Hook
On August 14, 2026, the SEC received a 13F filing from the Situational Awareness Fund, a $20.24 billion portfolio managed by Leopold Aschenbrenner—the former OpenAI safety researcher turned hedge fund manager. The filing, dated June 30, showed a staggering 55.5% in just two positions: SanDisk and Micron. But by the time the document was made public, the fund had already been dismantled. In July, a cascade of leverage-driven liquidations forced Citadel to take over its “problematic stock portfolio.” The market had already moved on, but the 13F remains a frozen echo of a strategy that aimed to capture the “AI bottleneck” thesis—and instead became a textbook case of how narrative-heavy, high-concentration bets can implode when liquidity dries up.
This is not a story about a bad thesis. It is a story about the silent code behind the noisy market: the structural fragility of leverage, illiquid positions, and a narrative that became too tightly wound into a single thread.
Context
Leopold Aschenbrenner rose to prominence in 2024 after leaving OpenAI’s “superalignment” team, citing ideological differences over AI safety. His subsequent essay, “Situational Awareness,” argued that compute—not just algorithms—would become the defining geopolitical asset of the AI era. He then launched a hedge fund to put that worldview into practice. The fund’s thesis was simple: invest in the physical infrastructure of AI—storage, chips, power, and data centers—and treat bitcoin miners as a low-cost entry point into the AI data center buildout.
The 13F filing, which covers 13 positions, reveals a portfolio that is both brilliant and reckless. Brilliant in its vertical integration of the AI supply chain: from storage (SanDisk, Micron) to foundry (TSMC) to cloud GPU (CoreWeave, Nebius) to power (Bloom Energy) to miner-converted data centers (Core Scientific, IREN, Riot, CleanSpark, Applied Digital). Reckless in its concentration: the top two holdings alone represent 55.5%, and the top seven—including TSMC, Bloom, CoreWeave, and Nebius—account for 84.3% of the portfolio. For reference, most institutional funds keep their top 10 holdings under 30-40%. This fund had a CR2 of 55.5% and a CR7 of 84.3%.
The fund’s structure was a conventional limited partnership, but its leverage was anything but conventional. By late July, as AI-related stocks fell, the fund faced margin calls that forced it to sell large portions of its public holdings. Citadel stepped in to take over the “problematic stock portfolio,” a move that hints at structured derivatives—perhaps total return swaps—being unwound.
Core Analysis: The Silent Code of Structural Fragility
When I first reviewed the 13F, I was struck by something that the market headlines missed. The collapse was not simply a function of AI stock declines. It was a function of portfolio architecture—a design that maximized beta in an uptrend but created a non-linear downward spiral during a drawdown.
Tracing the silent code behind the noisy market.
Let me break down the layers:
Layer 1: Concentration Risk as a Hidden Lever. The 55.5% in SanDisk and Micron is not just a bet on storage—it is a bet that the entire AI demand narrative will remain intact. Storage is a cyclical industry. HBM (high-bandwidth memory) is indeed a bottleneck for AI training, but the supply of HBM is expanding rapidly. Micron, SK Hynix, and Samsung are all ramping production. If the bottleneck eases, the pricing power of these companies erodes. The fund had no hedge—no short positions, no software layer exposure, no diversification into AI application stocks. From my experience auditing smart contracts, I learned that concentration is not a sign of conviction; it is a sign of untested assumptions. The code of the market has a way of exposing those assumptions under stress.
Layer 2: The Illiquidity of Miners as Tail Risk. The fund held roughly 7-8% of its portfolio in bitcoin miners that had pivoted to AI data center hosting: Core Scientific, IREN, Riot, CleanSpark, and Applied Digital. These are small-cap, high-volatility stocks with thin order books. In a forced liquidation, these positions suffer the most price impact. The 13F shows them as of June 30, but by July, the market capitalization of many of these names had likely dropped 30-50% as the fund unwound. The market absorbed the SanDisk and Micron sales with moderate impact, but the miner positions probably collapsed under their own weight.
Layer 3: Leverage as the Unseen Variable. The 13F does not disclose leverage, but the July events confirm it existed. The key question is: how much? The fact that Citadel took over the “problematic stock portfolio” rather than simply executing a margin call suggests that the fund had structured financing—perhaps through prime brokerage or total return swaps. In my years as a blockchain engineer, I saw similar patterns in DeFi protocols: when a leveraged position is concentrated in a correlated set of assets, a small drop in collateral value can trigger a cascade of liquidations. The Situational Awareness Fund was a real-world analog of a DeFi vault with no isolated risk parameters.
A hunter’s gaze into the algorithmic soul.
Let me quantify the fragility. The portfolio’s Sharpe ratio in an uptrend would be exceptionally high due to concentration. But during a downturn, the portfolio’s correlation structure becomes a death spiral. All the holdings are tied to the same narrative: “AI compute demand grows exponentially.” There is no diversification across different macro regimes. If AI capital expenditure slows—even temporarily—every position suffers simultaneously. The fund’s portfolio beta to the AI infrastructure theme was likely above 1.5, meaning that a 10% decline in the AI sector could translate into a 15-20% decline in the fund’s NAV. Add leverage of 2x or 3x, and a 10% sector decline becomes a 30-60% loss.
The core insight here is that the fund’s thesis was not wrong, but its execution was structurally fragile. The market is now pricing in the risk of similar leveraged-concentration portfolios. This is the silent code behind the noisy market: the 13F is not just a historical record; it is a warning signal for the entire AI infrastructure asset class.
Contrarian Angle: The Thesis Was Right, the Execution Was Flawed
Most market commentary will frame this as a failure of the “AI bottleneck” narrative. I disagree. The narrative is still valid. Storage, power, and data center capacity are real bottlenecks. The contrarian angle is that the market is overcorrecting: it will now assume that any concentrated AI infrastructure fund is risky, and that may lead to undervaluation of the underlying assets.
But the real blind spot is not the thesis—it is the portfolio construction.
Here is what the market is missing: the fund’s collapse was not a refutation of the AI infrastructure theme. It was a refutation of unhedged concentration combined with opaque leverage. The same thesis could be executed with better risk management: diversification across the supply chain, partial hedging via options or short positions on correlated sectors, and lower leverage. The market will now penalize funds that look like this one, but that may create opportunities for those who can separate the signal from the noise.
Another contrarian point: the inclusion of bitcoin miners was actually a clever idea—they are essentially call options on power and data center real estate. But the miners in this portfolio were too small and illiquid for a fund of this size. The mistake was not the asset class, but the sizing. A smaller allocation to more liquid miners (like Riot, which is relatively larger) or a passive approach (via an ETF) would have been less destructive.
The silent code here is that the market’s narrative about AI infrastructure is still intact, but the capital allocation mechanism is broken. The fund’s failure will create a temporary discount on the very assets it held, especially the miners. Astute investors who understand the structural integrity of the underlying thesis may find entry points.
Takeaway: The Next Narrative
What does this mean for the broader market? The Situational Awareness Fund’s 13F is a relic of a specific era—when high conviction, low diversification, and high leverage were rewarded. The market is now entering a new phase where risk management will be the differentiator, not narrative purity.
I expect to see two trends: first, a rotation toward more diversified AI infrastructure funds (e.g., ETFs or multi-asset vehicles) that offer exposure without the tail risk. Second, increased scrutiny of 13F filings for concentration and leverage proxies—investors will start looking for “Leopold-like” patterns in other funds.
The takeaway is not to abandon the AI bottleneck thesis, but to build portfolios that can survive a drawdown. The code of the market is rewriting itself: leverage amplifies both gains and losses, but concentration amplifies only the latter when the narrative shifts. The next narrative will be about resilience—not just which assets win, but how they are held.
One final thought: If the fund’s thesis was correct, then the assets it held—SanDisk, Micron, TSMC, Bloom, CoreWeave, and even the miners—are likely to recover. The collapse was a liquidity event, not a fundamental one. But the scars will remain. The market will remember that even the smartest narrative can be broken by the silent code of leverage.