Wall Street's hottest AI fund rode leverage to 439% | Wednesday, August 5, 2026

The book was concentrated and heavily leveraged, long AI memory and semiconductors, names like SK Hynix, Micron, and Nebius, reportedly with up to 400% leverage on some positions. When AI infrastructure stocks turned in late July, the same leverage that manufactured the returns worked in reverse. Chip stocks shed more than a trillion dollars in market value after SK Hynix missed and signaled softer memory demand. Margin calls arrived from three prime brokers. On July 30, the fund sold its entire public equity portfolio, about $16 billion, to Citadel at a discount of roughly 10%, and lost around two-thirds of its value on the month.

Here is the enterprise read. The AI capital cycle that funds your vendors' discounts and secures their compute is not a straight line, and it is more leveraged than the narrative admits. The very suppliers in that fund's book, the memory and infrastructure vendors, sit in your own supply chain. A single fund unwinding a $20 billion book in one trade, on the back of one earnings miss, is a signal: do not extrapolate today's compute prices and availability out to your three-year plan. Model for volatility in both.

The tempting read is that the fund's implosion means the AI trade is over, the bubble has popped, and the boom was always hollow. Slow down. What broke was a specific, extreme structure: a concentrated, correlated book levered up to 400%, meeting an ordinary correction. That is a risk-management failure, not a referendum on AI demand. In the same window that the stock cratered, SK Hynix grew revenue 257% year over year, private AI funding set records, and the fund itself stayed up about 80% on the year on its private holdings. The fundamentals did not collapse. The leverage did.

But do not take the opposite comfort either. The episode is a genuine signal about fragility, just not the one the headline implies. It shows how quickly a sector priced for perfection reprices when a single data point disappoints, and how tightly the AI infrastructure names are correlated on the way down. Your compute supply rides those same names. The lesson is not that AI is finished. It is that the road under it is bumpier than the last two years suggested.

Situational Awareness was the purest expression of that story, and its collapse is the wake-up. The enterprise takeaway is not about hedge funds. It is that most AI roadmaps quietly embed the same assumption the fund did: that compute keeps getting cheaper and more available on a smooth line, and that the vendors serving you will keep passing along discounts underwritten by an endlessly rising capital cycle. That assumption held beautifully until late July. It may hold again for years. But a plan that only works if the line keeps going up is not a plan, it is a bet with the same shape as a 4x-levered chip book. The teams that will be steady are the ones who diversified their model and compute suppliers before they had to, locked the capacity they genuinely need, and built a budget that survives a quarter where prices stop falling.