August 14, 2026. The SEC received a 13F filing from the Situational Awareness Fund. On paper, it showed a $20.2 billion portfolio as of June 30. But the market already knew this fund had collapsed in July. The filing is a post-mortem, not a preview. And what it reveals is a textbook case of conviction without risk management.
This is not a story about a failed crypto fund. It is a story about a fund that bet everything on a single narrative — the AI compute bottleneck — and lost. The 13F provides the raw data. I provide the forensic dissection.
Context: The Man and the Thesis
Leopold Aschenbrenner is not a traditional hedge fund manager. He came from OpenAI's “superalignment” team, left in 2024 over safety disagreements, and published a widely-read essay titled “Situational Awareness” that argued control of compute is the ultimate geopolitical lever. The fund was his attempt to turn that worldview into a portfolio.
The thesis was simple: AI scaling requires exponentially more compute, and the bottlenecks are not just GPUs but storage, power, and data center infrastructure.
Core: The 13F Breakdown
Let’s go straight to the numbers. The filing shows 13 positions as of June 30, 2026. Total disclosed US equity value: approximately $20.24 billion. The concentration is staggering:
- SanDisk: $5.674 billion (28.0%)
- Micron Technology: $5.574 billion (27.5%)
- Bloom Energy: $1.9 billion (9.4%)
- TSMC ADR: $1.25 billion (6.2%)
- Nebius: $1.0 billion (4.9%) – AI cloud native
- CoreWeave: $0.99 billion (4.9%) – GPU cloud
- Core Scientific: $0.8 billion (4.0%) – Bitcoin miner turned AI data center
- Applied Digital: $0.5 billion (2.5%) – Another miner/AI hybrid
- IREN: $0.4 billion (2.0%) – Miner
- Riot Platforms: $0.3 billion (1.5%) – Miner
- CleanSpark: $0.3 billion (1.5%) – Miner
Total: $20.24 billion. CR2 (SanDisk + Micron) = 55.5%. CR7 = ~84.3%.
This is not a diversified portfolio. This is a concentrated bet on three pillars: storage, power, and miners-as-data-centers.
Liquidity doesn't hide in large caps; it hides in the tails. The miner positions — Core Scientific, Applied Digital, IREN, Riot, CleanSpark — together represent about 11.5% of the portfolio. But they are small-cap, high-volatility names. In a forced liquidation, these positions would suffer the most slippage.
The Leverage Trap
The 13F does not disclose leverage. But the market knows the fund was forced to sell in July amid the AI stock rout. Multiple reports confirm “leverage pressure” led to the liquidation of most public holdings. Citadel stepped in to take over a “problem portfolio.” This is not a margin call. This is a structured unwind — likely total return swaps or prime brokerage financing that hit a trigger point.
Here’s the critical insight: the 13F shows $20.2 billion in equity. But the fund’s actual gross exposure was likely much higher. Arbitrage is the market’s way of correcting mispricing. Here, the arbitrage was between the fund’s conviction and reality. The market won.
Contrarian: The Blind Spot No One Is Talking About
The fund’s thesis — that storage, power, and physical infrastructure are the true AI bottlenecks — has surface-level logic. But it ignores a critical reality: the market prices AI infrastructure based on GPU supply, not storage or power. The marginal shock to AI sentiment in July came from reports of GPU oversupply and CapEx cuts by hyperscalers. When that hit, the entire thesis cracked.
Worse, the fund held zero AI application layer stocks. No OpenAI, no Anthropic, no software names. The fund was long the “pick and shovel” sellers but had no exposure to the gold miners themselves. That means it captured the beta of AI CapEx volatility without any offset from downstream revenue growth.
The Bitcoin miners were the kryptonite. They are not just AI plays; they are crypto plays. When AI sentiment turned, they got hit twice — once by the AI narrative, once by Bitcoin’s correlation with risk assets. The miner positions acted as a volatility multiplier, not a diversifier. The fund’s thesis assumed miners would be valued as data center operators, but the market still prices them as crypto proxies. The 13F shows clearly that the fund bought the narrative, not the market structure.
Takeaway: What to Watch Next
The 13F is a snapshot of a dead portfolio. Citadel now controls the residual positions. The key question: will Citadel liquidate the remaining miner holdings in the open market or via block trades? My experience in market surveillance says watch for off-exchange prints in Core Scientific and IREN. If those appear, it signals the final unwind.
For the broader market, this is a warning sign. The “AI bottleneck” trade is not immune to leverage cycles. The fund’s collapse is not an isolated event — it’s a symptom of overcrowding in a narrative that everyone agrees on. When everyone is betting on the same bottleneck, the bottleneck becomes the trade itself, not the underlying asset.
The next 13F filing from Citadel, due in mid-November, will show whether they kept any of these positions. If they dumped everything, expect a final wave of selling pressure on AI infrastructure stocks. The lesson: conviction without leverage is a thesis. Conviction with leverage is a time bomb.
I’ve spent years in market surveillance, tracking institutional flows. I’ve seen this pattern before — concentrated bets with leverage, a narrative that becomes a self-fulfilling prophecy until it isn’t. The 13F is not just a disclosure; it’s a forensic document. It tells us where the bodies are buried. Now the question is whether anyone else is buried in the same grave.