The 13F filing of Leopold Aschenbrenner's Situational Awareness fund, submitted on August 14, 2026, is less a portfolio disclosure and more a post-mortem of a failed macro thesis. The fund, which peaked at $20.2 billion in assets under management, collapsed in July under the weight of leverage and concentrated bets on AI infrastructure. But the story is not just about a hedge fund blowing up—it's about the structural fragility of the 'AI bottleneck' trade that has become the dominant narrative in both crypto and traditional markets.
Context: The Architect and the Collapse
Leopold Aschenbrenner, former OpenAI researcher and author of the influential essay Situational Awareness, launched his namesake fund in 2025 with a clear thesis: the bottleneck to AGI is not algorithmic innovation but physical infrastructure—compute, storage, power, and the industrial capacity to produce them. His 13F, filed with the SEC on August 14 (covering holdings as of June 30), confirms this worldview with surgical precision. The portfolio is a concentrated stack of seven core positions that represent 84.3% of the disclosed equity, with two storage giants—SanDisk and Micron—accounting for over half. Yet the fund's spectacular unraveling in July, when forced liquidation by prime broker Citadel occurred amid a broader AI stock sell-off, raises a critical question: Was the thesis itself flawed, or was its execution reckless?
From my perspective as a macro-focused researcher, the 13F is a document of both conviction and hubris. The fund's portfolio is a perfect expression of the 'infrastructure primacy' view I hold myself—but it also exposes the gap between having the right macro view and managing the micro risks of leveraged positions. The collapse was not a referendum on AI infrastructure; it was a stress test of portfolio construction in a world where liquidity can vanish overnight.
Core: The Portfolio as a Bottleneck Index
Let's dissect the positions. The 13F reveals a total of $20.24 billion in disclosed equity holdings as of June 30, but this is a snapshot taken before the July carnage. The breakdown reads like a list of critical nodes in the AI supply chain:
- Storage Layer (55.5%): SanDisk ($5.674B, 28.0%) and Micron ($5.574B, 27.5%). These are the fund's largest bets, reflecting the view that high-bandwidth memory (HBM) and NAND flash are the most binding constraints in AI training and inference. Micron is one of only three HBM3E suppliers alongside SK Hynix and Samsung. SanDisk provides enterprise SSDs for data centers. The concentration here is extreme—over half the portfolio in two cyclical semiconductor stocks.
- Compute Layer (16.0%): TSMC ADR ($1.256B, 6.2%) as the sole foundry play, along with two AI cloud providers: CoreWeave ($1.0B, 4.9%) and Nebius ($0.98B, 4.8%). CoreWeave is a GPU-as-a-service operator that has signed multi-billion dollar contracts with OpenAI; Nebius is a European AI cloud native. This layer captures the 'compute-as-a-service' model that is eating enterprise AI workloads.
- Power Layer (9.4%): Bloom Energy ($1.9B, 9.4%). A fuel cell manufacturer that provides onsite power generation for data centers. The fund's thesis: AI data centers need power that is reliable, scalable, and increasingly independent of grid constraints. Bloom's solid oxide fuel cells offer a natural gas-to-electricity pathway with 60%+ efficiency.
- Bitcoin Miner Layer (~7%): Core Scientific ($0.8B, 4.0%), Applied Digital ($0.3B, 1.5%), IREN ($0.2B, 1.0%), Riot Platforms ($0.1B, 0.5%), CleanSpark ($0.1B, 0.5%). These are no longer pure Bitcoin miners; they are pivoting to AI data center hosting. Core Scientific, for instance, has signed lease agreements with CoreWeave for GPU clusters. The fund bought these as 'energy-optioned real estate'—assets that own power capacity, cooling infrastructure, and data center sites.
What the portfolio is saying: The fund sees AI infrastructure as a vertical stack where the binding constraints are moving up the chain. First GPUs were scarce, now HBM is scarce, next power will be scarce. The inclusion of miners is the most innovative—and dangerous—part of the thesis. Miners are essentially derated power assets with existing data center shells. They can be repurposed for AI workloads faster than building new facilities. But they also carry Bitcoin price exposure, mining economics, and regulatory risk (e.g., the 2024 IRS rule on digital asset mining tax).
The leverage structure: The 13F does not disclose leverage, but the July collapse confirms it existed. Based on my experience modeling liquidity cascades in DeFi, the fund's decision to run a concentrated portfolio with leverage is analogous to a DeFi protocol with a single collateral asset and a high loan-to-value ratio. When the AI stock sell-off hit in July (triggered by a surprise Fed hawkish stance and a downgrade of AI chip orders by a major cloud provider), the portfolio's value dropped below margin thresholds. Citadel, acting as prime broker, likely demanded additional collateral or initiated a forced liquidation. The 'problem portfolio' that Citadel took over—as reported by Bloomberg—is consistent with a structured unwind of total return swaps or margin loans.
Contrarian: The Hidden Flaw in the Bottleneck Thesis
Here is where the macro consensus diverges from the fund's reality. The fundamental bet that AI infrastructure bottlenecks are long-duration and persistent is correct—I have argued this myself in my research on CBDC settlement layers for compute markets. But the fund's execution reveals a critical blind spot: the asymmetry between the thesis and the portfolio's risk profile.
First, the storage bet is pro-cyclical. Micron and SanDisk are semiconductor companies whose earnings are highly correlated with the global memory cycle. HBM demand is real, but supply is ramping fast. SK Hynix and Samsung are investing billions in new HBM capacity. Micron's own HBM3E production is expected to triple by 2027. The fund's thesis assumes that demand will outpace supply for years, but memory has historically been a boom-bust industry. A cyclical downturn would hit the fund's largest position hardest—and the lack of any hedging (e.g., shorting semiconductor ETFs or buying put options) is a glaring omission.
Second, the miner layer is a double exposure. The fund treats miners as AI infrastructure, but they are also Bitcoin miners. A drop in Bitcoin price—which occurred in late June as the market digested the Fed's hawkish stance—would depress miner revenues and force them to sell Bitcoin, further depressing the price. This creates a negative feedback loop that amplifies the AI sell-off. The fund's portfolio does not account for this correlation. It's a classic case of 'this time is different' thinking, where the AI narrative is assumed to decouple miners from Bitcoin's volatility. In reality, the decoupling is incomplete; miners still hold significant Bitcoin on their balance sheets and their stock prices are still correlated with BTC.
Third, the absence of downstream AI application stocks. The fund has no exposure to companies like Anthropic, OpenAI (not publicly traded), or even large-cap AI software plays like Microsoft or Salesforce. This means the fund is betting on the 'picks and shovels' of AI without any hedge if the gold rush slows. If AI model companies cut back on CapEx—as some did in July after realizing that scaling laws might be hitting diminishing returns—the entire infrastructure stack collapses. The fund's portfolio is a perfect storm of correlated tails: storage, compute, power, and miners all move in the same direction under the AI CapEx impulse.
Fourth, the leverage is the real killer. The 13F shows a portfolio that is structurally vulnerable to a margin call. The concentration in a few names means that a 20% drawdown in the top holdings (which happened in July) would wipe out most of the equity cushion. The fund's leverage, estimated at 2-3x based on the forced liquidation, turned a 20% market decline into a 60%+ loss. This is not a failure of the AI thesis; it's a failure of risk management. Yields dissolve; infrastructure remains, but only for those who survive the cycle.
Takeaway: What the Crypto Market Should Learn
From speculative frenzy to institutional ledger, the Situational Awareness fund's collapse is a cautionary tale for the crypto-AI crossover. The convergence of AI infrastructure and blockchain—whether through tokenized compute, decentralized GPU networks, or Bitcoin miners pivoting to AI—is real and will drive the next bull cycle. But the fund's fate shows that concentrated bets, even on a sound thesis, cannot survive without adequate risk controls.
Volatility is merely the tax on uncertainty. The fund paid the full premium because it mistook conviction for diversification. The 13F filing, delayed by 45 days, is a historical artifact of a thesis that was correct in direction but wrong in execution. As I analyze the macro liquidity environment for the remainder of 2026, I see the AI infrastructure trade as still valid—but the market will demand lower leverage, broader hedging, and a more nuanced understanding of the bottleneck dynamics.
For crypto-native investors, the lesson is clear: the next wave of institutional capital entering tokenized AI infrastructure will demand transparency not just in code, but in portfolio construction. The state does not compete; it absorbs. But the market also absorbs excesses. The Situational Awareness fund was a brilliant hypothesis that failed the stress test of reality. Let it be a footnote in the march toward a truly integrated AI-crypto infrastructure.