We didn’t just hunt alpha; we rewired the game. But sometimes the game rewires you.
Hook
On August 14th, 2026, a 13F filing landed on the SEC’s EDGAR system. It wasn't just any filing. It was the ghost of a $20.2 billion portfolio, frozen in time at the end of June. The fund was called "Situational Awareness." The manager was Leopold Aschenbrenner, a former OpenAI alignment researcher who wrote a viral essay about the geopolitical stakes of compute. By mid-July, that same fund was bleeding out, forced to sell a massive chunk of its holdings due to leverage pressure. The 13F wasn't a warning. It was an autopsy. And the body is fascinating.
Context
This isn't a DeFi protocol failure or a rug pull. It's a classic, high-conviction, concentrated bet on a single thesis: that the AI compute bottleneck is real, physical, and investable. Aschenbrenner didn't just buy Nvidia. He bought the entire supply chain. The 13F shows a portfolio that looks like a map of the AI infrastructure stack, from storage to power to the miners who pivoted to data centers. The core data: SanDisk (28%) and Micron (27.5%) alone make up 55.5% of the fund. Add in TSMC (6.2%), CoreWeave and Nebius (9.8%), Bloom Energy (9.4%), and a handful of Bitcoin miners like Core Scientific, IREN, and Riot, and you have a portfolio where the top 7 holdings account for roughly 84% of assets. This is not a diversified fund. This is a scalpel.
Core
Let’s break down the technical architecture of this bet. The fund’s thesis was clear: the bottleneck in AI isn't just the GPU. It's the high-bandwidth memory (HBM) from Micron, the NAND flash from SanDisk, the advanced packaging from TSMC, and the power from Bloom Energy. The Bitcoin miners (Core Scientific, Applied Digital, IREN, Riot, CleanSpark) were repurposed as “AI data center” plays, leveraging their power contracts and existing infrastructure. From a portfolio perspective, this is a “school of fish”—a tight cluster of correlated assets that all move in the same direction. When the thesis works, the beta is enormous. When it fails, the drawdown is catastrophic.
I’ve seen this pattern before. During the DeFi summer of 2020, I launched a localized AMM called UniBarter in Jakarta. We had 500 users in two weeks. The thesis was sound: localized liquidity for Indonesian traders. But the engineering maintenance was a nightmare, and the infrastructure wasn't ready. I learned the hard way that innovation often outpaces the underlying infrastructure. The Situational Awareness Fund made the same mistake, but at a $20 billion scale. They assumed that the physical constraints of the AI supply chain would remain tight for years. But the market is always building to solve those constraints. The moment a new fab comes online, or a new memory standard emerges, the thesis weakens.
Based on my audit experience, I can tell you that the hidden risk here is not just leverage. It's the lack of a hedge. The fund had no positions in AI application layer companies. No OpenAI, no Anthropic, no software. This means they were completely exposed to the capital expenditure (CapEx) cycle of the hyperscalers. If a major model developer decides to cut spending, the entire chain—from storage to power to miners—gets hit simultaneously. That’s exactly what happened in July. The 13F doesn't show the leverage, but the market knows it was there. The forced sale was a liquidity event triggered by a sentiment shift, not a fundamental breakdown of the AI thesis. But the distinction doesn't matter when you're getting margin-called.
Contrarian
Here’s the contrarian angle: this fund’s failure is not a signal that the AI compute thesis is wrong. It’s a signal that the market’s pricing of that thesis was too aggressive. The 13F reveals a portfolio that was built for a world where AI compute demand grows exponentially forever, with no cyclicality. But everything in hardware is cyclical. Memory chips are famously cyclical. Power contracts are long-term, but their valuation is tied to forward expectations. The Bitcoin miners, which I’ve analyzed extensively, are the most volatile of all. They are a double-edged sword: they benefit from AI demand, but they are also exposed to the price of Bitcoin and the cost of energy. The fund’s decision to bundle them into the “AI compute” narrative was clever, but it introduced a risk that wasn’t hedged.
From my trenches in Jakarta, I’ve seen a thousand projects that promised to “bridge” two worlds. The Situational Awareness Fund attempted to bridge the world of AI research with the world of public markets. The fund manager’s intellectual background—his paper on Situational Awareness—is brilliant. But the translation from that paper to a 13F file was flawed. The portfolio was a perfect reflection of a philosophical worldview, but it lacked the pragmatic adaptability that a 45-year-old market cycle teaches you. The market doesn’t care about your thesis when it’s panicking. It cares about your liquidity.
Takeaway
Education is the new mining rig for the mind. The Situational Awareness Fund’s 13F is a lesson in how conviction can become a liability. It’s a reminder that even the most brilliant technical analysis can be undone by a single overlooked variable: leverage. The fund wasn’t wrong about the AI compute bottleneck. It was wrong about the timing, the concentration, and the lack of a safety net. When the market sleeps, the architects wake up. But even the best architects need to build a fire escape. The question isn’t whether the AI compute thesis is valid. It’s whether your portfolio can survive the correction long enough to see it play out.