The code remembers what the market forgets. On August 14, 2026, the SEC published a 13F filing from a fund called Situational Awareness LP, managed by Leopold Aschenbrenner—the former OpenAI researcher turned AI narrative evangelist. The form showed a portfolio worth $20.24 billion as of June 30. But by July, the fund had already collapsed under leverage, its assets seized by Citadel. The filing arrived two weeks late, a corpse still wearing its smile.
I spent the morning in Buenos Aires tracing the ghost in the machine. The numbers tell a story that is both familiar and alien to anyone who has watched a high-conviction crypto fund blow up. SanDisk and Micron alone accounted for 55.5% of the portfolio—$11.25 billion in storage chips. The rest was a tightly woven bet on the AI infrastructure supply chain: TSMC (6.2%), CoreWeave and Nebius (9.8% combined), Bloom Energy (9.4%), and a tail of Bitcoin miners like Core Scientific, IREN, and Riot (roughly 15%). This is not a diverse portfolio. It is a single narrative, written in code and capital, waiting for the algorithm to break.
I have seen this pattern before. In 2021, I audited the Uniswap V1 whitepaper and learned that liquidity providers are not traders—they are trust machines. The same principle applies here: the fund’s owners were not investors. They were narrative participants, betting that the AI bottleneck would tighten for years. The thesis had intellectual merit. Aschenbrenner’s “Situational Awareness” essay argued that compute is the new currency of power, and that control over storage, power, and chip fabrication is the true leverage point. The 13F was a quantified version of that worldview. But the quiet ruin began when the market stopped believing in the timeline.
The core insight is the concentration. The Herfindahl-Hirschman Index of this portfolio is roughly 3,000—a level that would trigger antitrust scrutiny in a market, but here it is simply a signal of conviction. The fund’s top two holdings (SanDisk and Micron) are both cyclical memory stocks. HBM demand is real, but the storage cycle is notorious for violent swings. The fund had no hedge. No software, no AI application layer, no defensive assets. It was a pure beta play on the assumption that AI CapEx would grow exponentially forever. When the AI trade pulled back in July, the levered structure amplified the decline. The margin calls came. Citadel took over the “problematic portfolio.” The code broke.
The contrarian angle is that the thesis was right, but the structure was wrong. AI infrastructure is indeed a bottleneck. Storage, power, and data center space are real constraints. But the fund’s fatal flaw was not the narrative—it was the leverage and the lack of diversification. Every crypto veteran knows the smell of a levered, high-concentration fund. I felt it in the Terra collapse, when the algorithmic stablecoin narrative fractured under the weight of its own incentives. Here, the same pattern repeats: a brilliant idea, a high-conviction portfolio, and a financing structure that assumes the market will never test the thesis. The market always tests the thesis.
Quantitatively, the portfolio’s Sharpe ratio during the upswing must have been spectacular. But in a drawdown, the non-linearity of concentrated leverage creates a liquidity death spiral. The miners in the tail—Core Scientific, Applied Digital, IREN—are low-float, high-volatility names. Selling them in a panic would have crushed their prices, amplifying the loss. The 13F snapshot does not show the July carnage, but the math is inevitable. The fund’s “AI bottleneck” narrative became a self-fulfilling prophecy in reverse: as the market questioned the speed of AI adoption, the entire portfolio re-rated downward simultaneously.
The takeaway is not about AI, but about narrative risk management. The herd always wakes when the signal has already faded. The market’s next narrative will not be about storage or power—it will be about capital efficiency. Funds that survive the next cycle will be those that can hold their convictions while respecting the volatility of human belief. The code remembers what the market forgets: that every high-conviction bet is a bet against the noise. And the noise always wins, eventually.
I am writing this from a café in Palermo, where the silence between the blocks is thick with the memory of leverage. The 13F is a ghost, but it teaches a lesson that every token fund manager should learn: the smart contract doesn’t care about your thesis. The market does not care about your conviction. It only cares about liquidity. And when the liquidity disappears, the narrative is the first thing to die.