The Golden Boy's Multi-Billion Blowup: AGI Faith Meets the Margin Call

A $45 billion fund called Situational Awareness had its form as a leveraged hedge fund terminated within 24 hours. What genuinely worries me is what he learned from it.

“This recent selloff is one of the best buying opportunities since the start of last year.” — Leopold Aschenbrenner, in a letter to investors, written six days before the margin call

On July 24, a Friday, a fund manager sent his investors a letter. It said this recent market selloff was one of the best buying opportunities since the start of last year. The letter closed with a PS to the effect that: we occasionally flag some very particular moments suited to adding to positions — if you’ve been waiting for this opportunity.

The person who wrote that letter is 25. The fund he manages is, on some measures, the largest single-theme hedge fund in the world.

Six days later, on July 30, before Thursday’s US market open, he sold the overwhelming majority of the fund’s public equity positions — roughly $16 billion of book value, long plus short — in one package, at a discount, to Ken Griffin’s Citadel.

The person who wrote the letter is Leopold Aschenbrenner, and the fund is Situational Awareness. This piece does three things: first, nail down the facts — was it actually a blowup, did it actually go bankrupt; second, place it in historical coordinates against Bill Hwang, Three Arrows Capital and FTX; and third, the main point: what this actually means for this AI cycle.

A disclaimer first: I run a fund myself and currently still hold a net long position in AI stocks. So nothing below is investment advice.

1. Who this person is

Aschenbrenner is German, born in 2001, entered Columbia at 15 and graduated at 19 as valedictorian. After graduating he did a stretch of research at Oxford, then joined the philanthropy arm of FTX — the crypto exchange that later collapsed. In 2023 he joined OpenAI to work on Superalignment, and in April 2024 he was fired over a dispute about a security memo.

Two months later he published the 165-page essay Situational Awareness: The Decade Ahead, of which essentially everyone in Silicon Valley has a copy. The core argument is actually simple: AGI arrives around 2027, and compute, power and chips are the most important strategic resources of this decade.

Then he did something very few people do. He didn’t go start a model company or build AI applications; he took that paper and raised a hedge fund on it — if you believe my judgement about AI, give me the money and I’ll turn the judgement into positions in public markets. The LP list is extremely glamorous: Stripe’s Collison brothers, Daniel Gross, Nat Friedman, and even Jane Street, which had essentially never backed an outside manager. At the end of 2024, it started with a bit over $200 million.

2. From $200 million to $45 billion in 21 months

What followed can be understood as fully expressing AI faith. Through June 30 this year, a 439% net return in half a year — that’s the number in the investor letter the Financial Times obtained, not a rumor. Translated: an LP gave him 100 at the start of the year, and by the end of June it was 539.

Scale snowballed along with the eye-catching performance. In early July, CNBC’s figure was $45 billion — $200 million to $45 billion in 21 months. However you count it, that already puts it among the world’s largest hedge funds.

How does a young man do that? Three elements: concentration, leverage, and being right on direction. The book was essentially all in on one line, AI infrastructure — CoreWeave and Nebius in compute, Micron, SanDisk and SK Hynix in memory, Bloom Energy in power, plus a pile of bitcoin miners converting into data centers. At the same time he shorted the traditional software names he thought AI would disrupt, Adobe among them. And on top of that book, roughly four times leverage — Goldman Sachs, JPMorgan and Bank of America were all his prime brokers.

3. July: a textbook long-short double kill

In July 2026 the market reversed. The Philadelphia Semiconductor Index fell close to thirty percent in a month, and the stocks he was heavily long fell 35%, 45%, even over 50% in that single month. Worse was the other side: the software stocks he was short rose instead. Longs down hard, shorts up hard — a textbook long-short double kill.

Under four times leverage, what does that mean? In July alone, this fund lost 67% — the number he later confirmed himself in an investor letter. Continuing the earlier translation: the LP’s 539 became 180 in a month. And most of the positions were losses locked in after forced liquidation at a discount, so even if things rebound afterward, that money doesn’t come back.

What does four times leverage mean? Chinese readers aren’t unfamiliar with leverage: you buy a house with a 30% down payment, prices fall 30%, and you’re close to wiped out.

Around July 29, three prime brokers issued margin calls. Bloomberg later reconstructed those 24 hours: the banks looked for buyers overnight, Millennium saw the quote, Jane Street saw the quote, and finally it went at one price, at a discount, to Citadel.

That opening letter about “the best buying opportunity” went out six days before the margin call. The epic buying opportunity he described himself, he never got to see.

Reading this, I still find it somewhat lamentable. Because in early July I heard with my own ears from people at Jane Street that they were extremely, extremely proud of having invested in this fund — they felt it was the best investment they’d ever made, roughly five times in half a year. Five times in half a year; who wouldn’t love it, who wouldn’t be proud. But what I’m very curious about is: when the margin call came, when they knew the fund was in trouble and Jane Street might even have to lend a hand, were they still that proud?

Maybe not.

4. Blown up, but not dead

Now to the first question: was it a blowup? Was it bankruptcy?

English-language media were remarkably consistent in their wording. The Wall Street Journal said “sold the equity portfolio in a block trade,” CNBC said “forced fire sale,” Bloomberg said “forced liquidation” — not one used the word bankruptcy. The fact is: when the trade completed, the fund still held roughly $10 billion of assets, of which about $5 billion is unlisted Anthropic equity, not a share sold; plus a block of equity positions with leverage entirely removed, in their own words, fully paid for.

Late on July 31, Leopold wrote LPs a second letter whose opening line was: we’re sorry, we let you down this month. That letter clarified two things. First, the sale to Citadel was to remove all of the fund’s leverage: shorts fully closed, no further reliance on broker financing, and the remaining public positions all fully paid-for cash positions. Second, in his words, the fund has not closed, has not liquidated, and has not converted to a pure private fund; it will continue as a hybrid public-private fund — “we took the necessary steps so that we can live to fight another day.”

So the accurate one-sentence summary is: Situational Awareness genuinely blew up, with enormous losses, but it didn’t die. Its form as a leveraged equity hedge fund was terminated within 24 hours; but it can continue operating and fight another day.

5. If you were its LP, would you go or stay

At this point I want to hand the question to you. Suppose you are its LP: on January 1 this year you invested $1 million; on June 30 your account was $5.39 million; today, $1.8 million. By the unaudited numbers he reported himself, July alone fell 67%, but 2026 as a whole is still up 80%. The fund manager has just written to you saying he’ll keep going long AI, that public positions will be zero-leverage and fully paid for from here, and that he can take a one-on-one call with you next week and discuss any questions.

Do you redeem?

Don’t answer too fast, because there’s a more brutal detail here: you may not be able to redeem at all. This fund’s investors signed lockups running several years — the circulating account is two years. To this day, no report says the blowup triggered a redemption waiver, and not one LP has publicly demanded money back. None of that has happened. In other words, LPs can’t take out a cent — not because they have faith in AGI, purely and simply because the contract was signed.

The real question is: when the lockup expires — the earliest LPs may be at year-end or early next year — on the day the door opens, does that investor go, or stay?

6. Placed in historical coordinates

The second question: what magnitude is this blowup historically? Look at three precedents: Bill Hwang (Archegos) in 2021, Three Arrows Capital in 2022, and FTX.

Start with size. Situational Awareness peaked around $45 billion, already exceeding Bill Hwang’s $36 billion before he blew up. On an AUM basis, this is the largest fund blowup in five years — and going back another five years I can’t think of a fund of comparable size that blew up. But on notional exposure, the $160 billion of gross exposure Archegos stacked up with total return swaps remains the untouched record.

But on destructive power — destructive power to the financial system — Situational Awareness ranks very far down. One detail gives you the feel: in Bill Hwang’s case nine investment banks lost over $10 billion combined, with Credit Suisse alone eating $5.5 billion, against total fees of just $17.5 million earned from Archegos the previous year. Fifty-odd billion lost for a ten-million business; two years later Credit Suisse itself was gone. Three Arrows’ $3.5 billion of defaulted debt knocked over the dominoes of the entire crypto credit industry. FTX misappropriated over $8 billion of customer funds and SBF was sentenced to 25 years.

By comparison, Situational Awareness going from $45 billion to $10 billion overnight evaporated over $30 billion of AUM — on the scale of what vanished alone, higher than Bill Hwang, higher than Three Arrows, higher than FTX. But honestly, this summer this fund didn’t harm a bank, didn’t harm a broker, and certainly didn’t harm any company up or down the chain. The only thing it harmed was its own LPs. Those investors’ psychological experience in July was roughly: money that blew in on the wind has blown away again.

7. What it means for this AI cycle

This is the part I actually want to develop, on two time horizons, short and long.

Short term, I think this is a clearing of positioning, not a collapse of fundamentals. Look at one comparison: in July the Philadelphia Semiconductor Index fell just under thirty percent, but Morgan Stanley’s momentum TMT index fell 53.5%. What does that mean? What fell hardest was the most crowded book — a textbook momentum unwind. And Situational Awareness was itself the incarnation of that most crowded book; its liquidation was the climax of the unwind.

So you see a counterintuitive phenomenon: on the day the liquidation landed, the dumped stocks rose sharply as a group — Nebius, IREN and Bloom Energy all rebounded around 25%, though a good deal of that was short covering and a squeeze. Wall Street’s reading was very clear: the largest forced seller has cleared, and a short-term bottom may have appeared. Not one AI infrastructure order or capex plan was falsified in July — what was falsified is owning them with four times leverage. AGI is coming, that’s not wrong; you just can’t survive to that day. With four times leverage you certainly can’t survive to that day.

And historically, how do markets trade after a blowup? The answers from previous instances are more regular than you’d expect. LTCM blew up in 1998, the Fed convened a rescue, the S&P still rose 26% that year, and the Nasdaq rose for another 18 months; Bill Hwang blew up in 2021 and the broad market barely noticed, with the S&P up 27% that year. A single fund’s forced liquidation, as long as it doesn’t drag down the banking system or damage the financial system, gets digested by the market very quickly — that’s the side supporting a short-term bottom.

But at the same time, this kind of event is usually also a marker of the cycle’s second half. Eighteen months after LTCM, the internet bubble formally topped; nine months after Bill Hwang, US equities topped and entered a bear market. A fund blowup usually isn’t the end of a rally; it’s usually the starting gun for the second half.

Taking a longer view, it exposed how leveraged this cycle is — and that’s where the alarm should genuinely sound. A thematic fund less than two years old obtaining four to five times credit from three top investment banks tells you how loose the entire prime brokerage system’s risk appetite for the AI theme had become before July’s large correction. We can be confident Situational Awareness won’t be the only player expressing AI faith with leverage: have other funds borrowed? Is positioning crowded? Has retail bought leveraged ETFs? Have miners converting to data centers issued debt? Have the neoclouds borrowed to stockpile cards? The layers of leverage in this AI ecosystem are far thicker and far richer than two years ago.

And I believe that for some time ahead, whether or not Kevin Warsh raises rates, those days of easily obtained cheap money, easily obtained credit lines and easily obtained cheap leverage will be far scarcer than before. The marginal buyer’s firepower gets weakened, and even if the rally doesn’t end, it rises more slowly and falls more easily — the typical form of every deleveraging cycle.

8. It wasn’t trading the AI rally; it was the AI rally

Situational Awareness’s problem was never direction — after it liquidated, those stocks kept rising. Its problem was using four times leverage to express a ten-year judgement. The market doesn’t care whether you’re right or wrong; the market only proves one thing: when you blow up, you couldn’t hold on.

I’d go further and say it wasn’t trading the AI rally, it was the AI rally. A fund of tens of billions, betting on a few dozen small and mid-cap AI infrastructure stocks — in one, Core Scientific, it bought 8% of the float, becoming a major shareholder outright. So how much of the first half’s surge in these AI stocks was it buying up itself? Stepping on its own feet upward? You simply don’t know.

So I’d rather treat this as a margin test of the AI rally: the underlying assets may be fine, but everyone who over-levered got exposed. It fully tested how much generosity there was in the system, and exposed who was swimming naked.

And what comes next is the part that genuinely worries me. Situational Awareness wrote its positions into a paper, into annual reports, into 13Fs, into investor letters, onto Twitter — into every public place. The market knew perfectly well what it held, and so in July the whole market began hunting those names specifically. In his final investor letter on July 31 he called this a bank run.

What I want to ask is: when he said that, what exactly did he learn? Could what he learned be — next time I hide this information so you can’t see it, and then you can’t run on me and I quietly make the money? Does that mean the next round of crowded AI positioning will be less transparent and harder to predict? If he walks away from this intact, what he learned may not be “don’t use leverage” but “hide leverage cleverly.”

This has been done before. Bill Hwang used total return swaps to keep $160 billion of exposure below the waterline, and nobody found it — and when the chain of blowups actually came, it dragged Credit Suisse down with it.

Leopold has already committed that he won’t buy stocks with leverage again. But when other fund managers read that letter, what should their reaction be? As a fund manager myself, what should I learn?

Having said all that, I still have some worry: this thing may not end that simply.

If you're working on this too — or you think we've got it wrong — write to us at [email protected]; if you'd rather not write, just leave your email below.