The filing hit the SEC EDGAR system at 4:17 PM on August 14.
It was a 13F-HR, the routine quarterly snapshot of a fund's U.S. stock holdings. Nothing unusual on the surface. But the numbers inside told a story of a concentrated bet so large, so leveraged, that the entire structure collapsed under its own weight less than three weeks after the snapshot date.
This is the portfolio of Leopold Aschenbrenner's Situational Awareness Fund. A $202.4 billion (at June 30) pile of AI infrastructure plays, storage chips, and Bitcoin miners. And it's now a museum exhibit of what happens when high conviction meets high leverage in a bull market that suddenly turns.
Pulse on the chain, breath in the market.
Context: The Man and the Thesis
Leopold Aschenbrenner isn't your typical hedge fund manager. He was a researcher on OpenAI's superalignment team, left in 2024 over safety philosophy differences, and wrote a viral essay titled "Situational Awareness" about the geopolitical race for AI compute.
The thesis was simple: AI compute is the new oil, the new uranium, the new hard currency. Control the compute, control the future. And the bottlenecks aren't just GPUs—they're the entire supply chain: high-bandwidth memory (HBM) from Micron, NAND flash from SanDisk, advanced fabrication from TSMC, power generation from Bloom Energy, and cloud infrastructure from CoreWeave and Nebius.
But he added a twist: Bitcoin miners. Companies like Core Scientific, Applied Digital, IREN, Riot, and CleanSpark. His logic? They own power contracts, land, and existing data center infrastructure. They're the perfect vehicle to pivot from mining Bitcoin to hosting AI compute racks.
Running where the liquidity flows fastest. But also where it drains fastest.
Core: The Numbers That Tell the Story
The 13F reveals a portfolio of extreme concentration. The top two holdings—SanDisk at $56.74 billion (28.0%) and Micron at $55.74 billion (27.5%)—make up 55.5% of the entire fund. The top seven positions (adding Bloom Energy, TSMC ADR, Nebius, CoreWeave, and Core Scientific) account for roughly 84.3% of the portfolio.
For context, a typical institutional fund's top 10 concentration is usually 20-40%. This fund is 2-3 times that. It's not a portfolio; it's a single thesis expressed across 10-15 names.
The rest includes applied digital, IREN, Riot Platforms, and CleanSpark—all Bitcoin miners with AI pivot stories. Combined, these mining stocks probably represent around 15% of the portfolio. They are small-market-cap, high-volatility, low-liquidity names.
The fund's sector breakdown: - Storage (SanDisk + Micron): 55.5% - Power (Bloom Energy): 9.4% - Cloud GPU (CoreWeave + Nebius): 9.8% - Foundry (TSMC): 6.2% - Bitcoin Miners (Core Scientific, Applied Digital, IREN, Riot, CleanSpark): ~15% - Others: ~4%
This is a vertical integration of the AI compute stack—from raw materials (storage, power) to intermediate services (cloud) to the final physical infrastructure (miners). But missing: any AI application layer. No OpenAI, no Anthropic, no software names. The fund is betting on the "picks and shovels" of AI, not the miners themselves.
Caught in the flash, framed in fact.
The 13F is a snapshot as of June 30. By the end of July, the market had moved. AI stocks fell—not a crash, but a meaningful correction. The fund's leverage, which we can only estimate from the market reports, kicked in. Reports from July 25-30 describe the fund being forced to sell large portions of its public stock holdings due to "AI-related stock declines and margin pressure."
Then came the aftermath: Citadel, the giant market maker and prime broker, stepped in to take over the "problematic portfolio." The exact structure of the rescue is unclear—it could be a total return swap termination, a margin loan workout, or a custodial transfer. But the implication is clear: the fund's leverage was so high that a 15% drawdown in its core holdings triggered a liquidity crisis.
Seventy-two hours without sleep, zero doubts. But doubts don't matter when the margin call arrives.
Contrarian: The Unreported Angle
Everyone is focusing on the leverage and the concentration. But the real blind spot is the fund's assumption about the nature of the AI compute bottleneck.
The market narrative is that GPU supply is the constraint. But Leopold's thesis argued that storage (HBM, NAND) and power are the real bottlenecks. That's a plausible long-term view, but it's a short-term execution risk. Capacity expansions for storage and power take 18-24 months. If the market perceives that these bottlenecks are easing—or that AI CapEx is slowing—the entire thesis unwinds simultaneously across all positions.
Furthermore, the inclusion of Bitcoin miners as "AI infrastructure" is a creative narrative, but it's also a double-edged sword. These miners are not pure AI plays. They still have legacy Bitcoin mining operations, which are subject to Bitcoin price volatility, halving cycles, and energy price fluctuations. If the AI narrative cools, these stocks lose both their AI premium and their crypto floor. The fund's portfolio is effectively a massive leveraged bet on the continuation of the AI hype cycle.
Sensing the tremor before the earthquake hits. But the earthquake was the fund itself.
Takeaway: What to Watch Next
The 13F is a post-mortem. But the story isn't over. Citadel now holds a large portion of this portfolio. The next 13F filing (due in November for the quarter ending September 30) will reveal whether Citadel has liquidated, held, or restructured these positions. The market impact of a forced liquidation of these concentrated positions—especially the smaller miners—could still be ahead.
For the crypto market, the lesson is cautionary. The Bitcoin miners' AI pivot narrative has been a major driver of their stock prices. If this fund's collapse discourages other leveraged funds from making similar bets, the capital flow into these miners could slow. Conversely, if the underlying thesis (AI needs power and storage) survives this fund's failure, the miners with real AI contracts may emerge stronger.