The 13F That Killed a Fund: How a $20B AI Bet Imploded by August

On August 14, 2026, a 13F landed on the SEC's EDGAR system. It showed a $20.2 billion portfolio. By the time the filing was public, that portfolio was already under Citadel's control. The filing wasn't a warning. It was an autopsy. I didn't need the 13F to know this fund was doomed, but the numbers confirm the story. The fund was run by Leopold Aschenbrenner, former OpenAI researcher turned hedge fund manager. His thesis was simple: AI compute is the bottleneck of the decade. Bet on the hardware, the energy, and the infrastructure. The 13F, filed August 14, 2026, for the quarter ending June 30, reveals the exact structure of that bet. It's a concentrated nightmare. Two storage stocks—SanDisk and Micron—make up 55.5% of the portfolio. Add Bloom Energy, TSMC, CoreWeave, Nebius, and a handful of Bitcoin miners—Core Scientific, Applied Digital, IREN, Riot Platforms, CleanSpark—and you have 84% of the fund in seven names. The rest is scattered. No software. No AI application layer. No hedge. Let's break down the numbers. SanDisk: $5.674 billion, 28.0%. Micron: $5.574 billion, 27.5%. That's $11.2 billion in two stocks. Both are mature, cyclical semiconductor companies. They make NAND flash and HBM memory. Demand is strong, but the cycle is turning. The market already priced in the AI boom. The fund bought at the top of the narrative. Then you have Bloom Energy at $1.9 billion, 9.4%. Fuel cells for data centers. TSMC at $1.26 billion, 6.2%. The chipmaker. Nebius and CoreWeave together at $1.98 billion, 9.8%. Cloud GPU services. And the miners: Core Scientific, Applied Digital, IREN, Riot, CleanSpark—roughly 7% combined. These are small-cap, high-volatility names. Liquidity is thin. In a sell-off, they bleed first. Now, the time dimension. The 13F is a snapshot. June 30. The fund collapsed in July. The market reported that AI stocks fell sharply, and leverage forced the fund to sell. The 13F doesn't show the leverage. It doesn't show the shorts. It doesn't show the derivatives. But the sequence is clear. The fund was levered. The concentration meant that a 20% drop in SanDisk and Micron wiped out a huge chunk of equity. The miners, being illiquid, dropped faster. The margin calls came. Citadel stepped in to take over the "problematic portfolio." This is a standard rescue—a total return swap unwinding, or a prime broker stepping in to liquidate. The fund is dead. The 13F is the corpse. This is not a story of bad luck. It's a story of structural failure. The fund's thesis was that AI compute bottlenecks are permanent. But bottlenecks are temporary. HBM supply is ramping. New fabs are coming online. Energy constraints are real, but utilities are building. The narrative that the fund bought into was already peaking. The market was pricing in years of growth, but the fund bought at the top of the hype cycle. The 13F shows no hedging. No commodities. No short positions. No index puts. Just a long, concentrated bet on the hardware layer. When the AI trade rotated, the fund was the first to break. Let's talk about the Bitcoin miners. The market loves the AI transformation narrative. Core Scientific, IREN, Riot—they are all repositioning as AI data center operators. The 13F shows the fund bought into that story. But here's the problem: these are not AI companies. They are energy assets with a crypto hedge. Their revenue is still tied to Bitcoin mining, and their AI hosting contracts are early stage. The fund treated them as high-beta AI plays. That's a mistake. When the AI trade unwinds, the miners lose both the crypto and AI premium. They are the most fragile part of the portfolio. The 13F shows over $1.4 billion in miners. In a forced liquidation, those positions would take days to sell without moving the market. The fund couldn't exit fast enough. Leopold's story is a lesson in overconfidence. The fund's name—Situational Awareness—was supposed to reflect a deep understanding of the AI landscape. But the 13F reveals a lack of awareness of basic portfolio risk. A 55% concentration in two stocks is not a trade. It's a bet. The fund ignored the software layer. No OpenAI, no Microsoft, no AI application companies. The thinking was that the infrastructure is the better bet. But the infrastructure is the most capital-intensive, most cyclical part of the stack. In a downturn, the infrastructure gets cut first. The 13F shows no diversification across the value chain. The fund was betting on the entire AI compute ecosystem, but it only owned the most leveraged parts. Now, the market context. The 13F was filed on August 14, but the crisis happened in July. The market already knew about the fund's troubles. The AI sector had a sharp correction in late July, driven by liquidity concerns. The fund's forced selling likely amplified the drop. The 13F confirms the concentration, so the market now knows exactly which stocks are still under pressure. Citadel took over the portfolio, but they are not holding it. They will unwind it. The miners are the most at risk. Small-cap, low liquidity, high volatility. If you see a sudden spike in volume on Core Scientific or IREN, that's the liquidation. The market is already pricing in 60-70% of the event. But the tail risk remains. If Citadel is forced to sell into a thin market, the impact could be larger than expected. The infrastructure is the reality. The narrative is the noise. The 13F shows a fund that bought the narrative at the peak. The reality is that storage cycles turn, energy prices fluctuate, and AI capex is not infinite. The fund's thesis was correct in the long term, but the timing and structure were wrong. The market is a machine. You don't fight the machine. The machine demands diversification, risk management, and liquidity. This fund had none of that. Takeaway: Watch the next 13F filings from Citadel. They will show the unwinding. If the miners disappear from the portfolio, expect a multi-week overhang. If they appear as new positions, Citadel is holding for a better exit. Either way, the lesson is clear. High concentration plus leverage equals fragility. I've seen it in crypto, I've seen it in equities. The pattern is the same. The 13F is the post-mortem. The next one will be the execution. Final thought: When you see a 13F with a CR2 of 55%, ask yourself: what happens when the narrative shifts? The answer is already written. The fund's collapse is a warning for all of us. The AI trade is not dead, but it's wounded. The market will remember this. The next time a fund shows a similar concentration, short the liquidity. The margin call is coming.