The block confirms what the eyes missed. On August 14, 2025, Leopold Aschenbrenner—former OpenAI researcher, author of the influential Situational Awareness thesis, and manager of what was rumored to be a $450 million fund—did something that defies the retail narrative. He liquidated his entire position in Nvidia, Amazon, and other AI infrastructure stocks. The proceeds went into a single private equity stake: Anthropic.
This is not a hedge. It is a conviction trade. And it carries a signal that the market has not yet priced in.
Let me start with the facts. Aschenbrenner’s fund, which I will call the Situational Awareness Fund (SAF) for clarity, had been publicly long AI infrastructure since early 2024. In his writings, he argued that the path to AGI requires a $1 trillion compute cluster. Bulls assumed he would hold Nvidia forever. Instead, he sold every public share and concentrated his capital into a private company that does not even have a public path to liquidity. The reported size of the SAF is $450 million, though I flag that number with skepticism based on my own experience auditing crypto fund flows. The actual figure is likely lower, but the direction is unambiguous.
Context: The Man and His Thesis
Aschenbrenner is not a typical fund manager. He is a former OpenAI safety researcher who left to write a 165-page essay arguing that AGI is imminent—by 2027 to 2030—and that the world is unprepared. His followers call him the “Cassandra of AI.” In the crypto world, we would call him a “maximalist” with a specific religion: scaling laws will continue to hold, compute will become the most valuable commodity, and the first mover to AGI will capture an unprecedented share of global GDP.
Given that thesis, why sell the compute providers? Why abandon Nvidia, the literal monopoly on the chips that power the scaling laws? The answer lies in the second layer of his belief system: he believes that the model itself—the algorithmic architecture, the alignment technique, the proprietary data flywheel—will capture more value than the hardware. He is betting that Anthropic’s constitutional AI and responsible scaling policy give it a moat that Nvidia’s GPUs cannot replicate.
This is a bet on vertical integration. Nvidia sells picks and shovels. Anthropic builds the mine. And Aschenbrenner is wagering that the mine owner will extract more gold than the shovel seller, especially as the mine becomes more valuable over time.
Core: The Order Flow Analysis
Now, let me apply the forensic lens I use for on-chain trading patterns. I see three distinct signals in this trade.
First, the timing. Aschenbrenner sold Nvidia after its Q2 2025 earnings beat, when the stock was near all-time highs. This is a classic “sell the news” pattern, but from an insider who understands the AI supply chain better than most. He is not reacting to a quarterly number; he is reacting to a structural shift in the compute landscape. He knows that Nvidia’s H200 and B200 products are already facing competition from custom ASICs—Google’s TPU v6, AWS’s Trainium3, and even AMD’s MI400. The monopoly is eroding, and the market is pricing Nvidia as if it will last forever.
Second, the concentration. By moving from a diversified basket of AI infrastructure stocks to a single private company, Aschenbrenner is increasing his risk dramatically. But he is also revealing his conviction that the public market is mispricing the tail risk of AGI. In his view, if AGI arrives by 2028, Anthropic’s equity will be worth multiples of its current valuation, while Nvidia’s will be capped by competition and commoditization. If AGI is delayed, both will suffer, but the private equity will be more volatile. He is not hedging; he is doubling down on a specific outcome.
Third, the liquidity discount. Private equity in a pre-IPO AI company is illiquid. Aschenbrenner is accepting that he cannot exit for years. This is a signal that he believes the timeline for AGI is shorter than the typical VC lockup period. He is betting that Anthropic will either IPO or be acquired before 2028, at a valuation that justifies the risk. Based on my experience analyzing token unlock schedules and liquidity crunches in DeFi, this is a bet on compression of time. The market is not pricing that compression.
Contrarian: The Retail Blind Spot
Retail investors are piling into Nvidia. The narrative is simple: AI is the next industrial revolution, Nvidia is the only game in town, buy the dip. The contrarian angle is that the “picks and shovels” thesis is already priced in, and the smart money is rotating into the actual value creators.
Consider the parallel to crypto. In 2020, retail bought $ETH because it was the “gas” for DeFi. But the real value accrued to the protocols that built moats—Uniswap, Aave, Compound. The infrastructure narrative (gas) was a lagging indicator. The real alpha was in the application layer. Aschenbrenner is making the same trade: sell the gas, buy the app.
But there is a deeper blind spot. Retail investors do not understand the fragility of Nvidia’s monopoly. The company’s competitive advantage is its CUDA ecosystem and the speed of its hardware iteration. But the hyperscalers—Google, Amazon, Microsoft—are all designing their own chips to reduce dependence. The inference market is already fragmenting, and training is next. Aschenbrenner’s move suggests that he sees the peak of Nvidia’s dominance within 18 months.
Moreover, the regulatory environment is shifting. The Tornado Cash sanctions set a precedent that code can be illegal. In the AI space, similar regulation could target open-source models or mandate safety standards that favor incumbents like Anthropic. Aschenbrenner, who has written extensively on AI safety, likely sees regulation as a moat for Anthropic, not a headwind. That is a contrarian view that most retail investors miss.
Takeaway: Actionable Price Levels
Trace the anomaly, ignore the noise. The market has not yet absorbed the signal from Aschenbrenner’s trade. Here are the actionable levels to watch:
- Nvidia (NVDA): If the stock breaks below $120, the move will accelerate. The $100 level is the final support. If Aschenbrenner is right, NVDA will revisit $80 within 12 months.
- Anthropic secondary market: Look for the next round of secondary trading on platforms like Forge or EquityZen. If the valuation reaches $100 billion, that is a confirmation of the thesis. A drop below $60 billion is a rejection.
Front-run the narrative, not just the chain. The narrative is shifting from “compute is king” to “alignment is the moat.” The smart money is already positioning. The question is whether you have the patience to hold through the volatility.
Silence is the safest ledger. The block confirms what the eyes missed. The eyes saw a fund manager selling Nvidia. The block saw a vote of no confidence in the infrastructure narrative.