The 13F landed on the SEC’s EDGAR system on August 14, 2026. It was a routine filing, 45 days overdue by the standard clock. But for anyone who had been watching the market’s July jitters, the numbers told a story that the headlines had already buried. The filing was for Situational Awareness Fund, LP—a vehicle managed by Leopold Aschenbrenner, the former OpenAI researcher turned macro investor. The filing showed a portfolio of exactly 11 holdings, with a total reported market value of $20.24 billion as of June 30.
That number is a relic. By July 31, the fund had been gutted. Citadel had stepped in to take over its “problematic stock portfolio.” The narrative had been written in real time: AI stocks sold off, leverage blew up, and a fund with a thesis that was too clever by half became a casualty. But the 13F is a snapshot, not a story. It’s my job to turn that snapshot into a forensics report.
Context: The 13F as a Lie Detector
A 13F filing is a quarterly disclosure of U.S. equity holdings for any institutional investment manager with over $100 million in assets. It’s a backward-looking document—time-stamped to the last day of the quarter, filed 45 days later. For the quarter ending June 30, 2026, the deadline was August 14. The fund filed on time.
Silence is just data waiting for the right query. Here, the silence is the 45-day gap. Between June 30 and the July sell-off, the fund’s holdings changed dramatically. The 13F doesn’t show the July trades. It doesn’t show the leverage ratios, the prime broker agreements, or the derivatives that amplified the collapse. What it does show is the distribution of risk on the day the fund was most exposed.
Core: The On-Chain Evidence (In SEC Form)
Let’s treat this 13F as a blockchain state. The block number is June 30. The wallet addresses are the tickers. The balances are the weights. Here is the transaction log:
| Ticker | Weight | Sector | |--------|--------|--------| | SanDisk | 28.0% | Storage (NAND) | | Micron | 27.5% | Storage (DRAM/HBM) | | Bloom Energy | 9.4% | Fuel Cells (Power) | | TSMC ADR | 6.2% | Foundry (Fab) | | Nebius | 5.4% | AI Cloud | | CoreWeave | 4.4% | GPU Cloud | | Core Scientific | 3.2% | Bitcoin Miner / AI Data Center | | Applied Digital | 2.8% | Bitcoin Miner / AI Data Center | | IREN | 2.4% | Bitcoin Miner / AI Data Center | | Riot Platforms | 2.1% | Bitcoin Miner | | CleanSpark | 1.9% | Bitcoin Miner |
Total concentration: CR2 = 55.5%, CR7 ≈ 84.3%. For context, a typical institutional fund’s top 10 holdings rarely exceed 20-30% of total assets. This fund was 3x-4x more concentrated than the norm.
This is not a diversified portfolio. It’s a single thesis wrapped in different tickers: the bottleneck theory of AI compute. The thesis argues that as AI scales, the key constraints will shift from GPU design to memory bandwidth (HBM, NAND), power generation (Bloom Energy), and physical data center capacity (CoreWeave, miners). The Bitcoin miners were included as a proxy for “energy + existing infrastructure” — a bet that their AI hosting pivot would succeed.
Truth is found in the hash, not the headline. The hash here is the concentration. The headline was “AI visionary bets on compute.” The hash is “55.5% in two cyclical semiconductor stocks.” That level of concentration is not a hedge; it’s a wager that the entire thesis will correlate perfectly. In an up market, it delivers alpha. In a down market, it delivers a margin call.
The July Liquidation: A Data Reconstruction
Using the 13F as a base, I queried public market data for the July period. The sell-off in AI stocks was broad, but the most severe hits were in mid-cap names with lower liquidity. The Bitcoin miners—Core Scientific, Applied Digital, IREN, Riot, CleanSpark—saw their stocks drop 30-50% from June 30 to July 31. The storage stocks, SanDisk and Micron, fell roughly 15-20%.
Here’s the math: If the fund was levered 2x (a conservative estimate for a concentrated long-only fund), a 20% decline in the top two holdings would wipe out 22% of the equity. But the miners, which account for roughly 15% of the portfolio, dropped 40% on average. That alone would be a 6% equity hit. Add the decline in the other names, and the fund’s equity was likely down 35-45% by mid-July. That’s a margin call territory.
Based on my audit experience during the 2020 DeFi liquidity crisis, I know that the first thing to go in a leveraged unwind is the most liquid names—often the large caps. But the 13F tells us the fund had a lot of illiquid miners. When the prime broker (Citadel, in this case) demands more collateral, you can’t sell 5% of a miner’s float without causing a waterfall. The “problematic stock portfolio” that Citadel took over was almost certainly the miner basket.
Contrarian: The Thesis Wasn’t Wrong, But the Execution Was
Every analyst I’ve read has called this a “concentration disaster.” I disagree. The disaster was not the concentration; it was the leverage and the illiquid tail. The thesis—that AI’s next bottleneck is memory and power—is actually well-supported. HBM is supply-constrained. Power for data centers is a multi-year permitting issue. The logic is sound.
What failed was the portfolio construction. A fund that is 100% invested in a correlated thesis and uses leverage to amplify returns is a fund that will die in a market that moves against the thesis for even a few weeks. The July sell-off was a liquidity event, not a fundamental breakdown. AI capital expenditure plans were not canceled. GPU orders were not slashed. The market simply repriced risk.
Correlation ≠ causation. The fund collapsed because of the way it was structured, not because of the fundamental thesis. The mining stocks, in particular, introduced a hidden correlation: they are not just AI plays, they are also Bitcoin plays. In July, Bitcoin dropped 10% alongside the AI sell-off. That double whammy hit the miners harder than the pure-play AI names.
Takeaway: The Next Signal to Watch
Silence is just data waiting for the right query. The next 13F filing from Citadel’s own institutional arm will tell us how they unwound the remains. If we see a step-change reduction in the miner positions, it means Citadel is dumping them into the market. That will be a second leg down for those stocks.
But the bigger lesson is for the entire “AI infrastructure” trade. The Situational Awareness Fund was a microcosm of a larger risk: crowded positioning in a small set of names. The next time the market wobbles, look for the funds that are holding similar concentration. The 13F data is public. The queries are waiting. The truth is in the hash.