How AI Oracle Leopold Aschenbrenner Lost His Public Book to Citadel

Leopold Aschenbrenner’s Situational Awareness fund sold the bulk of its roughly $16 billion public equities portfolio to Ken Griffin’s Citadel after a brutal July for AI stocks, sources told Reuters. The mid-20s former OpenAI researcher had turned a viral 2024 essay into one of the fastest-growing hedge funds on record, only to watch leverage and a sharp sector rout force a block unwind.

The deal leaves the firm with about $10 billion, mostly private stakes including Anthropic, and strips out the high-profile public book other traders once shadowed. It is a sharp turn for a vehicle that returned 439 percent net through June.

The Essay That Built the Thesis

In June 2024 Aschenbrenner published the SITUATIONAL AWARENESS essay series on his For Our Posterity site. The long document argued that almost no one was pricing the coming decade of AI progress, from compute scaling to geopolitical stakes and what he called The Project.

He had just left OpenAI’s Superalignment team. Before that he graduated valedictorian from Columbia at 19, did growth research at Oxford’s Global Priorities Institute, and moved to San Francisco. German-born, he framed the work as securing liberty for posterity.

The piece drew wide attention in tech and policy circles. It also became the name and investment mandate for the fund he launched months later.

From Seed Capital to Twenty Billion

Situational Awareness Partners LP filed its 2024 Form D registration that September. Early capital came in around $200 million to a few hundred million, with anchors from Patrick and John Collison, Nat Friedman, Daniel Gross and later Jane Street.

By early 2026 the firm managed well over $20 billion. Some tallies put the peak near $24 billion or higher. Public 13F snapshots alone reached $13.7 billion in reported value by the March 31 filing. Staff stayed tiny, roughly 8 to 20 people.

Aschenbrenner ran concentrated, leveraged bets on AI infrastructure: power, storage, chips, data-center names and related options. Other managers tracked the positions as a leading signal on the buildout trade.

The same leverage that multiplied the first-half gains left little cushion when prices reversed.

July’s Cascade Hit Longs and Shorts

AI-linked equities sold off hard. Names in the book such as CoreWeave, SanDisk, Micron and Nebius each dropped more than 35 percent in the month. Infrastructure holdings including SK Hynix were hit. Short positions against software names like Adobe moved the wrong way.

The Nasdaq 100 fell about 10 percent in July. Broader hedge-fund AI books saw their worst monthly drawdowns on record. Prime brokers including Goldman Sachs, JPMorgan, Bank of America and others began marketing pieces of the book and pressed for capital or sales.

• Long side pressure: AI infrastructure and related equities reversed after the H1 run.
• Short side pain: Software shorts that had worked earlier flipped against the book.
• Leverage multiplier: Reports put gearing near 4x, turning a 30-40 percent asset drop into a funding crisis.
• Broker response: Margin demands and orderly unwind talks replaced free rein.

Aschenbrenner had written investors in late July that the selloff looked like one of the best AI buying windows since early 2025. Days later the forced sale closed. He later told LPs the firm had let them down that month and partly blamed short sellers for amplifying the pressure, comparing the dynamic to a bank run.

Citadel Takes the Public Book

Citadel bought most of the financed public portfolio in a block facilitated by the primes. Situational Awareness kept private holdings, notably its Anthropic stake valued by some accounts near $5 billion, plus a residual public book.

The fund’s last major 13F showed a mix of long equities in Bloom Energy, SanDisk, CoreWeave, IREN and others alongside large put exposures on semiconductors, Nvidia, Oracle, Broadcom and the VanEck Semiconductor ETF. A recent Schedule 13G filing confirmed ongoing reporting activity under Aschenbrenner’s name as managing partner.

Existing investors were given a short window into early August to add capital. The firm is not in full liquidation. It has simply de-levered the most visible, most volatile sleeve.

Stats snapshot

• Public book sold: bulk of ~$16B equities and options
• Remaining AUM: roughly $10B (stocks + privates)
• Key private hold: Anthropic stake retained
• YTD context: still positive after the H1 surge, per some tallies near 80%

Who Put Money Behind the Bet

Early backers mixed Silicon Valley operators and trading heavyweights. The Collison brothers of Stripe, former GitHub CEO Nat Friedman, and investor Daniel Gross provided brand and capital. Jane Street’s involvement signaled professional trading confidence in the thesis and the manager.

That roster helped the fund scale far faster than a typical first-time vehicle. It also raised the stakes when the public book had to be sold. Aschenbrenner’s own skin in the game and the essay’s reputation had made the fund a proxy for conviction on the AI infrastructure supercycle.

Why Leverage Turned Foresight Into a Margin Call

The core irony is structural. Aschenbrenner’s essay mapped decade-scale OOMs of compute, energy bottlenecks and national projects. The fund expressed that view with concentrated, leveraged public positions that live or die on weekly margin and quarterly flows.

When the July rout arrived, the long-horizon thesis had no time to work. Selling to meet brokers became the only path that preserved the private book and the firm. Crowd reaction on X noted the parallel to earlier leverage blowups without claiming a full Archegos-style collapse. Others pointed out that Citadel acquired the positions after the pain and before any rebound.

We let you down this month.

That line, attributed to Aschenbrenner’s investor letter and circulating widely, captured the tone. He still frames the period as a buying opportunity for the underlying AI buildout. The mechanics of 4x gearing on crowded names simply overrode the timeline.

Similar concentration risks appear across the AI supply chain. Chipmakers and memory suppliers face their own cycle questions; Micron’s take-or-pay contracts and cycle risks illustrate how even contracted demand can leave markets pricing a peak.

What the Remaining Book Still Holds

Situational Awareness is smaller, less levered and still open. The Anthropic stake keeps it tied to frontier model economics. Residual public names and any new capital will show whether the thesis can be expressed with more durable funding.

Prime brokers have already moved on. Other AI-focused funds are reassessing gearing after the same monthly drawdowns. Aschenbrenner’s essay remains widely read. The fund that carried its name just demonstrated how quickly a celebrated long-term view can meet short-term market plumbing.

The public book is gone. The private convictions and the manager remain.