The Situational Awareness Collapse: How a 25-Year-Old's $45 Billion AI Bet Unraveled in Three Weeks

The Headline

In late July 2026, one of Wall Street's most talked-about young managers watched his hedge fund collapse from $45 billion in assets to roughly $10 billion in a matter of weeks — and then watched almost his entire public stock portfolio get sold, in a single pre-market block trade, to Ken Griffin's Citadel.

The manager was 25-year-old Leopold Aschenbrenner. The fund was called Situational Awareness. And the story, once you strip away the ticker symbols, is a very old one: a smart person with a correct idea used too much borrowed money to express it, and the market didn't give him time to be right.

Who Is Leopold Aschenbrenner?

Aschenbrenner is a German-born former OpenAI researcher who graduated from Columbia University as valedictorian at 19, then worked on OpenAI's Superalignment team before leaving the company in 2024. He became a household name in AI and tech circles after publishing a 165-page essay in June 2024 that argued the world was seriously underestimating how fast artificial intelligence was going to scale — and what that scaling would require in chips, data centers, energy and capital.

That essay was called "Situational Awareness: The Decade Ahead." It didn't just make him famous — it became the name and the thesis of his own investment fund.

What Is "Situational Awareness," Exactly?

Two things, and it's worth separating them:

• The essay — Aschenbrenner's 2024 manifesto arguing that AI progress would compound quickly, and that the physical inputs to that progress (semiconductors, memory, data centers, power generation) would become the real bottleneck and the real investment opportunity.
• The fund — Situational Awareness LP, the hedge fund Aschenbrenner launched later in 2024 with backing from Stripe's Patrick and John Collison, Daniel Gross and Nat Friedman, built to invest directly on that thesis.

The fund put money behind the "picks and shovels" of the AI buildout — neocloud GPU providers, memory manufacturers, data center builders and energy companies — while shorting some of the more established incumbents it believed would lose ground. It also held a large private stake in Anthropic.

It worked, spectacularly, for a while. The fund reportedly launched with around $225 million, returned roughly 439% after fees in the first half of 2025, and by July 1, 2026 had grown to a peak of about $45 billion in assets — a gain of more than 1,000% since inception.

What Actually Happened in July 2026

The Peak

By July 1, 2026, Situational Awareness had grown to roughly $45 billion. Its largest disclosed long positions — Nebius, Bloom Energy, SanDisk, CoreWeave, IREN, Core Scientific and Applied Digital — had been on a tear, and the fund was running with reported leverage of up to 400% (4x), amplifying every move.

The First Crack: The SK Hynix IPO (July 10)

SK Hynix, the Korean memory giant riding the AI memory boom, listed in the US. It was heavily anticipated — and it ended up functioning as a classic "sell the news" event. Within days, memory and AI semiconductor stocks started sliding.

The Meta Rotation (July 1–20)

Reports emerged that Meta was planning to sell excess cloud computing capacity to third parties through a new "Meta Compute" division. That was a direct threat to the "neocloud" business model — the idea that companies like CoreWeave, Nebius and IREN could rent out GPU capacity at a premium. CoreWeave, Nebius and IREN fell sharply on the news.

The Cascade (July 20–29)

By late July the sell-off had become self-reinforcing. The Philadelphia Semiconductor Index fell 28.6% from its peak on June 22 until July 29th. Individual neocloud names like CoreWeave and Nebius fell 45.4% and 47.7% respectively. Crucially, capital didn't just leave the AI trade — it rotated into large-cap software names like Adobe which went up by 35.2%, which happened to be exactly the type of stocks Situational Awareness was short. The fund was losing money on its longs and its shorts at the same time.

The Letter (July 24)

Six days before the forced sale, Aschenbrenner wrote to investors describing the drawdown as "one of the best buying opportunities since early 2025" and invited additional capital starting August 1. He never got to August 1.

The Margin Calls (July 30)

On the morning of July 30, all three of the fund's prime brokers — Goldman Sachs, JPMorgan and Bank of America — issued margin calls simultaneously. With assets down and leverage still in place, the fund had little room to negotiate. Its public portfolio — long positions and short positions together — was packaged and sold as a single block trade before the market opened.

Who Bought It

Two firms reportedly looked at the distressed book and passed: Millennium Capital, a multi-strategy fund with strict risk limits, and Jane Street, an original investor in the fund. Ken Griffin's Citadel bought the entire thing — every long, every short — in one transaction.

The Bounce (July 30, One Day Later)

The very next trading day, US stocks rallied. The Nasdaq Composite rose 2.8%, and the seven positions Citadel had just acquired rose roughly 25% in a single session.

Where Things Stand

After the sale, Situational Awareness still held an estimated $5 billion private stake in Anthropic — an illiquid position not affected by the margin calls. For the month, the fund lost roughly 67%. Since inception, it remained up an estimated 78%.

The Reported Portfolio, at a Glance

As of the fund's Q1 2026 SEC filings, before the July collapse:

Why Did Citadel Buy It?

This is the part people find counterintuitive: Citadel didn't buy Situational Awareness because it believed the AI thesis was wrong. If anything, the opposite. A few things made this a rational trade for Citadel specifically:

• Distressed assets trade at a discount. Block sales of this size, sold under margin-call pressure with three prime brokers acting in concert, typically execute below the prior closing price.
• Citadel had the balance sheet and infrastructure. Absorbing $13–14 billion of combined long and short exposure in one transaction requires serious capital and fast hedging capability — exactly why Millennium and Jane Street passed.
• Buying both sides gave flexibility. Because Citadel bought the longs and the shorts together, it wasn't stuck holding a single directional bet.
• It fit a pattern. Citadel has a track record of stepping in when leveraged players are forced to unwind, because forced sellers create mispricing that patient, well-capitalized buyers can exploit.

None of this proves Citadel made a profit — its hedges aren't public, the exact discount on the block trade wasn't disclosed, and a one-day bounce doesn't establish a realized gain. But it does explain the logic.

The Real Lesson: Leverage Doesn't Wait for You to Be Right

Here's the part that matters far more than any single ticker. Aschenbrenner's underlying thesis — that AI's growth would require a massive build-out of chips, data centers and power — was not obviously wrong. Microsoft's July 30 earnings, posted the day after the liquidation, showed Azure cloud growth of 43%, arguably validating the demand side of his argument. The trade may have been right. The leverage was not.

That's the mechanism worth internalizing:

• Leverage magnifies the move, not the odds of being right. At 4x leverage, a 25% decline in your positions doesn't cost you 25% of your capital — it can wipe out 100% of it.
• Concentration plus leverage removes your margin for error twice over. When the long book and the short book both move against you at once, there is no offsetting cushion.
• Being early looks identical to being wrong, until it doesn't. Margin calls arrive on the market's schedule, not the thesis's.
• A forced seller is not a strategic seller. Three prime brokers, acting simultaneously, removed the choice entirely — the structural risk of borrowed money is that it transfers control of your exit timing to your lenders.
• A good idea, badly sized, can still destroy a portfolio. This is really the whole story.

For anyone building or monitoring an investment portfolio, the takeaway isn't "avoid conviction" — it's "size conviction so a temporary drawdown can't force you out of a position you'd otherwise be right to hold." Diversification, leverage discipline and liquidity planning aren't the opposite of having a strong view. They're what let you keep the view long enough for it to play out.

FAQs