Situational Awareness: How a +439% AI Bet Became a Fire Sale in Three Weeks

There is a specific kind of story markets produce every few years, and it always teaches the same lesson in a new costume. This month’s costume is artificial intelligence. Leopold Aschenbrenner — a 24-year-old former OpenAI researcher who became briefly famous in 2024 for a 165-page essay arguing that artificial general intelligence could arrive by around 2027 — did what almost no essayist ever does: he put real money, and a great deal of other people’s money, behind the thesis. He launched a hedge fund named after the essay, Situational Awareness, and for eighteen months it looked like one of the great trades of the era. Then, in about three weeks, it became one of the great blow-ups.

The details are dramatic, and we’ll walk through them. But strip away the AI novelty and what’s left is a textbook that could have been written about Long-Term Capital Management in 1998, or a leveraged trader on Reddit last Tuesday. The instrument changes. The mechanism never does.

Key takeaways

• Leopold Aschenbrenner’s Situational Awareness fund returned ~439% in the first half of 2026 and reportedly peaked near $45 billion — a generational run.
• In late July it unwound its entire public book — longs and shorts — in a single forced sale to Citadel to meet margin calls, reportedly down ~67% on the month.
• The cause wasn’t a wrong idea about AI. It was ~4x leverage, extreme concentration, and a book that was really one correlated bet — long AI infrastructure, short “AI losers.”
• The lesson is the oldest one there is: you can be right about the decade and still be liquidated in three weeks. Survival is a sizing decision, not a forecasting one.

The setup: an essay that became a balance sheet

Aschenbrenner’s résumé reads like a thesis about the age itself. He entered Columbia at 15, graduated valedictorian at 19, and landed on OpenAI’s Superalignment team under Ilya Sutskever before being fired in 2024 in a dispute over whether he had improperly shared information or, as he maintains, raised legitimate security concerns. His June 2024 essay, “Situational Awareness: The Decade Ahead,” became required reading in tech circles for its confident, specific case that AI progress was faster than governments understood.

Within weeks he had turned that fame into capital, reportedly raising around $225 million to seed a hedge fund from a roster of technology heavyweights — Stripe’s Patrick and John Collison, former GitHub CEO Nat Friedman, and investor Daniel Gross among them, with Jane Street later cited as an investor. The fund scaled at a speed that should itself have been a warning: from under a billion dollars to a reported peak in the tens of billions — accounts range from around $20 billion for its equity book to roughly $45 billion in total assets — in little more than a year.

The trade: one idea, expressed with maximum force

The strategy was the essay, leveraged. On the long side: the physical picks-and-shovels of the AI build-out — the chipmakers, data-center operators, and power suppliers expected to profit from the infrastructure boom, names reportedly including SK Hynix and CoreWeave. On the short side: the software companies the thesis judged most vulnerable to being disrupted by AI, reportedly including firms like Adobe. Long the winners of the AI age, short the losers. And, crucially, all of it geared up with roughly four times leverage borrowed from Goldman Sachs, JPMorgan, and Bank of America.

Read that structure carefully, because its flaw is the whole story. A long book of AI-infrastructure names and a short book of “AI-disrupted” software names is not two bets. It is one bet — a single, enormous wager on the AI trade continuing in a straight line — expressed twice. The longs and the shorts were the same thesis wearing two hats, which meant they could, and eventually did, lose money at the same time. The diversification was an illusion. There was really only one position.

A portfolio of correlated bets is not a portfolio. It is a single position in a costume, and it takes off the costume at the worst possible moment.

The unwind: three weeks from triumph to fire sale

That final forced sale is the part every trader should study, because it is where leverage turns a loss into a death. Once the margin calls came, the fund had to sell — and in markets, being a known forced seller is catastrophic. The whole market can see that you must transact, so liquidity vanishes exactly when you need it and buyers only step in at punitive prices. This is not bad luck; it is a documented dynamic. It is the “predatory trading” of Brunnermeier and Pedersen — sell into the forced seller, then buy it back cheap — and it is the same mechanism that finished off LTCM in 1998 and that sweeps a retail trader’s clustered stop-losses. Citadel didn’t need to be cruel. It just needed to be the only buyer when someone else had no choice.

The crucial point: he may be right and it doesn’t matter

Here is what makes this more than schadenfreude, and where the honest lesson lives. Aschenbrenner’s underlying thesis might be entirely correct. AI may well be the defining economic force of the decade; the infrastructure names may resume climbing next month; the Anthropic stake he still holds could turn out to be the best position in the whole book. None of that would change what happened. The fund was liquidated before its thesis could be proven. Leverage doesn’t just magnify returns — it collapses your time horizon to the length of your lenders’ patience, and a margin call does not care that you’ll be vindicated in 2027.

This is the single most important idea in trading, and it keeps recurring on this blog because it keeps recurring in the world. We wrote it about Michael Burry: being early is indistinguishable from being wrong. We wrote it about LTCM: two Nobel laureates whose trades were mostly correct, denied the time to be proven right by their own leverage. Situational Awareness is the 2026 edition, at internet speed. Correct analysis plus fatal sizing equals zero.

Being right about the decade is worthless if your leverage only bought you three weeks. The market can stay irrational — or simply volatile — longer than a margin call will let you stay solvent.

The behavioral autopsy: four ordinary mistakes at extraordinary scale

What’s striking is that none of the errors here are exotic. They are the same four that show up in ordinary retail accounts, just with nine more zeros.

Concentration disguised as conviction. The entire fund was one idea. High conviction is a reason to take a position, never a reason to make it your only one; the moment your book is a single bet, a single bad month is existential. Leverage that removed the time to be right. Four-times gearing turned a large-but-survivable drawdown into a forced liquidation. Recency and extrapolation. A 439% run is exactly the kind of recent history that makes a manager — and their investors — believe the line goes up forever, the extrapolation error that peaks in confidence right before the reversal. And a track record mistaken for proof. A spectacular short run is weak evidence of a durable edge and can simply be the fat, volatile tail before the fall — precisely the warning of Fooled by Randomness.

Put them together and you get the anti-Thorp. Where Ed Thorp spent two decades sizing every bet so that no single one could end the fund, Situational Awareness made one bet large enough to end everything — and, for a while, was rewarded lavishly for it. That reward is the trap. The market pays you generously for reckless concentration right up until the afternoon it takes it all back.

What it means for your account

You will never run $45 billion, and you may feel this has nothing to do with you. It has everything to do with you, because the mistakes don’t require billions — they scale all the way down to a single margin-enabled brokerage account. If all your positions rise and fall together, you don’t have a portfolio, you have one trade. If leverage means a normal drawdown can force you to sell, you’ve handed control of your timing to your broker. If a hot streak has you sizing up because “it keeps working,” you’re extrapolating at the worst moment. And if you’re treating a good few months as proof you’ve cracked it, you may just be standing on the fat part of the tail. The defense is not a better view of AI, or of anything. It’s sizing and diversification that let you survive being early, wrong, or merely unlucky — which, sooner or later, you will be.

From belief to behavior: is your book really one bet?

Resources and further reading

• The reporting: CNBC, “Leopold Aschenbrenner’s Situational Awareness fund: $45B to fire sale” and “Why Situational Awareness hedge fund imploded” (July 30–31, 2026); Bloomberg, “Situational Awareness Assets Fall to $10 Billion After Losses” (July 30, 2026).
• The Citadel sale and Anthropic stake: WSJ / Yahoo Finance on Citadel’s purchase of the equity book; TechCrunch, “AI hedge fund Situational Awareness may have sold its public portfolio, but it still has its Anthropic shares” (July 30, 2026).
• The origin: Aschenbrenner, L. (June 2024), Situational Awareness: The Decade Ahead; and biographical background via Wikipedia and Fortune.
• The historical parallel: Gecko, When Genius Failed — LTCM’s leverage-and-forced-liquidation blueprint, and Brunnermeier & Pedersen (2005), “Predatory Trading.”
• The sizing counter-example: Gecko, Ed Thorp and the Kelly criterion — how to size so no single bet can end you.

Frequently asked questions

Essay in Gecko’s trading psychology series. This is a developing story; figures on assets, leverage, returns, and the July 2026 liquidation are drawn from contemporaneous reporting (CNBC, Bloomberg, WSJ, TechCrunch, Fortune and others) and may be revised — verify current details at the source. Accounts of the fund’s peak size differ (roughly $20 billion for the equity book to about $45 billion in total assets). This piece analyzes market behavior and risk and is not a judgment of, or allegation against, any individual or firm. Gecko is an educational and informational tool. Nothing here is financial, investment, or trading advice, or a recommendation for or against any security, strategy, or the use of leverage. Trading and leverage carry substantial risk of loss.