Leopold Ashenbrenner: What does the collapse of the $45 billion AI fund in a few days mean for crypt

At the end of July 2026, one of the most watched funds in the global market evaporated about three-quarters of its assets in just a few days.

The artificial intelligence fund "Situational Awareness" founded by former OpenAI researcher Leopold Ashenbrenner was forced to sell its entire public stock portfolio at a discount to Ken Griffin's Citadel after receiving an unsatisfied margin call from its main broker.

This story is of great significance to cryptocurrency readers, for reasons that go far beyond schadenfreude over the explosion of leverage in another asset class. The mechanism of this crash will be instantly recognized by anyone who has experienced trading in 2022; it involves a person whose career began in the FTX circle; and it is happening at the same time that Bitcoin is quietly breaking the correlation with artificial intelligence transactions. Today, multiple cryptocurrency native companies are taking on artificial intelligence infrastructure risks directly on their balance sheets.

Who is Leopold Ashenbrenner?

Ashenbrenner is a German investor and former artificial intelligence researcher. He was born in 2001 or 2002 to both doctors and attended the John F. Kennedy School in Berlin. He entered the University of Colombia at the age of 15 and graduated in 2021 at the age of 19 as a valedictorian, with a degree in economics and mathematics and statistics.

His early career included serving at the FTX Future Fund (the charity arm of the Sam Bankman Fried Exchange), where he helped run charity projects in Bahamas. He later joined OpenAI's "Super Alignment" team, responsible for controlling systems with capabilities beyond humans. OpenAI fired him in April 2024 on suspicion of information leakage. Ashenbrenner disagreed with this statement. He said he had shared a largely unclassified planning document with outside researchers for comment, and that his firing was the result of tensions after he warned about the company's security practices. OpenAI said these concerns had nothing to do with his departure.

In June 2024, he published a 165-page article "Situation Awareness: The Next Decade", believing that the arrival of general artificial intelligence far exceeds the understanding of most people, and the resulting computing, energy and hardware needs will be historic. The article became must-read material in Silicon Valley. The next month, he converted it into a fund of the same name.

The transaction itself is the article. If artificial intelligence capabilities continue to expand, semiconductors, memory, data centers and power infrastructure are bottlenecks, and leveraging leverage to have this bottleneck is the most confident expression in this argument. Backers include Stripe co-founders Patrick and John Collison, former GitHub CEO Nat Friedman and investor Daniel Gross. Jane Street is also an investor. The Wall Street Journal reported that its revenue since its establishment exceeded 1000%.

Reported peak assets vary by source. CNBC put the fund's all-time high at about $45 billion, while other reports cited peak assets under management of about $20 billion. Either number is an extremely large amount of capital for a manager who had never managed funds before founding the fund at the age of 22.

What happened to Situational Awareness?

The liquidation process lasted approximately two weeks in late July. The fund's concentrated AI infrastructure stocks, which reportedly include SK Hynix, CoreWeave, Nebius, Micron and Bloom Energy, fell 35% to 47% for the month. The Philadelphia Semiconductor Index fell 28.6% from its June 22 peak as investors began to question whether ultra-large capital spending could generate sufficient returns. It is said that short positions in software stocks were also detrimental to the fund, exacerbating losses in two ways. Then, leverage played its due role. Reports say the fund has a leverage ratio of as high as 400%. At four-fold leverage, a 25% drop in the underlying position is mathematically enough to erase an investor's entire equity contribution. The actual position fell by far more than 25%.

Prime brokers Goldman Sachs, JPMorgan Chase and Bank of America issued margin calls. The fund has tried a variety of ways to get out of its way: issuing capital raising letters to existing investors, discussing with lenders, and negotiating with Millennium Management and Jane Street Group. According to the Financial Times, all these attempts failed. Citadel stepped in and acquired the entire public portfolio at a discount. Assets fell from approximately $45 billion to approximately $10 billion. Reports since then suggest that the fund may still be forced to liquidate its holdings further.

There is a meaningful postscript. Once Citadel absorbed positions, the Nasdaq rose 3.30%, and the semiconductor index rebounded sharply. Most of the decline in AI infrastructure stocks in late July came as the market was pricing a large and obvious forced seller. Remove him and remove the discount. The timing was unusual in another way: Ashenbrenner married Avitar Balvert, chief of staff to Anthropic CEO Dario Amodi, in California, a weekend that coincided with the fund's liquidation.

Why is AI fund exposure important to the cryptocurrency market?

There are three reasons, in order of increasing importance.

First of all, it's a familiar storyline, but with a different code. A young, mathematically gifted manager constructs an all-encompassing argument about the future, executing through extremely concentrated positions and high leverage, producing surprising returns that attract huge amounts of capital, and then discovering that the leverage is symmetrical. The cryptocurrency market has repeatedly conducted such experiments. The specific detail that closes the cycle is that Ashenbrenner's first important job was at the FTX Future Fund, and Jane Street, where Bankman-Fried himself trained, appears in this story as both an investor and a failed rescue counterparty. The comparison should not be excessive. There are no allegations of fraud, no customer funds, no assets lost. Situation Awareness appears to be a legitimate fund that took directional bets and suffered losses, which is a very different thing from what happened to FTX. But the underlying behavioral pattern-narrative beliefs plus leverage minus risk management-is exactly the one that costs cryptocurrency investors more money than any hacker attack.

Secondly, its operating mechanism is exactly the same as the clearing cascade. Concentrated leveraged bulls, falling prices, margin calls, sellers forced to sell in falling markets, and well-capitalized buyers waiting to take orders at discounts. Cryptocurrency traders continue to see this on the chain and in exchange clearing information. On July 13, when the South Korean Composite Index fell 8.95% and SK Hynix fell a record 15.37%, leveraged cryptocurrency positions worth US$253 million were simultaneously forced to liquidate, of which long positions accounted for 76% of the total. Same physical principles, different places.

The third and most important thing is that cryptocurrency didn't do much this time. Why didn't Bitcoin follow AI stocks down this time? For most of 2026, cryptocurrencies will be traded as a high-beta expression of AI trading. It rises as chip stocks rise and falls as chip stocks fall. The relationship was broken in July and twice in five trading days. When about $797 billion evaporated from the largest U.S. technology stocks on a Thursday in late July, Bitcoin was barely moving. On July 29, Bitcoin rose about 1% to $63,800, as Asian stock markets experienced one of the worst two-day losses of the year and SK Hynix's share price fell by nearly one-fifth despite a more than sixfold increase in quarterly profits. Ethereum rose 1% to $1,899, XRP rose 2% to $1.07, and Solana held steady around $73. While Citadel absorbed Situation Awareness's portfolio and AI infrastructure stocks rebounded sharply, the cryptocurrency market was largely unaffected in the other direction. Throughout July, Ethereum rose 16.29%, Bitcoin rose 5.61%, and the AI infrastructure sector is being re-priced lower.

This interpretation is crucial. One interpretation is that Bitcoin is regaining its independence as an asset class and is currently driven by interest rate expectations, ETF funding flows and its own regulatory agenda rather than sentiment about Nvidia's supply chain. Analysts are increasingly describing cryptocurrencies as behaving like a liquidity sponge, expanding and contracting with the global money supply and real interest rates, rather than following the narrative of any individual stock. Research shows that about 45% of Bitcoin's weekly price movements in 2026 will be driven only by ETF funds flows. A more cautious interpretation is that two weeks does not constitute a trend, and since 2020, the idea of decoupling has been repeatedly raised and then abandoned. The honest stance is that correlations have diminished significantly and significantly, and the next real risk avoidance event will test whether this weakening is structural or coincidental.

Which cryptocurrency companies are actually exposed to AI transactions?

Here, the story is no longer an analogy, but becomes a direct exposure. A large part of the Bitcoin mining industry has spent two years transforming itself into AI infrastructure, and now its price reflects this accordingly. Miners have two things that AI companies need most: massive contracted power capacity and physical data center space. After halving the mining economy in 2024, shifting this capacity to high-performance computing and AI hosting has become the industry's leading strategy. Leasing activity increased from 95 MW in the first quarter of 2026 to 1.19 GW in the second quarter, and a further 928 MW were announced in the third quarter as of July 27, bringing the year-to-date total to 2.21 GW. TeraWulf signed a $19 billion lease with Anthropic. Hut8, IREN and Applied Digital accounted for the majority of this year's contracted capacity.

This steering is two-way. When AI infrastructure sentiment collapsed in July, these stocks fell more than the underlying assets they named. IREN fell 33% in a month, TeraWulf fell 38%, Applied Digital fell 36%, and the broader Global X Data Center and Digital Infrastructure ETF fell 13%. In July alone, MARA Holdings fell 18.14%, IREN fell 19.40%, Riot Platforms fell 23.08%, while spot Bitcoin rose. Their beta coefficients explain this sensitivity: IREN has a beta coefficient of 4.28 in May, TeraWulf is 4.26, and Applied Digital is 5.68.

KBW analysts made the sharpest observations of what was actually being re-priced. They believe the sell-off mainly removes the value the market allocates to future AI and high-performance computing leases, rather than repricing completed projects. In other words, the market stopped paying for project pipelines and started paying only for actual contracts signed with creditworthy tenants. KBW downgraded Core Scientific to "par with the broader market" and pointed to a new danger category it calls "model level risk": developers holding leases will be at risk if AI lab tenants fail to meet expectations. CoreWeave, one of the core positions Ashenbrenner reported, shows the entire cycle. It started as an Ethereum mining business and later became an AI cloud provider. It tried to merge with Bitcoin miner Core Scientific but failed. Since then, its share price has fallen 61% from a mid-year high of $187, and its market value has evaporated by about $33 billion in six weeks. At the same time, it faces criticism from short sellers and questions about its GAAP profitability. A company born out of cryptocurrency mining has become the most crowded position in AI trading, and then one of its biggest victims.