The 13F hit the SEC feed on August 14. A snapshot of June 30. A fund that was already dead on arrival.
Leopold Aschenbrenner’s Situational Awareness Fund — named after his sprawling essay on AI geopolitics — had filed its required disclosure. The numbers: 55.5% of total holdings in two storage chip stocks. SanDisk and Micron. Another 9.8% in cloud providers CoreWeave and Nebius. 9.4% in fuel cell maker Bloom Energy. And a tail of Bitcoin miners transitioning to AI data centers: Core Scientific, Applied Digital, IREN, Riot, CleanSpark.
Total reported value: $20.24 billion. A portfolio that reads like a manifesto on AI compute bottlenecks.
Code is law, but logic is fragile.
It’s also a post-mortem. By July, the fund was bleeding. Margin calls. Leverage pressure. Forced liquidation. Citadel stepped in to take over the “problem portfolio.” The 13F is a corpse with a timestamp.
Context: The Narrative Hunter’s Trap
Aschenbrenner was no trader. He was an AI safety researcher from OpenAI’s superalignment team. He left in 2024, wrote a viral essay about compute as the ultimate strategic resource, and launched a fund to monetize that conviction. The 13F shows he was all-in on the bottleneck thesis: storage (HBM, NAND), power (fuel cells, grid capacity), and compute rental (GPU cloud). He even wrapped Bitcoin miners into the same narrative — their underutilized power and data centers could be repurposed for AI.
It was a clean story. Too clean.
Core: The Machinery of Collapse
Let’s audit the structure. The portfolio had a Herfindahl index that would make a single-asset ETF blush. The top two holdings — SanDisk and Micron — accounted for $11.25 billion combined. That’s 55.5% of the entire fund. Add in Bloom Energy, TSMC ADR, Nebius, CoreWeave, and Core Scientific, and the top seven hit 84.3%.
Concentration is not a sin. Leverage is.
The 13F doesn’t show leverage. It never does. But the July liquidation tells us the margin was there. My own experience auditing DeFi composability in 2020 taught me a simple rule: when a fund’s net asset value is tied to a single narrative vector, any shock to that vector becomes a blowtorch. The AI stock correction in July was that blowtorch.
Trust no one. Verify everything.
The technical logic behind the portfolio was not wrong. AI training does rely on HBM from Micron, on NAND from SanDisk, on power from Bloom, on compute from CoreWeave. The problem is that these are all supply-side bets. There is no hedge. No software layer. No AI application stocks. The fund was long the entire compute stack, from sand to server.
When the narrative shifted from “AI infrastructure is scarce” to “AI spending may slow,” every position moved in the same direction. The miners were the worst of both worlds: small-cap, low liquidity, and tied to Bitcoin’s idiosyncratic volatility. Riot and CleanSpark probably suffered the most during the liquidation.
⚠️ Deep article forbidden
Contrarian: The Blind Spot in the Corpse
The market will read this 13F and conclude: avoid concentrated AI infrastructure plays. That’s the obvious take. But the contrarian angle is that the bottleneck thesis itself is still intact. The fund failed not because the thesis was wrong, but because the execution was structurally unsound.
Aschenbrenner’s mistake was not the narrative. It was the packaging. He used a traditional hedge fund structure with opaque prime broker leverage, no downside protection, and a portfolio that was essentially a single-factor model. In a bull narrative, such funds outperform. In a correction, they disintegrate.
Another blind spot: the Bitcoin miners were not a crypto play. They were a power play. The fund was betting on AI’s hunger for electricity. But by holding miners, Aschenbrenner inadvertently introduced a second narrative dependency — if Bitcoin drops, miner profitability drops, and their AI transformation story becomes harder to fund. The sell-off in July may have been amplified by exactly that cross-contamination.
Takeaway: The Next Narrative Will Be More Careful
This 13F filing is a warning, not a guide. It shows that even the most intelligent conviction can be destroyed by leverage and concentration. The next version of this trade will be more modular: lower leverage, broader diversification, and a hedge against narrative reversal.
What happens when the next narrative shift hits an equally concentrated portfolio?
