Hook
A 13F filing dropped on August 14, 2026, and it wasn't just another quarterly disclosure. It was the autopsy of a conviction so deep it took down a $20 billion fund. The Situational Awareness Fund, run by former OpenAI researcher Leopold Aschenbrenner, revealed a portfolio where two chip stocks—SanDisk and Micron—made up 55.5% of total holdings. By the time the SEC made this public, the fund had already been gutted. Citadel had stepped in to take over a "problematic stock portfolio" after a forced liquidation in July. We didn't see the crash coming—we saw the wreckage.
Context
Leopold Aschenbrenner isn't your typical hedge fund manager. He was a key figure in OpenAI's "superalignment" team, leaving in 2024 over philosophical disagreements about AI safety. His 2024 essay "Situational Awareness" argued that AI compute is the new geopolitical currency. That worldview became his investment thesis. The fund, structured as a standard LP, piled into the infrastructure layer of AI: storage, power, cloud, and—most controversially—bitcoin miners pivoting to AI data centers. The 13F, filed on August 14, 2026, shows a snapshot as of June 30, 2026: total disclosed equity holdings of ~$20.24 billion. But the market already knew the fund had blown up in July, triggered by a pullback in AI stocks and margin pressure. Open source isn't just a code philosophy—it's a transparency that reveals brutal truths after the damage is done.
Core
Let's break down what this fund actually owned, because the numbers are staggering.
Layer 1: Storage gorillas. SanDisk ($5.674B, 28.0%) and Micron ($5.574B, 27.5%) together account for 55.5% of the portfolio. This is a bet on HBM (high-bandwidth memory) and NAND flash as the bottlenecks for AI training and inference. Micron, along with SK Hynix and Samsung, dominates the HBM market. SanDisk is a leader in NAND. The logic: as AI models grow, storage demand grows exponentially. But the semiconductor industry is brutally cyclical. One demand shock and these two stocks can drop 30% in a quarter.
Layer 2: Foundry and power. TSMC ADR (6.2%) and Bloom Energy (9.4%). TSMC makes the chips that run AI. Bloom Energy provides fuel cells for data center power. This is a bet on the physical constraints of AI expansion: you need chips and you need electricity.
Layer 3: Cloud and compute. CoreWeave (5.3%) and Nebius (4.5%). CoreWeave is a GPU cloud provider that has signed multi-billion-dollar contracts with OpenAI. Nebius is an AI-native cloud platform. These are the direct beneficiaries of GPU rental demand.
Layer 4: Bitcoin miners turned AI hosts. Core Scientific, Applied Digital, IREN, Riot Platforms, CleanSpark—together roughly 15% of the portfolio. This is the most creative part of the thesis. Bitcoin miners own land, power substations, and cooling infrastructure. In 2024-2025, many began repurposing facilities for AI data center hosting. The fund bought into that narrative. But these are small-cap, high-volatility stocks. In a forced liquidation, they are the first to get crushed.
Total top-7 holdings concentration: ~84.3%. The top two alone are 55.5%. A typical institutional fund has a CR10 of 20-40%. This fund had concentration levels 2-3x normal. A philosophy of transparency means we can see the risk, but we can't see the leverage. The 13F doesn't disclose margin loans, derivatives, or short positions. Based on my audit experience with early Ethereum prediction markets, I know that when a portfolio is this concentrated, any leverage above 2x becomes a death spiral during a 15% drawdown. The market reported that the fund was forced to sell most of its public equity holdings in July due to "leverage pressure." That's a euphemism for margin calls.
The missing piece: no AI application layer. No OpenAI, no Anthropic, no Microsoft, no software stocks. The fund was a pure "picks and shovels" play on AI infrastructure. That works in a bull market where AI capex grows 50% YoY. But when the narrative shifts—when GPU supply catches up or when model companies cut spending—the entire stack collapses simultaneously. There's no hedge. There's no income from protocol fees. The fund's return depended entirely on capital appreciation. Art isn't just about who owns it—it's about who can survive the winter.
Contrarian
Now for the uncomfortable truth. Most analysts will call this a "smart money" conviction that got unlucky. I see it differently. This wasn't bad luck—it was a structural design flaw embedded in the worldview itself.
First, the thesis assumed AI compute demand would grow linearly forever. It didn't model the possibility of a sharp capex pause. In July 2026, a combination of Fed hawkishness, export controls on AI chips to China, and disappointing earnings from some hyperscalers triggered a 15-20% correction in AI stocks. The fund's entire portfolio was correlated. There was no diversification into software, no defensive assets, no short positions (at least not disclosed). The leverage turned a moderate correction into a catastrophic unwind.
Second, the bitcoin miner inclusion is a double-edged sword. The AI hosting narrative is real, but it introduces two layers of volatility: crypto price risk (because miners still hold Bitcoin on their balance sheets) and AI capex risk. When the AI narrative weakens, miners lose both the crypto and the AI valuation support. The fund owned Core Scientific, which had a market cap under $3 billion. Selling a position like that in a panic can move the stock 40% intraday.
Third, the 13F timeline is a joke for risk management. The filing was due 45 days after quarter end. The fund blew up in July. The filing came in August. By the time the public saw the concentration, it was already too late. This is a systemic weakness in the regulatory framework. The SEC requires disclosure for transparency, but the 45-day lag means these filings are history, not warnings. Decentralization is not a tech stack; it's a philosophy of transparency—but only if the data is timely.
Fourth, the fund's manager was an AI researcher, not a risk manager. Leopold Aschenbrenner is brilliant. His "Situational Awareness" essay is a must-read. But managing a levered, concentrated portfolio requires a different skill set. The absence of any hedging, any cash buffer, any position sizing discipline suggests that the fund was run on conviction rather than risk frameworks. This is the same pattern I saw in the 2022 crypto crashes: brilliant founders who don't understand tail risk.
Takeaway
So what do we learn? The Situational Awareness Fund's 13F is a cautionary tale for every bull market narrative. The AI infrastructure theme is real. The storage and power bottlenecks are real. But the market has a way of punishing those who forget that leverage is a two-way street. The fund's collapse didn't need a black swan—it needed a 15% drawdown in a single sector. That's not a black swan; that's a Tuesday.
Going forward, every institutional investor looking at AI infrastructure should ask: Is my portfolio designed to survive a 30% drawdown in my top holding? If the answer is no, you're not investing—you're gambling on a narrative. The philosophy of transparency demands that we look at the wreckage and build better ships. Trust, but verify. Build, but share. And don't let conviction blind you to the math.
— A former crypto auditor who once found three critical logic flaws in Augur's oracle