Why Citadel's Massive AI Unwind Is the Story

The Quick Buy, The Quicker Exit

Citadel bought a distressed hedge fund’s portfolio in late July. Within weeks, it had liquidated more than 80% of the risk. The sequence matters more than either move alone.

According to a client letter obtained by CNBC, Ken Griffin disclosed that Citadel entered discussions with Situational Awareness on July 29 to acquire some of the fund’s holdings. By that point, Situational Awareness was already in distress. The firm had concentrated long positions in the AI trade and short positions in software companies, a pair of bets that relied on a specific worldview: AI would disrupt software but the disruption wouldn’t be clean or linear.

Both sides of that bet lost money simultaneously. Stocks like Sandisk and Bloom Energy — part of Situational Awareness’ long book — dropped more than 50% in June and July. Meanwhile, software names like Adobe rebounded, crushing the shorts. The fund faced margin calls and was forced into compulsory sales. Citadel stepped in as the buyer of last resort, absorbing more than $4 billion in market value across over 100 block trades.

Then Citadel sold almost all of it back out.

Who Gets Hurt by the Narrative

Situational Awareness was founded by Leopold Aschenbrenner, a 25-year-old whose thesis was built on the conviction that AI would fundamentally reshape the global economy. That kind of conviction usually comes with two things: deep expertise and dangerous rigidity. Aschenbrenner was not just betting on AI — he was betting against the incumbents who would supposedly get disrupted by it. The fund’s structure reflected that binary thinking.

What happened in June and July exposed the flaw. AI stocks did not move in a straight line up. Some names cratered. The software companies they had shorted did not simply evaporate. The thesis was not wrong about the direction of travel — it was wrong about the speed, the winners, and the paths to disruption. And having concentrated positions in a thesis that can only work if everything goes right is a structural vulnerability, not a strategy.

Citadel understood this immediately. Griffin’s letter noted the extraordinary cooperation required to complete a transaction of this size, but the real story is what happened after the acquisition closed. Griffin did not hold the position. He did not take a long view. He unloaded the bulk of it.

The Signal the Market Missed

There is a detail buried in Griffin’s disclosure that deserves more attention than it is getting. Since Citadel bought Situational Awareness’ assets, the AI trade has rebounded. The fund’s forced selling represented a local bottom for the sell-off that began in June. In market terms, Citadel bought panic and sold recovery.

This is not unusual for a firm like Citadel. Multistrategy funds are built to profit from dislocation, not to hold convictions. But the move carries a signal that extends beyond Citadel’s own P&L. When the most sophisticated money in the world buys a concentrated AI thesis at fire-sale prices and then exits rapidly, it is not necessarily saying AI is dead. It is saying the particular way those positions were structured was untenable.

The difference matters. Aschenbrenner’s fund was a pure-play bet on a specific narrative of AI disruption. Citadel’s response suggests that the institutional view has shifted from narrative conviction to selective exposure. The money is still interested in AI, but the interest is narrower, more defensive, and far less enthusiastic about the long positions that got crushed in June and July.

Second-Order Effects: The Dominoes Keep Falling

The ripple effects of this trade extend well beyond Citadel’s balance sheet. Other hedge funds that ran similar concentrated AI narratives likely feel the pressure now. When Citadel disposes of large blocks, it creates selling overhang that depresses prices across the sector, which in turn triggers margin calls for other leveraged players. This creates a feedback loop: forced selling begets more selling, and Citadel’s rapid exit essentially became a reference point for how aggressively one can extract value from a broken thesis.

Pension funds and endowments with AI exposure are watching closely. Many of these institutions allocate to hedge funds as a proxy for direct market access, and a high-profile failure of this magnitude forces a reassessment of how much capital is deployed toward single-thesis strategies versus diversified approaches. The trend toward multi-strategy vehicles — exactly the kind Citadel popularized — is likely to accelerate.

There is also a psychological dimension. The AI trade had enjoyed near-universal bullishness among institutional allocators throughout 2024 and early 2025. A 25-year-old founder running a concentrated bet that lost decisively sends a uncomfortable message: conviction without flexibility is indistinguishable from hubris. Portfolio managers who resisted the AI consensus earlier may find their caution rewarded, while those who piled in at peak enthusiasm face difficult conversations with their limited partners.

What’s Left and Where the Money Goes

Griffin also confirmed that Citadel’s flagship Wellington fund returned 5.94% in July — its best month since 2022. That performance came despite, or perhaps because of, the disciplined risk management that defined the Situational Awareness response. The multistrategy approach does not require a single thesis to work. It requires the firm to be right about many small things and wrong about nothing catastrophic.

The AI sector is now facing a broader reckoning. The stocks that fell hardest during the June-July selloff — particularly hardware-adjacent names like Sandisk and energy plays like Bloom Energy — will need to rebuild confidence. The software names that rallied on the short squeeze are the ones institutional money is now more cautious about. The disruption narrative has not died, but it has lost its simplicity.

For Situational Awareness, the forced sale was devastating. The fund was likely wound down entirely, given that it had to divest all of its public positions. For Aschenbrenner, the episode is a lesson in how quickly a concentrated thesis can become a trap when the market does not move on schedule.

For Citadel, the trade was textbook. Buy dislocated assets from forced sellers, capture the spread, exit before the crowd returns. Griffin’s letter thanked the banking teams for their cooperation, but the real skill was knowing when to stop being cooperative and start being profitable.

The Bottom Line

The easy money in the AI trade has been made and lost. The next phase belongs to firms that can navigate ambiguity rather than bet everything on a single timeline. Citadel’s swift exit from Situational Awareness’ portfolio is not a bet against artificial intelligence — it is a bet against the kind of thinking that treats a complex, unfolding transformation as a simple bet to be won. That is a space Citadel occupies comfortably. The rest of the market is just beginning to understand the difference.