AI-Focused Fund SA Collapses Amid Market Correction

By Cui Peng CP

We are one step away from permanent capital loss.

At the end of July, when the hotly anticipated fund Situational Awareness (SA) sent this email to its investors, it was slipping toward collapse.

This fund, founded by former OpenAI researcher Leopold Aschenbrenner, grew its assets under management from $1.5 billion to $45 billion within a year—just six months ago, it was Wall Street’s most coveted success story.

Over the past few years, when attending various dinners in San Francisco, Ashenbrenner often spoke about his dream of buying a galaxy. He believed that with AI advancing so rapidly, humans would soon be able to colonize other planets, and he needed to start saving now.

To invest in extraterrestrial galaxies, he needs a lot of money, and SA is the best tool for earning it.

The capital markets have also shown great enthusiasm for this 24-year-old trading prodigy, with investors including Stripe co-founders the Collison brothers, former GitHub CEO Nat Friedman, and quant giant Jane Street all competing to invest in his fund and secure a share of the upside.

However, during the major correction in the semiconductor sector in July, this fund lost 67% of its investment gains within a week, narrowly avoiding a cascading liquidation disaster. In the end, SA was forced to urgently sell its public market positions to alleviate the margin crisis and prevent forced liquidation by the bank.

The current AI investment narrative has lasted nearly two years, during which various criticisms have failed to deter successive waves of investors. Only after the SA collapse did the market begin to seriously consider a question long overlooked: Does the way capital is wildly betting on AI inherently contain systemic risk?

Ashenbrenner’s story initially sounded like another Silicon Valley and Wall Street fairytale. Public reports stated that he graduated from Columbia University at age 19 as the top student in both economics and mathematical statistics, before joining OpenAI’s renowned Superalignment team.

The team, co-led by Ilya Sutskever and Jan Leike, former chief scientist at OpenAI, aims to tackle cutting-edge safety challenges around aligning superintelligent AI with human values.

In April 2024, OpenAI fired him for "improper disclosure of internal information." Two months later, Ashenbrenner released a 165-page report titled "Situational Awareness: The Decade Ahead," predicting the evolution of AI capabilities—it stunned the entire industry and quickly became required reading among Silicon Valley peers.

Many investors firmly believed in his ability to predict the trajectory of AI technology, and this reputation later directly translated into fundraising capability, leading to the founding of the Situational Awareness hedge fund.

Even in Wall Street, a hub brimming with geniuses, the fund's growth rate is astonishing. According to holdings tracking data, SA began in June 2024 with $383 million in seed capital, and by the first quarter of 2026, its assets under management had surged to approximately $45 billion.

A team of fewer than twenty people matched the assets under management of many established funds with decades of operation in less than two years—a rarity in the history of the fund industry.

SA's strategy is straightforward: go long on hardware and computing stocks benefiting from the AI infrastructure boom (such as SK Hynix, CoreWeave, and Micron), while shorting software companies perceived to be at risk of being displaced by AI (such as Adobe and AppLovin).

This sounds like a hedging structure, but it's actually an amplifier for one-sided bets—both long and short positions are driven by the same factor: market confidence in AI.

During the market rally, this strategy generated remarkable excess returns, with a peak annual yield of 439%. According to an exclusive report by the WSJ last week, SA demonstrated extremely high leverage efficiency, previously able to leverage $1 of its own capital to secure approximately $3 in borrowed funds. Major institutions including Goldman Sachs, JPMorgan Chase, Bank of America, and Citibank provided financing for it, with Goldman Sachs even showcasing this fund as a benchmark case to clients at the Emerging Managers Conference in March 2025.

Reports indicate that both Jefferies and Barclays declined the business; after meeting with Ashenbrenner, a prime brokerage executive said that his unshakable confidence itself was a red flag. In hindsight, this proved to be well-founded.

As mid-July approached, Wall Street began reassessing the capital expenditure (AI Capex) rationale for AI infrastructure, and the semiconductor sector officially began to correct on the 24th.

According to FT, on that day, Ashenbrenner sent an安抚 email to investors, confidently stating that the recent stock price decline was "the best opportunity to add positions in the past year," and invited investors to increase their investments starting August 1.

Everyone now realizes that this was not the best opportunity to add to positions, but rather the final chance to exit.

Within a few days, the Philadelphia Semiconductor Index fell 28.6% from its late-June high, with SK Hynix, Sandisk, Bloom Energy, Nebius, and other holdings heavily weighted in SA all declining more than 30% in a single month.

SA's position gains have plummeted consecutively, with the shadow of high leverage looming over the entire fund. On the morning of July 30, Goldman Sachs, JPMorgan Chase, and Bank of America simultaneously issued margin calls to SA, leaving Ashenbrenner with few options remaining.

According to the WSJ, the day before, SA had urgently contacted Anthropic shareholders such as Sequoia and Greenoaks to discuss a discounted sale of private equity shares, while also engaging in separate negotiations with Citadel, whose founder and CEO, Ken Griffin, joined the call from London.

The two parties finalized the transaction at 3:00 AM, with Citadel purchasing all of SA’s public market positions at a discount of approximately 10%. The agreement was signed at 9:10 AM, just 20 minutes before U.S. markets opened.

This is the real deadline: After the stock market opens, banks no longer need to gradually sell SA's holdings into the open market, preventing a chain reaction of forced liquidations and further fund margin calls.

After the transaction, SA's assets under management dropped from a peak of $45 billion to approximately $10 billion; however, public reports indicate that its private equity stake in Anthropic was not included in this liquidation, and this holding is widely regarded as SA's foundation for a potential comeback.

Some narratives have sensationalized this event as "a genius trader falling just before dawn," but this billion-dollar cautionary tale should not be simplistically reduced to a matter of luck—it may be a sign that the entire AI investment narrative needs to be reevaluated.

When news of SA's crash-style losses emerged, the market's first reaction was disbelief—how could the fund's hedging mechanisms have failed to halt the cascade of losses?

Because a standard hedging strategy requires two positions to react oppositely to the same shock, SA’s strategy involves going long on AI-benefiting stocks while shorting AI-hurting stocks.

The biggest issue with this strategy is that both long and short positions react identically to shifts in "AI confidence," creating the illusion of a short position while actually increasing leverage and multiplying the risk.

John Pfeffer, founder of Pfeffer Capital and one of the SA investors, previously defended the strategy by stating that it was never designed for conservative pension funds, and everyone understood it would experience significant volatility.

However, he and Ashenbrenner clearly underestimated the magnitude of this volatility. When the semiconductor sector declined in July, SA’s long positions sustained losses, while its short positions also lost money due to the rise in AI-affected stocks (the software sector). The double blow caused SA’s returns to plummet sharply, falling from a peak of 439% to just 80%.

Ashenbrenner’s gambling-like actions have rendered the short-selling mechanism ineffective, depriving it of its intended hedging function. This type of "pseudo-hedging" structure has a bitterly ironic name on Wall Street—the "Texas hedge"—akin to gambling at Texas Hold'em poker.

Wall Street has long witnessed its power, whether in the stock market crash triggered by the dot-com bubble burst at the turn of the millennium or the collective collapse of quantitative funds in 2008—its influence was present in both.

As LPs, investors should closely monitor the health of the fund’s portfolio, but they often become complacent about the fund’s risk exposure due to the illusion of being "hedged" on paper. During periods of one-sided market rallies, both the fund and investors benefit greatly; however, once the market turns downward, both parties face losses far exceeding normal levels.

According to the WSJ, the renowned quantitative fund Jane Street, as an investor in SA, suffered an estimated $6.5 to $7 billion in paper losses due to SA’s drawdown in July alone, surprising the market.

In SA’s public holdings for the first quarter of this year, over 60% of the risk exposure was concentrated on a few AI-related stocks, further amplified by approximately 3x leverage. Additionally, SA’s private equity stake in Anthropic in the primary market cannot be quickly liquidated and therefore cannot be used to meet intraday margin requirements, making it highly susceptible to liquidity crises at critical moments.

These are typical risk factors associated with high concentration, high leverage, and medium-liquidity标的. When they occur individually, many funds have sufficient buffers to absorb the risks.

But when all three are pushed to their limits, the fund will simultaneously face the pressures of losing money quickly, being unable to sell, and failing to meet margin requirements—even in the event of a brief market reversal.

So when semiconductor stock prices corrected in July, SA went from incurring losses to being forced to liquidate in less than two weeks.

Bank of America CEO Brian Moynihan said in a recent CNBC interview that this is a wake-up call for mainstream financial firms to reassess their exposure to AI-themed funds.

There are also optimistic voices in the market; the Financial Times believes that despite the significant correction in this cycle, SA’s core assessment regarding growing demand for memory chips, data centers, and computing power has not been disproven.

Unfortunately, even if the long-term logic behind trading stocks is sound, the repayment pressure from high leverage can cause investors to fail before the logic is realized.

SA's full-year return was still around 80% by the end of July, which may seem substantial—but this gain assumes the fund remained solvent. Had Citadel's rescue not arrived in time, SA would most likely have faced a forced liquidation by banks.

For a long time, Wall Street has had this investment adage: Markets can remain irrational longer than you can remain solvent.

So the real question left by this crisis has long since moved beyond Ashenbrenner’s views on AI—it’s whether capital markets will continue to ignore this systemic risk in their frantic pursuit of the next “certainty” AI narrative fairy tale. (Featured image source: Margin Call)