According to Woofun AI, Leopold Aschenbrenner, a young trader dubbed the 'AI Stock God', is trying to rebuild order in the rubble after experiencing a sharp shock from $10 billion in leverage to liquidity traps. This German teenager went to Columbia University in New York at age 15 and was amazed by economist Tyler Cowen with his extraordinary intellectual performance; at age 19, he graduated from Columbia University as a representative graduate and immediately joined the FTX Future Fund, a charitable fund under SBF, where he met his later wife Avital Balwit, who is now Anthropic CEO Dario Amodei's chief of staff.
After the FTX crash, he briefly joined the OpenAI Super Alignment team, but left his job less than a year later due to an internal security memorandum incident. In June 2024, he published the 165-page long article “Situational Awareness: The Decade Ahead” (Situational Awareness: The Decade Ahead), which used this as an opportunity to establish a hedge fund of the same name and begin his legendary and controversial journey in the capital market.
The rise of the Situational Awareness Fund is based on surprising predictions of macroeconomic trends. The initial capital of the fund was $225 million, and the main investors included Stripe's founding partners, the Collison Brothers. In that 165-page long article, Aschenbrenner predicts that the world's largest AI training cluster will reach about 1 million H100 equivalent sheets in 2026, consume 1 gigawatt of power, cost tens of billions of dollars, and use Hoover Dam as a reference.
This prediction, once viewed as a sci-fi fantasy, was proven two years later: Anthropic and Amazon's New Carlisle (1037 megawatts) in Indiana, XAI's Colossus 2 (946 megawatts, approximately 1.11 million copies of H100 equivalent), and OpenAI's Stargate Abilene (824 megawatts, approximately 1.02 million H100) in Texas — their scale and power consumption have been Approaching or surpassing their predictions.
Furthermore, he predicts that the total annual investment related to global AI will exceed $1 trillion by 2027, while Moody's estimates in July 2025 show that the total capital expenditure of the six companies Microsoft, Amazon, Meta, Alphabet, Oracle, and CoreWeave is expected to reach $785 billion, and is expected to approach $1 trillion by 2027. The precision of macro deduction provides strong logical support for its transactions in the secondary market.
Aschenbrenner's trading style combines a macro vision with extreme leverage. He once resolutely shorted the stock market when the mayor of New York advised the public to look at Broadway at the end of February 2020, and when DeepSeek R1 caused a panic in chip stocks in January 2025, he worked with fund research director Carl Shulman to determine that the market was overreacting and bucked the trend. According to fund advisor Graham Duncan, to “Fortune” magazine, a large technology fund at the time gave up selling because analysts said 'Leopold says it's fine'.
This confidence has paid off amazingly: in the first half of 2025, the fund's return after deduction reached 47%; as of May 2025, the revenue after deduction during the year was about 270%, more than tenfold since inception. Even quantitative giant Jane Street joined the LP ranks. An investor letter in the middle of the year also revealed that the return after deduction in the first half of the year was as high as 439%.
However, behind this 439% return, lies up to 4x leverage and the 'AGI shock theory' described in the 165-page PDF. Aschenbrenner sees the secondary market as a primary market, uses a VC style of play, and fluctuates heavily after adhering to an ambitious direction. According to the Form ADV document updated in June 2025, Situational Awareness, which manages assets worth tens of billion dollars, has only 8 employees, of which only 4 are responsible for the investment. However, this “caucus team” borrowed tens of billions of dollars from the bank through the main broker business. Before the incident, Barclays Bank rejected its business due to excessive industry concentration, and Goldman Sachs took over and made it the largest hedge fund main brokerage customer, paying more than 200 million US dollars in fees in the first half of the year alone. Even on the eve of the July crash, he was still in negotiations to increase his leverage from 4x to 10x.
As leverage climbed, position concentration also increased dramatically. At the end of the first quarter of 2025, Form 13F showed that it held put options with a nominal value of approximately US$8.46 billion, covering semiconductor ETFs (SMH), Nvidia, Oracle, Broadcom, AMD, Micron, TSMC, and ASML to insure long positions. However, by the end of the second quarter, these umbrellas were completely removed, and the total declared holdings surged to US$20.04 billion, of which Sandisk (SNDK.US) and Micron (MU.US) two memory chip stocks accounted for 55%. In the private market not covered by 13F, he was not only the cornerstone investor of SK Hynix's NASDAQ listing in July, but also purchased 16.61% of the shares of Taiyo Yuden, a major Japanese passive components manufacturer, in the Tokyo market.
Hedging was removed, leverage was increased, and 10 billion US dollars were fully invested in storage and power infrastructure. The July crisis then broke out: from July 1 to 29, Sandisk, the largest stock company, fell from $2,032 to $1,016, Australian computing power stock SharonAI fell by nearly half, fuel cell leader Bloom Energy (BE.US) fell by more than 40%, and Adobe, which had gone short, bucked the trend. Due to the transparency of the documents disclosed by 13F and major shareholders, Wall Street was aware of its security deposit recovery pressure and targeted with precision.
The 48-hour crisis management at the end of July became a classic case in financial history. Citadel executives contacted Sven Khatri, who had just switched jobs, on Tuesday, July 28, according to WSJ's review. To avoid open market stocks being trampled on, the fund attempted to sell Anthropic private equity shares worth about US$3.5 billion at 20% off for cash on Wednesday, July 29, and sought quotes from Sequoia Capital, Greenoaks, and the Michael Dell family offices, requiring a deal within 12 hours and cash delivery before 8 a.m. the next morning.
The Greenoaks team stayed up late to do their best, and Anthropic officially signed and released overnight. The turning point came on Thursday, when negotiations between the fund and Citadel narrowed down to only taking over leveraged stock positions. Citadel head Ken Griffin made a remote deal and bought it at 10% off the market price. The two sides signed a contract about 20 minutes before the opening of the US stock market. After receiving the life-saving money, the fund immediately withdrew from the Anthropic old stock deal, leaving top Silicon Valley buyers happy. There is a 10% discount on meat cuts in the open market, and the primary market chips are preserved.
The shares taken over by Citadel rebounded violently after opening. Nebius (NBIS.US) and Bloom soared 27% on the same day, and IREN (IREN.US) surged 31%. On August 3, 10.81% of his shares in Taiyo Yuden were split in the Japanese OTC market. Citadel took 3.41% (at a price of 9397 yen per share), and J.P. Morgan, Barclays, and Jane Street split the remaining shares (at a price of 10,225 yen per share). This loss of around $30 billion to $35 billion is among the largest single transaction losses in financial history.
Reborn after the heist, Aschenbrenner did not leave the table. On August 25, the SEC issued subpoenas to four major Wall Street banks regarding fund transactions and leverage; in September, J.P. Morgan cut off financing support. Instead, he sought the support of boutique brokerage firm Clear Street to return to the options market. The third-quarter 13F submitted at 4:30 p.m. on November 16 shows that the position structure for the rest of the life after the hijacking has changed drastically: the entire form has only 23 lines, and the total value of declared positions is US$1,577 billion, which is a sharp decrease from $20.04 billion three months ago. In a letter to investors at the end of July, he announced that all of his short positions had been closed and that his open market positions had been changed to using only his own funds to buy spot goods and options that paid the premium in full, and completely bid farewell to bank guarantees.
Although options come with high leverage, there is an upper limit on losses, which avoids the risk of being weakened. Of the total declared value of $1,577 billion, $1,045 million were call options, and the targets were still Bloom, Nebius, Sandisk, and a memory chip ETF — these were the stocks that almost crashed in July. Of the remaining US$530 million in stock, there are only five stocks: the three Bitcoin mining companies that have transformed computing power totaling about US$80 million, the Texas energy company Fermi accounts for 11.25 million, and the largest sum of US$441 million is wholly invested in Australian computing power company SharonAI.
Furthermore, the private market not covered by 13F still holds about $10 billion in assets, including Anthropic shares valued at around $5 billion, and chip lithography startup Source Foundry, which has invested a cumulative investment of $500 million. According to the Nebius 13G/A disclosure document submitted on the same day, as of September 30, he effectively held 8.01 million Nebius shares, equivalent to approximately US$1.89 billion at the stock price at the time.
Data compiled by Woofun AI showed that SharonAI's case profoundly revealed the liquidity trap in Aschenbrenner's strategy. Media reports on November 16 interpreted his position as rebetting on AI after bursting out, but SEC historical documents show that he already held about 800,000 shares at the end of the first quarter. SharonAI is a small Australian company listed under a backdoor liquidation SPAC. It was announced on June 12, 2025 that it had obtained Nvidia's cooperation to deploy up to 40,000 GB300 chips. Five days later, it completed huge financing of US$1.6 billion, of which US$900 million was a fixed increase of US$68.73 per share, US$700 million was convertible bonds, and the lead investors were Aschenbrenner's fund and Oak Capital. He invested $524 million to buy 21.1% of the shares.
However, shares acquired through PIPE are in a lockdown period until the SEC resale registration takes effect and cannot be sold in the secondary market. As a result, SharonAI was not on the emergency list sold to Citadel at the end of July. It's not that they don't want to sell it, but that they can't. The 5.396 million shares at the end of the second quarter remained unchanged until the third quarter, with only 2.67 million pre-financing warrants exercised at a symbolic price of $0.0001 per share on September 15. Fixed increases, warrants, and lock-up periods, these common methods of primary market VC soldering liquidity to the top of the mountain when taking heavy positions in small-cap stocks in the secondary market. According to the three-quarter report handed over by SharonAI on November 16, revenue for the quarter was only 9.93 million US dollars. In September 2025, the company spent approximately 50 million Australian dollars to buy an abandoned paper mill site near Albury, New South Wales, Australia to transform a data center. By the close of the day, the stock price had fallen to 20.13 US dollars, and the 8.07 million shares in Aschenbrenner's hands were worth only 160 million US dollars, down nearly 70% from the investment made five months ago.
More seriously, electricity bottlenecks and political resistance are becoming fatal constraints on AI infrastructure. In a long article, Aschenbrenner asserted that “the biggest hard constraint on the supply side is probably electricity,” and believes that the US has sufficient natural gas reserves and only needs 1,200 additional gas wells to meet the 100 gigawatt computing power cluster needs. As long as environmental approvals are loosened and federal regulations make way, everything will be solved, and even emphasized that “no one can match America's construction capacity, at least in Red State.”
However, reality has proven that electricity is a boring physical and administrative issue that cannot be queued. The largest single campus planned by the end of 2028 is only about 2 gigawatts, far from 10 gigawatts. In July 2025, PJM, the largest regional power grid operator in the US, held a capacity auction. Electricity prices reached the federal approved limit. The grid gap reached 6831 megawatts, but only 525 megawatts of new or upgraded power generation capacity were added or upgraded. Of the $164 billion in grid capacity costs, about $6.3 billion was driven up by data centers.
The cost was transmitted to residents. Electricity prices for US residents increased by an average of 5.4% in 2025; on the industrial side, large-scale power transformers needed to wait about 128 weeks for delivery, and the median queue for new power generation projects to be connected to the grid exceeded 5 years. Even Red State, where he had high hopes, gave an impetus: on September 21, Texas Republican Governor Greg Abbott ordered the suspension of approval of all state-level licenses relating to data centers; in the November 3 midterm elections, Texas elected its first Democratic governor, Gina Hinojosa. One of its core promises was to freeze data center expansion. Goldman Sachs estimates that similar moratoriums across the US could jam 45% of planned data center capacity.
Fermi, which is heavily stocked by Aschenbrenner, led by former Texas Governor and former US Secretary of Energy Rick Perry, leased about 5,000 acres of land from Texas Tech University near Amarillo, Texas, and plans to build an 11 to 17 gigawatt superenergy park, first using natural gas, and then launching four AP1000 nuclear power plants. Fermi (FRMI.US) was listed in October 2025, and the stock price reached $36.99 on the second trading day.
Trouble ensued: discussions with Amazon's computing power crashed, the co-founder was fired, Perry failed to be re-elected as director, tenant TensorWave tore up the lease, and the 5-gigawatt gas emissions license collided with Abbott's moratorium. On November 5, Fermi's stock price dropped to $1.44, breaking the ankle. Another major stock, Nebius, is a computing power park designed for Microsoft in Vineland, New Jersey. Some deliveries were delayed until the second quarter of 2027 as the site operator withdrew the temporary diesel power supply and waited for Bloom fuel cells to be installed.
On the second trading day after the midterm elections on November 5, Sandisk plummeted 22% in a single day, and SharonAI fell 20%. Even though Bloom Energy announced that NTT DATA and Google had ordered more than 1,600 megawatts, the stock price still smashed 16%. According to an August poll by the University of Massachusetts, only 11% of residents support building AI data centers near the community; Motley Fool statistics, in 2026, 41 data center site referendums and hearings across the US, the opposition won 40.
Stock options such as Bloom, which Aschenbrenner bought back in early September, were 68% more expensive than before cutting meat on July 29, and Nebius was 52% more expensive, but after September 30, AI infrastructure stocks fell sharply again, and both Nebius and Sandisk were slashed. On November 16, Nebius shares were 28% lower than the price it sold to Citadel at the end of July, and 20% lower than the price of Sandisk. The July crash can be blamed on leverage and sniper, but the declines in October and November revealed deeper problems when bank security deposits were removed and there was no large-scale consolidation.
The underlying reason is that AI has crashed from the 'bit' industry into the 'atomic' world, and Aschenbrenner's understanding of 'bottlenecks' is misplaced. In economics and VC logic, bottlenecks mean scarcity and monopoly profits; from an industrial perspective, bottlenecks mean work stoppages, overspending, frauds, and a long payback cycle. His heavy storage chip inventory is a strong cycle business, and gross margin can drop from 80% to -30% within a year. The AI business is getting heavier, and the required knowledge structure has changed. In addition to engineering and political frictions, the atomic world is extremely cash-intensive, especially borrowed money.
Two years ago, Aschenbrenner predicted in the Dwarkesh Patel podcast that people in the computing power cluster would “borrow money like crazy” and push up actual interest rates. The borrowing wave is on schedule: According to the Federal Reserve's November 13 “Financial Stability Report”, the global issuance of AI computing power-related bonds and debts exceeded 680 billion US dollars in the first 10 months of 2026, compared to only about 130 billion dollars for the full year of 2023. He predicted that around 2026, Google or Microsoft could reach 100 billion US dollars in annual revenue from AI alone, but Microsoft's latest financial report revealed that AI's annual revenue is only about 37 billion US dollars.
Most of the revenue remains with OpenAI, Anthropic, and Nvidia, while midstream second-tier Neocloud, fuel cell plants, and transition mining companies receive high depreciation and interest on debt issuance. The $100 billion gap between upstream revenue and downstream infrastructure investment was bridged by the Wall Street bond issuing machine. The Nebius, SharonAI, Bloom, and mining companies he is betting on are the weakest part of the industry chain that relies the most on borrowing the old and new. He accurately predicted that the industry would borrow money like crazy, but he was betting his net worth on borrowers who were most afraid of bank suspensions.
According to Wall Street rules, young fund managers who have taken tens of billion dollars in positions within half a year and lose money once again have basically come to an end. But Silicon Valley was abnormal: On July 30, the day news of Citadel's 10% off harvest broke, former legendary hedge fund manager John Arnold wrote on X: “The best number of times to close a position in the past is once.” The next day, the title of the Silicon Valley podcast TBPN directly typed “Leopold is Back,” and Richard Craib, founder of quantitative fund Numerai, talked about “Investor Hero Cult” in the second half.
On August 7, Bloomberg reported that top investors in Silicon Valley are lining up to put new money into his fund. Sequoia Capital partner Pat Grady publicly stated that he will be an important figure in Silicon Valley for many years to come, and the fund's official response is that it will not accept new external funding yet. Wall Street reads jokes, and Silicon Valley hands out checks, showing that Silicon Valley is sick. First, I am superstitious about Zaohui. Aschenbrenner once said he can't stand Europe's 'high poppy syndrome', but Silicon Valley loves to give the first bird a tenfold leverage. The 17-year-old was funded by Tyler Cowen's Emergent Ventures. Cowen called him one of the top young economists and advised him not to study.
From the 19-year-old Columbia No. 1, to the 22-year-old OpenAI whistleblower, to the 23-year-old head of the $10 billion fund, Silicon Valley needed the MSI script. Second, the late Japanese view of time. Former OpenAI director Helen Toner pointed out to The Atlantic Monthly that its huge losses stemmed from a single worldview with 'only one definite end'. Now that the road to the super-smart end in 2027 has been nailed, the security deposit recovery, EPA hearings, and transformer delays have all become noise, and risk control seems timid. Third, addiction to 'grand insight'.
Amjad Masad, founder of Replit, says Silicon Valley is hungry for 'Insight Porn (Insight Porn) ', and anyone who can write a grand story that looks into the next ten years can suck money and attention away. Bestselling author Tim Ferriss calls him “Nostradamus in the AI world,” and domestic shareholders call him “AI stock god.” His best craft is to attract attention. Fourth, ideology stands up. After the sale was sold out, podcast host Dwarkesh Patel posted a strong post, and the accelerationist (e/acc) camp leader Guillaume Verdon saw the collapse as a slap in the face against “rationalists who are close to effective altruism,” and both sides used their net worth curve as a weapon.
Those who clap on the shore are not the same group as those who pay their tuition fees in the water. Most of the old LPs who paid the money were tech upstarts and top family offices in the Bay Area. After a 67% retreat in July, their net worth still had positive returns of about +80% since 2026. They have doubled several times since the fund was established, so naturally they can sit back. What was really buried at the top of the mountain was a follower of the wind later. According to the regulatory document Form D, the cumulative subscription amount of funds surged by US$7.57 billion from March to September, when the myth of “making a profit of 439% in half a year” was circulating, with 79 new investors. Quantification giant Jane Street lost about $15 billion overall in a single month in July 2026, and it is reported that a significant portion was discounted to this fund.
As for Aschenbrenner himself at the center of the storm, he calmly dissected the chips in a podcast two years ago: On the eve of AGI, the rapidly depreciating assets in his hands were human capital. ' I'm a graduate representative at Columbia University. What made me unique in the past was my cleverness. ' After another four years, smart or automated by machines. A friend joked that setting up a hedge fund was a “perfect hedge”: if AGI arrived within three to five years, he would have exchanged depreciated human capital for tens of billions of dollars of financial capital; if AGI hadn't come and the fund was destroyed, “I'm still in my 20s, and I'm still that extremely smart person.” There's nothing wrong with this statement. After turning Wall Street upside down, he will actually be in his 20s in 2026.
The historical mirror overlaps at this moment. In 2006, Brian Hunter, a 32-year-old Canadian genius trader, bet on the winter and spring gas futures spreads at the hedge fund Amaranth Advisors. One person supported the company's profits, and the fund's assets once reached 9.2 billion US dollars. In September of that year, the weather forecast and market spread were reversed, and Amaranth went up in smoke within a month, losing more than 6 billion US dollars. Citadel and J.P. Morgan Chase are the two organizations that took advantage of the flames to rob and take over Amaranth's energy holdings at a discount.
Another 32-year-old Texas gas trader, John Arnold, made a huge profit from Hunter, whose Centaurus Energy fund recorded an impressive 317% return. Later, Hunter tried to make a comeback. In 2007, Zhang Luoxin Fund reportedly received a subscription of 800 million US dollars, but was unable to open due to regulatory lawsuits; in 2014, he reached a settlement with the US Commodity Futures Trading Commission, handed over fines, and accepted a permanent trading ban, and completely withdrew.
Exactly 20 years later, on X, understate that “the best number of times to close a position in the past is once” to break the siege for young people. It was John Arnold who won that year. He even told the “Atlantic Monthly” reporter that Leopold's 10 billion dollar gold cutoff in July was just a “small episode” in a long cycle. Further ahead, there is also a groundwork laid by Leopold himself. On the podcast two years ago, Dwarkesh talked about the biography of Peter Thiel, the godfather of Silicon Valley, and mentioned that Thiel personally went to work as a macro hedge fund in his early years and had poor performance.
Aschenbrenner took it on: 'It blew up. This is the opposite case; it's too theoretical. ' Dwarkesh quoted the book as saying that Thiel's fund is essentially a 'think tank hidden in a hedge fund shell'. 23-year-old Aschenbrenner's eyes lit up: 'This is exactly what I wanted to build. ' Peter Thiel's fund that lost $7 billion and went into liquidation was called Clarium Capital.
Coincidentally, Carl Shulman, now co-portfolio manager with Leopold, worked at Clarium Capital in his early years. As for “Situational Awareness (Situational Awareness),” which he used as the title of his famous long article and printed on the fund's door, the concept first dates back to World War I air combat pilots, and was called by the US Air Force after the North Korea and Vietnam Wars. In her 1995 paper, engineer Mica Endsley broke it down into three steps: the first layer, sensing the basic elements of the environment; the second layer, understanding the meaning of combining elements; and the third layer, predicting the next second state.
Each layer must step on top of the previous one. Endsley counted American Airlines air crashes from 1989 to 1992 due to pilots' human error, and found that 88% of the accidents had problems with pilots' 'situation sense', staring dead at distant targets or instrument readings, but missed freezing wings under the eyelids and rapid drop of the altitude pressure gauge. Two years ago, 23-year-old Leopold Aschenbrenner sat in front of a microphone in San Francisco and enthusiastically said to the world, “Our perception of the situation will far surpass those who manage money in New York.” Now it seems that he may have seen superintelligence in the distance, but he is ignoring the frozen layer of fluid ice underneath his feet.