The post-2000s AI stock guru has liquidated all positions, suffering a staggering loss of 140 billion yuan
On July 24, Leopold Aschenbrenner, the post-2000s AI stock guru, wrote in a letter to investors that if you had been waiting for the right moment to increase your positions, now might be the time.
Six days later, he liquidated all his holdings. Under duress.
It is reported that Leopold Aschenbrenner manages a fund named Situational Awareness, and sold off all its US stock positions at one go, which accounts for about two-thirds of the fund's total position. The total loss of the fund is reportedly around 200 billion US dollars, equivalent to 1.4 trillion RMB. The buyer is another large hedge fund whose name has not been disclosed. Market rumors say the buyer is Citadel, while Millennium participated in the bidding.
It took him less than two years to grow from 200 million US dollars to 240 billion US dollars. The process from 240 billion to full liquidation was even faster, taking only six days.
Leopold is truly a remarkable figure. Born in 2001, he is a child prodigy who entered Columbia University at 15, graduated with the best grades at 19, and has an impressive resume. He is the most dazzling and profitable Silicon Valley star in the past two years, yet there is a sense of awkwardness lingering around him.
For example, he worked at SBF's FTX Future Fund for nine months, which is the crypto platform that later collapsed. The platform had a foundation under the banner of "Effective Altruism", and its founder SBF often said that "making money is for giving". After FTX collapsed, it was found that SBF embezzled tens of billions of dollars of client funds and was sentenced to 25 years in prison. That set of ethics was called "mostly a front" by SBF, nothing more than a facade.
Later, Leopold left FTX and joined the "Superalignment Team" at OpenAI. The task of this team was to ensure that AI would not get out of control, and to put the brakes on superintelligence. In April 2024, the 23-year-old Leopold was fired by OpenAI on the grounds of leaking confidential information, after he wrote a security memo warning that OpenAI's security was insufficient.
However, one month later, the Superalignment Team was also disbanded. Then Leopold turned around and launched the most aggressive AI acceleration fund in the whole market, betting all his chips on the rapid growth of AI. Well-known hedge fund Jane Street is among his investors.
The "Superalignment Team" was supposed to be the one stepping on the brakes, and no one knows whether Leopold was indignant or not, but after leaving, he directly became the one stepping on the accelerator, and the most radical one at that, frantically adding leverage.
Only two years after being fired by OpenAI at the age of 23, the 25-year-old Leopold became the "AI stock guru" who grew his capital from 200 million to more than 24 billion US dollars, and his AUM reportedly exceeded 400 billion US dollars at its peak. When he left OpenAI, he wrote a famous article predicting that AGI would arrive in 2027. As a result, AGI has not arrived yet, but the market has shifted from rewarding to punishing.
Moreover, this incident is strikingly similar to what happened to Bill Hwang, who was forced to liquidate after encountering a black swan event when he held heavy positions in Chinese concept stocks with high leverage. Now the market is highly concerned about the excessive investment in AI infrastructure and its ROI, and Leopold's positions plummeted from high levels, leading to his collapse due to leverage. The Chinese concept stock market has barely recovered since Bill Hwang's incident, and no one can yet falsify the current AI narrative. However, Leopold's forced liquidation is a landmark event, and it remains to be seen whether it will become a watershed for the current AI market rally.
From Braking to Acceleration
Leopold Aschenbrenner was born in 2001 into a family of doctors in Germany. He was admitted to Columbia University at 15, and graduated as the top student in his department at 19, with three degrees in mathematics, statistics and economics. This kind of start is dazzling in any era.
While studying at Columbia University, Leopold co-founded the on-campus Effective Altruism (EA) chapter, stepping into this ideological circle very early. From 2021 to 2022, he went to the University of Oxford's Future of Humanity Institute as a researcher. This institution is the core think tank of the EA movement, where he was able to systematically delve into EA theories.
EA was proposed by Oxford philosophers William MacAskill and Toby Ord, advocating to "maximize good" in a data-driven and utilitarian way, and encouraging the idea that "making money is for giving". The founder of FTX was influenced by MacAskill when he studied at MIT, and donated about half of his salary after joining Jane Street. The FTX charity department also promised to allocate more than 160 million US dollars to fund hundreds of non-profit organizations. Later, Leopold also joined the FTX Future Fund to engage in EA funding work.
Leopold worked for nine months and left on the eve of the FTX collapse in November 2022. Later, FTX was found to have embezzled tens of billions of dollars of client funds, and SBF was sentenced to 25 years in prison, with that set of ethics described by himself as "mostly a front", nothing more than a facade. Leopold was not a bystander, he was right behind that facade. This experience of being at the center of a financial scam was described by multiple sources as something that made him "perceive the complexity and risks of the market at an early stage".
In 2023, he moved to OpenAI and joined the Superalignment Team.
This team is also very interesting. At that time, due to the huge public opinion impact caused by ChatGPT, under internal and external pressure, it was established in the summer of 2023 under the leadership of OpenAI's core figures Ilya Sutskever and Jan Leike. Its goal is to solve an almost unsolvable problem: "When AI's intelligence surpasses human beings, how to ensure it does not get out of control?"
OpenAI promised to invest 20% of its computing power, set a four-year deadline, and aimed to develop technologies to align superintelligence by 2027. The core challenges include scalable supervision, goal generalization, internal representation, and adversarial robustness. However, the problem is that traditional alignment methods such as RLHF (Reinforcement Learning from Human Feedback) will no longer make humans reliable supervisors once the reasoning ability and scale of AI exceed the scope of human capacity.
Then came the famous OpenAI "coup" at the end of 2023. In short, Sam Altman won, and the accelerationists won. Looking back now, this superalignment plan is a total joke.
In April 2024, a sudden change occurred. Leopold submitted a security memo to the board of directors, warning that the company's security measures were "extremely insufficient", but OpenAI decided to fire him on the grounds of "leaking internal information". A month later, Jan Leike left the company, and later Ilya left to found SSI, and the Superalignment Team was disbanded.
All the people who wanted to step on the brakes for AI were kicked out of the driver's seat.
After being fired, in June 2024, he published a 165-page long article titled "Situational Awareness: The Decade Ahead", predicting that AGI would arrive around 2027, and putting forward judgments such as "intelligence explosion" and "superintelligence". This article went viral in Silicon Valley's tech circle, and paved the way for his subsequent fundraising. A person who had participated in superalignment research and published a long security paper turned around and went to Wall Street to do AI acceleration.
If you can't beat them, join them. It seems this is the case all over the world, and the smartest minds are switching sides faster than ever.
Moreover, according to Fortune magazine, Leopold is engaged to Avital Balwit, chief of staff of Anthropic CEO Dario Amodei. It is unknown whether this relationship is related to the change in his views on AI, but the result is clear: Anthropic is exactly the largest single position in his fund.
At the end of 2024, Leopold officially established the Situational Awareness fund. A top student from Columbia University, EA believer, FTX affiliate, superalignment researcher, finally became an AI acceleration fund manager. Leopold's resume has gone full circle around AI, and the next step is to raise capital.
How the 240 Billion Fortune Was Built
Situational Awareness is a hedge fund that bets almost 100% on the AI theme. The team is extremely lean, with only 4 investment professionals and a total of 8 employees. Its official website has only one sentence: It is an AGI-focused investment advisor founded by Leopold Aschenbrenner.
Looking at the list of early investors of this fund, it is almost a who's who of Silicon Valley power players. Patrick and John Collison, co-founders of Stripe, Nat Friedman, former CEO of GitHub, tech investor Daniel Gross, plus quant giant Jane Street.
These people are no ordinary LPs. They are the core nodes of the EA movement, the Effective Altruism network, and the tech accelerationist circle. In other words, Leopold's fund was endorsed by this circle from day one. The Jane Street connection is particularly convoluted: it was SBF's old employer, and bought shares of Anthropic from the bankrupt assets of FTX, while Anthropic is exactly the largest single position in Leopold's fund, accounting for about one-fifth of its total assets.
The fund's strategy is also very straightforward. It makes concentrated bets on AI infrastructure, including computing power, electricity, data centers and storage, and uses options and short selling to hedge or amplify returns. The fund's prime brokers are Bank of America, Goldman Sachs and JPMorgan Chase. The 13F filing for the first quarter of 2026 shows that the fund's long positions are all hard assets in AI infrastructure:
It holds approximately 879 million US dollars in Bloom Energy, 724 million US dollars in SanDisk, 556 million US dollars in CoreWeave, 12.41 million shares of Dutch AI infrastructure provider Nebius worth approximately 2.6 billion US dollars. It also led the 1.6 billion US dollar financing of Sharon AI, taking a 19.9% stake, and became a cornerstone investor of SK Hynix's Nasdaq ADR.
The positions also include a number of Bitcoin mining enterprises and data center targets, including Riot Platforms, CleanSpark, Core Scientific, IREN, and Applied Digital. In addition, the early-stage investment in Anthropic accounts for about one-fifth of the fund's assets, with a valuation of about 60 billion US dollars when it entered in February 2025. Calculated based on the current valuation of one trillion US dollars, the floating profit is about 18 times.
As AI advanced triumphantly, the performance of the Situational Awareness fund once became a myth.
By the end of June 2026, the fund's net return for the year was about 439%, and its cumulative return since inception exceeded 1000%. Its scale expanded from 200-300 million US dollars at the time of establishment to about 240 billion US dollars by mid-2026. A young man in his early 20s turned a small sum of money into a behemoth that caught Wall Street's attention in just two years.
The key is that Leopold's judgments were indeed accurate.
Back when the market was frantically chasing the narrative of "power shortage and computing power shortage", he said that software can be iterated rapidly with AI, but physical world assets such as factories, energy and infrastructure cannot be created by AI, and these are the most valuable assets in the future. This is the confidence behind his heavy positions in the AI infrastructure chain.
Research institution Citrini once did the math: if you invested 100 million US dollars when the fund was established, even if you lost 90% in July, this investment would still be worth 230 million US dollars. Scale itself is part of the narrative.
On July 24, 2026, in the face of sharp market volatility, he wrote in his letter to investors that the fund "did not escape" the recent market turbulence, but he said that this sell-off of tech stocks was "the most attractive investment opportunity since the beginning of 2025". At the end of the letter, he added a line: If you have been waiting for the right moment to increase your positions, now might be the time.
However, Leopold did not make it to the point of adding positions. Instead, six days later, he liquidated all his holdings.
The Reflexivity of the Market
Soros has a well-known financial theory called reflexivity. It roughly means that what determines stock prices is not only fundamentals, but also the expectations and trading behaviors of participants, which in turn change fundamentals and prices, forming a two-way cycle and self-reinforcement.
Applied to the AI sector, this was a typical upward reflexivity cycle before.
The long-term narrative of AGI brought the market expectation of infinite expansion of computing power demand, and a large amount of capital poured into the AI hardware sector, pushing stock prices to continue to rise, and the fund's net value soared accordingly, constantly confirming Leopold's judgment, prompting him to further increase leverage and concentrate positions. Leopold himself firmly believed in the narrative that AGI will eventually arrive, and used huge amounts of capital to trade, relying on capital to boost the price trend corresponding to this narrative.
But the problem is that the most fatal feature of the reflexivity cycle is that the more extreme the rise, the more severe the fall. The long-term industrial vision can be infinitely romantic, but the leverage and valuation realization in the capital market will not wait for anyone. The higher the stock price, the lower the tolerance of speculative capital and the fault tolerance rate of leveraged capital.
Looking at Leopold's 13F positions, almost every heavyweight holding is bleeding.
SanDisk fell about 46% for the whole month, making it the biggest losing position. Bloom Energy fell about 33%, CoreWeave fell about 25%, Nebius once fell more than 46% with a monthly pullback of about 43%, SK Hynix's ADR pulled back about 35% from its highest point, and Oracle and AMD each fell about 20%. The tech-heavy Nasdaq 100 index posted a cumulative decline of about 10% for the month, and the Korea Composite Stock Price Index fell by about one-third.
One of the triggers was Meta's plan to rent out its AI computing power to external parties, which made the market start to debate whether computing power has become oversupplied. The credit market is also tightening: CoreWeave's CDS rate once soared to about 855 basis points, and the five-year default probability calculated by the market model is as high as 50%. Oracle's CDS has also risen significantly.
The real culprit is still leverage.
Media reports said that Leopold used up to about 4x leverage, which is of course far less than Bill Hwang's leverage, but the rapid drop in stock prices still triggered margin calls, and he could not meet the margin calls so forced liquidation was the only option.
What is even more ironic is Leopold's so-called hedging. At the end of the first quarter, the top five positions of the fund were all put options, covering Nvidia, semiconductor ETFs, Oracle, Broadcom and AMD, with a nominal value of tens of billions of US dollars. Among them, the put options of VanEck semiconductor ETF alone were worth about 2 billion US dollars, and Nvidia puts were about 1.6 billion US dollars.
But 13F is only a snapshot at the end of the quarter, which cannot show the strike price and expiration date, and no one knows how much role these hedges played in the July decline. What can be confirmed is that his hedges did not work, or the effect was not enough to withstand the extreme pullback, leading to a liquidity crisis. Otherwise, Leopold could have adjusted his positions to save himself.
Facing the pressure of margin calls, Leopold raised emergency funds from LPs and lenders while negotiating to transfer private equity assets such as Anthropic to raise cash. But all rescue efforts failed. On July 30, Leopold chose to liquidate all his US stock positions and sell the entire public market portfolio to another large hedge fund.
After the public positions were cleared, the fund was left with only unlisted private equity assets, taking a form closer to a concentrated holding growth equity vehicle, and it was no longer the flag-bearer of the all-in AI hedge fund it used to be.
However, it is worth noting that right after Leopold's liquidation, the US semiconductor and storage sectors rebounded sharply, with Western Digital rising more than 20%, and Micron and SK Hynix rising more than 15%. After Leopold's selling pressure was released all at once, buy orders came in to take the heavily discounted shares.
This also shows that the current market's disagreements and games are far from over, and both bulls and bears are betting with real money. At the bottom of this game is a major valuation debate by global capital over the AI narrative.
In the past two years, the narrative was simple: the one who burns more money and expands