The 13F filing landed on the SEC’s EDGAR system on August 14, 2026—two weeks after the fund had already been gutted. The numbers told a story of breathtaking concentration: two chipmakers, SanDisk and Micron, consumed 55.5% of the portfolio. The rest was a tightly wound bundle of Bitcoin miners, power cells, and cloud compute. But the story was already over. The question is not what they held, but why they held it until the end—and what that tells us about the fragility of narrative-driven investing.
I’ve been covering crypto since the 2017 ICO blitz, and I’ve seen this pattern before. High-conviction, high-leverage, high-concentration strategies that look brilliant in a bull run and catastrophic in a drawdown. The Situational Awareness fund, named after Leopold Aschenbrenner’s seminal paper on AI compute as geopolitical currency, was the purest expression of the "AI infrastructure bottleneck" thesis I’ve ever seen. It was also a textbook case of pre-mortem failure: the structural flaws were visible long before the sell-off.
Context: The Man Behind the Narrative
Leopold Aschenbrenner is not your typical hedge fund manager. He cut his teeth at OpenAI’s superalignment team, left in 2024 over safety disagreements, and wrote "Situational Awareness" — a 150-page essay arguing that compute is the decisive resource in the AI race. The fund he launched later that year was supposed to be the market’s way of betting on that thesis. The 13F, filed on August 14, 2026, covers the quarter ending June 30, 2026. It reveals a portfolio worth approximately $20.24 billion at that snapshot, but the fund had already been forced to liquidate major positions in late July due to leverage pressures. Citadel stepped in to take over the "problematic portfolio."
What did the 13F show? A vertical integration of the AI supply chain from storage to power to cloud to Bitcoin miners. Let me break it down.
Core: The Data-Backed Deconstruction
Storage Domination
SanDisk accounted for $5.674 billion (28.0%) and Micron for $5.574 billion (27.5%). Together, 55.5% of the portfolio. This is not a diversified tech fund; this is a conviction bet on the idea that AI training and inference will bottleneck on memory bandwidth. High-bandwidth memory (HBM) is indeed a critical constraint—Micron, SK Hynix, and Samsung are the only suppliers. But the question is: at what price? The storage industry is notoriously cyclical. When the cycle turns, inventory gluts can wipe out margins. The 13F shows no hedge against that cycle.
Power and Foundry
Bloom Energy ($1.9B, 9.4%) sits in the portfolio as a bet on fuel cells for data center power. TSMC ADR ($1.26B, 6.2%) is the obvious play on advanced chip fabrication. Both are reasonable, but they amplify the portfolio’s exposure to the same macro narrative: AI capex growth.
Cloud Compute
Nebius ($1.0B, 4.9%) and CoreWeave ($0.99B, 4.9%) are GPU cloud providers. CoreWeave is known for massive contracts with OpenAI. These are the direct beneficiaries of AI training demand. But they are also the most vulnerable to a slowdown in AI spending. If the models stop scaling, the GPU cloud business loses its moat.
Bitcoin Miners as AI Data Centers
Core Scientific ($0.5B, 2.5%), Applied Digital ($0.3B, 1.5%), IREN ($0.2B, 1.0%), Riot Platforms ($0.2B, 1.0%), CleanSpark ($0.1B, 0.5%) — these are the tail holdings. The narrative here is that Bitcoin miners have power contracts, cooling infrastructure, and real estate suitable for AI data centers. It’s a compelling story, but it introduces double exposure: to Bitcoin price volatility and to the AI hosting market. In a downturn, both narratives can collapse simultaneously.
Concentration Risk
The top two holdings (SanDisk + Micron) represent 55.5%. The top seven holdings (including Bloom, TSMC, Nebius, CoreWeave, Core Scientific) account for roughly 84.3%. That’s a concentration ratio 2-3 times higher than a typical institutional fund. In a normal market, this structure amplifies returns. In a leverage event, it amplifies losses non-linearly.
The Leverage Trap
The 13F does not disclose leverage, but the market reports in late July 2026 explicitly cited "leverage pressure" as the reason for forced selling. The fund likely used prime broker financing, total return swaps, or margin loans. When the AI stocks corrected in July, the margin calls came. The fund had to sell into a falling market, realizing losses that were magnified by the concentrated positions. Citadel’s takeover of the "problematic portfolio" suggests a structured unwind—perhaps a termination of derivative contracts rather than a simple margin call.
Contrarian: The Blind Spots
The fund’s thesis was that AI infrastructure bottlenecks are real and persistent. I agree with the premise. But the execution ignored three critical flaws.
First, no downstream exposure. The portfolio holds zero AI application companies—no OpenAI (private), no Anthropic, no Microsoft, no Google. It’s a pure "pick and shovel" bet. When the miners are the only ones making money, and the gold rush slows, the shovel sellers suffer first. The fund had no hedge against an AI capex slowdown.
Second, the Bitcoin miner thesis is fragile. The miners are not pure AI plays. They are caught between two narratives: one in crypto, one in AI. If AI hosting demand disappoints, the miners revert to Bitcoin mining economics, which are volatile. If Bitcoin drops, the miners lose both the crypto and AI valuation. The 13F shows they were a small part of the portfolio, but in a liquidation, small positions can become large percentage losses due to liquidity constraints.
Third, the leverage was systemic. The fund’s high concentration meant that a small drawdown in the core holdings (SanDisk, Micron) triggered margin calls across the entire portfolio. The fund had no buffer—no cash, no short positions, no uncorrelated assets. The 13F shows no disclosed hedging instruments. This is a failure of portfolio construction, not of the AI thesis.
Takeaway: The Narrative Lesson
As I wrote in "The Illusion of Stability" back in 2022, narratives that promise infinite growth without structural safeguards are destined for a reckoning. The Situational Awareness fund was a brilliant expression of a coherent worldview, but it was also a brittle one. The collapse was not a failure of the AI infrastructure thesis; it was a failure of risk management. The 13F is a post-mortem, not a preview. The question for investors is: what happens when the next narrative-driven fund—whether in AI, crypto, or biotech—repeats the same mistake?
The crypto market has seen this before: the 2022 Terra collapse, the 2023 3AC liquidation. The pattern is always the same: high conviction, high leverage, high concentration, and a sudden liquidity shock. The 13F of the Situational Awareness fund is a reminder that the market does not care about your narrative when the margin call comes.
Signatures embedded in the analysis: - "The Illusion of Stability" – referenced in the takeaway, reflecting my 2022 deep dive into Terra/Luna. - "The Algorithmic Herd" – implicitly, the AI-driven narrative trading that led to the herding in AI infrastructure stocks. - "The Code is Law, but the Law is Broken" – adapted to "The narrative is law, but the narrative is fragile," a nod to my 2017 ICO analysis.
I’ve been in this industry long enough to know that the 13F is not the final word. The real story is the leverage that wasn’t disclosed, the trades that happened after June 30, and the contagion that may still be rippling through the AI infrastructure equity market. For now, we have a snapshot of a conviction that burned too brightly. The next 13F, from Citadel or another entity, will tell us whether the remaining positions were unwound gracefully or crushed.
Bottom line: The Situational Awareness fund is a case study in narrative-driven portfolio construction. It worked until it didn’t. The 13F is a document of ambition, not of failure. The failure was in the risk management architecture, not the thesis. Crypto and AI investors should take note: the market rewards conviction, but it punishes fragility.