconductor ETF. It also showed smaller direct shareholdings in some of the same names and call positions in Micron, Sandisk, CoreWeave and TSMC.
That combination is better understood as a relative-value expression than a simple bet against AI chips. Situational Awareness could favor electricity, memory and compute-hosting businesses while using puts to protect against expensive or crowded semiconductor exposure. It could also use those positions to express a view that profits would migrate from one layer of the AI supply chain to another.
The public report cannot reveal the live portfolio that entered July. A Form 13F is a delayed quarter-end snapshot, and it omits foreign-listed securities, private investments, cash, bonds, futures and short sales. CNBC’s reporting about an Adobe short, for example, describes a position that cannot be reconstructed from the March filing.
How both sides of a trade can demand cash
A long-short portfolio is designed to reduce some market risk, but the word “hedged” does not mean safe.
Suppose a fund owns AI-infrastructure stocks and shorts software companies it expects AI to disrupt. If the infrastructure stocks fall while the software shorts rise, both sides lose money at once. The long positions shrink in value. The short positions require more collateral because buying them back has become more expensive.
Borrowing intensifies the cash demand. A prime broker does not need to decide that a fund’s long-term thesis is wrong. It needs enough collateral to protect the loan today. Falling asset values can trigger higher margin requirements, which force the client to supply cash, pledge more assets or reduce positions.
Selling into a weak market can then create a loop:
- Falling prices reduce collateral value.
- Prime brokers ask for more margin or less exposure.
- The fund sells positions to raise cash.
- Those sales can push prices lower, weakening the remaining collateral.
- The fund must sell more or find a buyer for a large block.
The problem becomes harder when a portfolio combines liquid public securities with valuable but illiquid private stakes. An Anthropic holding may be attractive to buyers, but it cannot be converted into cash as quickly or transparently as a listed stock. Its estimated value does not automatically satisfy a margin call due that morning.
This is why an investor can be directionally right over several years and still become a forced seller. The lender’s timetable is shorter than the investment thesis.
What Citadel bought remains disputed
The most consequential factual disagreement concerns the size of the transfer.
CNBC and Axios say all public-stock holdings were sold. The Journal says Citadel bought the bulk of the stock portfolio. The FT described a large portion of roughly $16 billion in public-equity holdings. Bloomberg reported a big chunk and said some public stocks remained. Reuters said Citadel bought most of the holdings.
Those versions may reflect different definitions or different moments in a fast transaction. “Public investments” might include foreign-listed shares, derivatives and short positions in one account but not another. A report describing the borrowed portion of a book can also sound inconsistent with one describing the client-capital-funded remainder. Without a first-party statement or transaction documents, the cleanest description is that Citadel acquired most or possibly all of the public-equity portfolio.
The buyer was Citadel, Ken Griffin’s hedge-fund and investment firm. Citadel Securities is a separate market-making company. Social-media posts have repeatedly blurred the two while alleging that Citadel engineered the sell-off, spread interest-rate fears to cause margin calls or bought the portfolio at a specific discount. No credible public evidence currently supports those claims.
There is also no verified report of an investor-redemption run, a formal bankruptcy filing or a lender seizure. The reported pressure came from trading losses, borrowed financing, margin requirements and the need to raise cash.
The $13.68 billion 13F number is not what you think
Situational Awareness’s March 31 Form 13F reported 42 entries with a total value of $13.676 billion. That number is often mislabeled as the fund’s AUM or the amount it had invested in stocks.
The underlying SEC information table contains approximately $3.86 billion of ordinary shares, $8.46 billion associated with put positions and $1.36 billion associated with calls. Form 13F reports option positions largely in terms of the securities underlying them. Those figures are not the premiums paid, the net exposure or the maximum capital at risk.
The filing also excludes private assets such as Anthropic, foreign-listed holdings such as SK Hynix, cash and short positions. It describes March 31, not the portfolio transferred on July 30.
Situational Awareness LP and Situational Awareness Partners LP filed identical $13.676 billion tables. Adding them together would double-count the same reported book.
The AI thesis and the financing structure face different tests
Situational Awareness’s losses do not settle the debate about how quickly AI capabilities will advance or how much power and compute future models will consume. Citadel’s willingness to buy the portfolio could even indicate that a sophisticated counterparty saw value in the assets at the negotiated price, although the purchase reveals nothing conclusive about their long-term prospects.
The episode does expose weaknesses in turning a sweeping technological forecast into a financed portfolio.
Concentrated thematic trades can contain hidden correlation. A fuel-cell company, a memory manufacturer, a GPU cloud provider and a bitcoin miner may look like different businesses. In a broad retreat from AI capital spending, investors may sell all of them together. Their trading liquidity can also deteriorate at the same time.
Options add another layer. A large notional put position may cap losses in one scenario yet provide little protection against a different combination of moves, especially when strike prices, expiries and offsetting positions are unknown. Public 13F data is too coarse to calculate that protection.
Finally, a private stake can make a fund look wealthy while leaving it short of cash. If private holdings rise in estimated value as public collateral falls, headline AUM can remain impressive even as the fund’s ability to meet daily financing demands deteriorates.
The lesson for AI investors is narrower than “the boom is over.” A long-duration view requires financing that can survive short-duration volatility. Conviction does not extend a margin deadline.
The documents that can resolve the open questions
Public records lag a crisis like this one. No SEC filing available at the time of publication confirms the Citadel transaction, the exact loss, the price paid or the post-sale value of the fund.
Amendments to beneficial-ownership filings could arrive sooner. Situational Awareness had recently disclosed stakes above reporting thresholds in Nebius and SharonAI, and it previously reported a large Core Scientific position. Schedule 13D or 13G amendments may show when ownership falls below those thresholds.
The next Form 13F will be less useful than it may appear. The report due August 14 covers the quarter ended June 30 and therefore predates the July transaction. The September 30 snapshot, due November 16, should provide the first quarterly public view capable of reflecting a July 30 exit. It will still omit private investments, foreign securities and short positions.
A Form ADV amendment or withdrawal could reveal a change in the adviser’s status. A first-party statement from Situational Awareness or Citadel could settle whether the public book was sold in full and whether the firm will manage outside capital. Until then, “forced deleveraging” is supported by the reporting. “Liquidation of the fund” is not.
Disclosure: This article is reporting and analysis based on public filings and attributed news reports. It is not investment advice.
Kingy Launch Brief
Put the week’s verified AI launches in your inbox.
One source-checked edition every Friday, with a clear try, watch or skip verdict. After subscribing, check your inbox and confirm your address.
Free · Fridays · Double opt-in · Unsubscribe anytime