Author: Long Yue, Wall Street View
A 24-year-old young man raised $45 billion on Wall Street based on a lengthy article predicting that AI would transform the world, earning a staggering 439% gain in the first half of this year. Yet, over the past month, as the AI sector plunged, he lost all of his two-year-long accumulated gains in one fell swoop.
On July 30, a rumor spread on Wall Street:
A hedge fund with only eight employees sold its entire public market position—approximately $16 billion in stocks—in a single trade to a counterparty over the past 36 hours.
The buyer is Citadel, one of the world's largest hedge funds, founded by Ken Griffin, with assets under management of $71 billion.
The fund that was forced to cut its losses is called Situational Awareness (SA), founded by Leopold Aschenbrenner.
Notably, Aschenbrenner’s fiancée, Avital Balwit, is the chief of staff to Anthropic CEO Dario Amodei. This AI crash and his forced liquidation occurred just before their wedding.
Just a few weeks ago, he was the most sought-after name on Wall Street and in Silicon Valley.
The "AI Prophet" championed by Silicon Valley
Leopold Aschenbrenner, German, 24 years old.
Entered Columbia University at age 15, graduated first in his class at age 19. After graduation, he joined OpenAI’s “Superalignment” team, specializing in research on aligning superintelligence with human values. In 2024, OpenAI fired him citing “improper disclosure of internal information.” He denies the claim, stating he merely shared a largely non-sensitive planning document with external researchers for feedback; the real trigger was an internal memo he wrote criticizing OpenAI’s safety measures.
After being fired, he did something many people didn’t expect—he wrote a 165-page essay titled “Situational Awareness.”
The core argument of the article is that the pace of AI development far exceeds public expectations: by 2027, AI will be capable of conducting AI research on its own, and humanity will face true superintelligence. Achieving this will require an explosive expansion of chips, memory, data centers, and power infrastructure.
This article caused a sensation in Silicon Valley. Some have called it "the most important article in a decade." Podcast host Tim Ferriss has dubbed him "the Nostradamus of AI."
In 2024, he founded the namesake fund SA with approximately $225 million in seed funding. Investors include Stripe co-founders Patrick and John Collison, former GitHub CEO Nat Friedman, and institutions such as Jane Street.
With no prior industry experience, he stepped in to manage tens of billions of dollars. The fund's assets rapidly expanded, reaching a peak of nearly $45 billion in early July 2026.
The entire fund has only eight employees, four of whom are investment professionals.
The "AI Oracle" goes all-in on AI with 4x leverage, earning 439% in six months
SA's investment logic is straightforward: since AI requires massive computing power and infrastructure, heavily invest in companies that provide computing power and infrastructure.
Bullish positions: SK Hynix (South Korean semiconductors/memory), Nebius (AI cloud platform), SanDisk (storage chips), Micron (memory chips), CoreWeave (AI cloud computing), Bloom Energy (energy).
Short position: Short traditional software companies like Adobe—the rationale is that AI will disrupt them.
At the same time, the fund amplifies returns by borrowing from banks (i.e., leverage), with a leverage ratio of approximately 4x.
This structure has a key feature: banks hold physical shares, while SA gains economic exposure and leverage synthetically through a "total return swap" (TRS). The benefit is amplified returns, but the downside is that if the stock price falls, the bank may require additional margin (a margin call).
From 2025 to the first half of 2026, this strategy yielded astonishing results.
As of the end of June 2026, SA achieved a net return rate of 439%. According to reports, the fund has generated cumulative returns exceeding 2000% since its inception. The fund's size grew from less than $1 billion to a peak of approximately $45 billion in early July.
Some early investors have achieved returns of up to 200% this year.
Within a month, hit from both sides
However, the good times didn't last. In July, AI stocks began to decline.
The market is beginning to question: Can the high valuations of AI infrastructure be sustained? Significant capital is flowing out of AI hardware stocks.
SA's long positions are hit first:
Nebius has dropped approximately 48% from its June high, with a market cap loss of about $35 billion.
SanDisk has dropped over 56% in more than a month.
CoreWeave, Micron, and SK Hynix have all declined more than 35% this month.
Meanwhile, software stocks shorted by SA, such as Adobe, have risen in the opposite direction, causing losses for short positions.
Both long and short positions are losing money, and 4x leverage has amplified the losses. The bank has started issuing margin calls.
This is what Wall Street calls a "margin call": you borrowed money to buy stocks, the stock price dropped, and the bank demands you deposit additional collateral, or else your position will be forcibly liquidated. This is the same logic as margin liquidation in China's A-share market.
SA's prime brokers include Bank of America, Goldman Sachs, and JPMorgan Chase, all of which are involved to assist the fund in meeting margin calls or executing orderly liquidations.
Interestingly, on July 24, Aschenbrenner wrote in a letter to investors: "Sometimes we specifically point out that now is a great time to add to your position—if you’ve been waiting." The letter concluded with a postscript inviting investors to add funds on August 1.
36 hours, one-night auction
The crisis evolved rapidly within 36 hours.
In the late hours of July 29, SA sent urgent appeals to multiple potential buyers seeking to sell over $10 billion in shares to stabilize the situation. This news was widely reported by media outlets including the Financial Times and Bloomberg.
Bidders include: Millennium Management (a top global multi-strategy hedge fund) and Jane Street (an investor in SA).
Millennium made a bid but considered the position too risky and was unwilling to offer a high price.
Ultimately, Citadel emerged victorious.
On the morning of July 30, the trade was completed. Citadel acquired all of SA’s publicly traded stock positions—both long and short—amounting to approximately $16 billion at a significant discount. This was one of the largest emergency stock trades in Wall Street history.
After the news of the completed transaction, the Nasdaq 100 surged more than 3% that day—market participants breathed a sigh of relief: SA no longer needed to dump more assets into the market.
This is not the first time Citadel has done something like this. Historically, it has acquired assets from companies like Enron during crises at discounted prices. In Ken Griffin’s own words, one reason Citadel has sustained success is: “Experience—the wisdom gained through losses and pain. My leadership team and I have been through many of the market’s toughest moments and learned many hard lessons. But it has made us more effective investors during times of turbulence and crisis.”
Citadel was founded in 1990. SA was founded in 2024.
What's left?
After the transaction, SA's assets under management dropped sharply from a peak of $45 billion to approximately $10 billion, a decline of more than half.
But Aschenbrenner was not completely emptied.
SA retained its entire private placement position, approximately $10 billion, with the most significant portion being its equity stake in Anthropic, valued at around $5 billion. SA will continue to operate as a private investment firm.
The ability to liquidate this private placement portfolio depends on the valuation of companies like Anthropic when they go public.
This is an unresolved issue.
By the way, Aschenbrenner’s fiancée, Avital Balwit, is the Chief of Staff to Anthropic CEO Dario Amodei. SA was previously rumored to be selling its Anthropic equity stake, but a spokesperson for SA denied this claim.
Where did he go wrong?
The most important question to consider is: Was Aschenbrenner’s assessment of AI wrong?
Not necessarily.
The long-term demand rationale for AI infrastructure is still widely accepted by numerous institutions.
The issue lies in the position structure.
On X, someone quoted Ken Griffin’s previous description of why fund managers fail: "Your portfolio is highly concentrated with massive positions, yet you cannot clearly articulate what competitive advantage you have in holding these positions."
The problem with SA is exactly this: heavily weighting a single theme (AI infrastructure) while using 4x leverage to bet on both sides of the same narrative. When market sentiment shifts, there’s no buffer.
This structure can generate a 439% return during an uptrend and can wipe out the entire open position within weeks during a downtrend.
As someone summarized: "His judgment on AI was correct. He was wrong about how he constructed his position."
The other side of the myth
Aschenbrenner's story is one of the most extreme examples of this AI bull market.
One article, one narrative, eight people, $45 billion, 439% returns—and then, within a month, crushed by margin calls.
This is not the end of AI, nor the end of Aschenbrenner.
But it clearly illustrates one thing: seeing the right direction doesn't mean you'll live to see it realized.