Abstract
Leopold Aschenbrenner, a German-born former OpenAI researcher who graduated as Columbia University valedictorian at age 19, rose rapidly as a public intellectual on artificial general intelligence (AGI) timelines after publishing his influential June 2024 essay series Situational Awareness. He founded the AI-themed hedge fund Situational Awareness LP later that year. The fund delivered extraordinary early returns on concentrated bets on AI infrastructure (semiconductors, high-bandwidth memory, data-centre operators, and power assets), grew to a peak reported as high as approximately $45 billion in assets under management by early July 2026 according to sources cited by CNBC, and then suffered a severe drawdown. In July 2026, leveraged long positions in AI infrastructure stocks and adverse short positions triggered margin calls; the fund was forced to sell the bulk of its public equity portfolio to Citadel. The portfolio fell roughly 67% that month. The fund remained substantially positive year-to-date (around +80%) thanks to earlier gains and retained private holdings (notably a significant stake in Anthropic), and it continues to operate in a de-levered form. The episode illustrates both the power of a high-conviction technological thesis and the classic risks of concentration, extreme leverage, limited traditional risk-management experience, and short-term market cycles.
1. Early Life, Education, and Pre-Fund Career
Aschenbrenner was born in Germany (circa 2001/2002) to physician parents and attended the John F. Kennedy School in Berlin. He entered Columbia University at age 15 and graduated in 2021 at age 19 as valedictorian with a BA in economics, mathematics, and statistics. He received academic prizes, including recognition for best senior thesis in economics. While at Columbia he co-founded the university’s effective-altruism chapter and later conducted research at Oxford’s Global Priorities Institute.
He spent a period with the FTX Future Fund (an effective-altruism philanthropic vehicle linked to Sam Bankman-Fried) before resigning ahead of FTX’s 2022 collapse. In 2023 he joined OpenAI’s Superalignment team (led at the time by figures including Ilya Sutskever and Jan Leike), which focused on technical approaches to controlling systems more capable than humans. He co-authored work including the paper on “Weak to Strong Generalization.”
In April 2024 OpenAI terminated his employment. The company cited alleged improper disclosure of internal information (sharing of a planning document with external researchers). Aschenbrenner has disputed the characterisation, stating that tensions arose after he circulated a security memo to board members following an internal security incident; the memo warned of industrial-espionage risks (particularly from China) regarding model weights and algorithmic secrets. He has said the security memo was explicitly given as a major reason the matter was treated as a firing rather than a warning. OpenAI has denied any connection between the security concerns and the termination. The Superalignment team itself largely dissolved later in 2024 amid broader departures.
2. The Situational Awareness Essays (June 2024)
Shortly after leaving OpenAI, Aschenbrenner published the long-form essay series Situational Awareness: The Decade Ahead (available at situational-awareness.ai). The core argument rests on observed scaling trends: roughly half an order of magnitude per year in effective compute and another half from algorithmic progress and “unhobbling.” Extrapolating from GPT-2 to GPT-4, he projected systems that would outperform many college graduates by 2025/26 and AGI around 2027, followed potentially by a rapid intelligence explosion as AI systems automate AI research itself.
He forecast massive industrial mobilisation—trillions in GPU, data-centre, and power capacity—along with acute national-security stakes in the US–China competition, inadequate current lab security against state actors, and unresolved challenges in controlling superintelligent systems. The essays became widely read in Silicon Valley and among investors seeking an AI-infrastructure thesis.
3. Formation and Ascent of Situational Awareness LP
Aschenbrenner founded Situational Awareness LP in mid-2024 (seed capital reported around $225 million from Patrick and John Collison of Stripe, Nat Friedman, and Daniel Gross; later capital included Jane Street and others). The fund was explicitly named after the essay series and structured around the infrastructure thesis: long exposure to semiconductors, high-bandwidth memory, data-centre operators, and electricity/power-generation assets, supplemented by private-company stakes (most prominently a large position in Anthropic) and some short positions against software names expected to face disruption.
Early performance was exceptional. Reports indicate gains exceeding 1,000% since inception prior to the July 2026 events, with first-half 2026 returns cited in the region of 439% net. Assets under management expanded rapidly; CNBC sources placed the peak near $45 billion earlier in July 2026, while other contemporaneous accounts referenced figures in the $15–20+ billion range at various points. Roughly two-thirds of the book was in public equities (long and short), the balance in privates. The fund operated with a very small team and employed substantial leverage—contemporaneous reporting frequently cited gross exposure approaching 4×.
Aschenbrenner had no prior professional portfolio-management or risk-management experience. Investors were attracted by the intellectual coherence of the thesis, his perceived proximity to frontier AI developments, and the early returns.
4. The July 2026 Stress and Forced Unwind
In July 2026, AI-infrastructure names experienced sharp declines. Key holdings (including SK Hynix, Micron, SanDisk, CoreWeave, and Nebius Group) fell more than 35% during the month. Simultaneously, short positions (for example against Adobe and other software names) moved adversely as those stocks rallied. Liquidity thinned and positions associated with the fund came under additional pressure—dynamics Aschenbrenner later likened to a bank run (“vulnerability begetting more vulnerability”).
Prime brokers (Bank of America, Goldman Sachs, and JPMorgan) issued margin calls. The fund was forced into rapid reduction of leveraged public positions. Citadel acquired the bulk of the publicly traded equity portfolio in a block/fire-sale transaction. The overall portfolio declined approximately 67% in July. Assets fell from the reported early-July peak toward roughly $10 billion (largely the remaining private holdings, including the Anthropic stake, which was retained). Leverage was removed.
In a letter to limited partners, Aschenbrenner wrote: “We let you down this month” and acknowledged that the firm “came closer to permanent capital impairment than is acceptable to us.” He stated that the underlying fundamentals of the positions remained strong, that the fund was still up around 80% year-to-date 2026, and that he took full responsibility. He indicated he would strengthen portfolio-management and risk teams. The events coincided with the weekend of his wedding to Avital Balwit (then chief of staff to Anthropic CEO Dario Amodei).
5. Structural Causes and Lessons
Several interlocking factors produced the severity of the drawdown:
- Concentration and thematic purity. The book was heavily exposed to a narrow set of AI-infrastructure equities whose prices had already advanced dramatically.
- Leverage. At roughly 4× gross exposure, a 25–35% decline in the long book is mathematically capable of eliminating equity capital before hedges or short-side gains can offset it. Shorts that were intended as hedges instead compounded losses.
- Liquidity and margin dynamics. Forced selling by a large, concentrated, leveraged holder amplified price impact and invited further pressure.
- Experience and organisational scale. Aschenbrenner’s exceptional analytical and forecasting strengths did not equate to institutional risk-management infrastructure, stress-testing against short-term cycles, or the operational capacity typical of multi-billion-dollar funds.
- Timing mismatch. A multi-year (or decade) infrastructure thesis is compatible with severe interim drawdowns; high leverage and public-market liquidity constraints convert those drawdowns into existential threats to the vehicle.
The long-term directional correctness of the AI-infrastructure demand thesis remains an open empirical question; the near-term market cycle and financing structure proved decisive.
6. Aftermath and Assessment
As of early August 2026 the fund continues, de-levered and with a greater relative weight on private holdings. Aschenbrenner’s standing as an AI prognosticator is largely intact among those who shared his scaling views; his credibility as a steward of leveraged public capital has been substantially impaired. Critics (including venture investors such as Bill Gurley in contemporaneous commentary) highlighted the recklessness of the leverage. Supporters note that the fund was still solidly profitable on a year-to-date basis and that the core private positions were preserved.
The episode is a compact case study in the intersection of technological narrative, concentrated capital, and market microstructure. Visionary insight into a secular trend does not automatically confer mastery of portfolio construction, leverage, or liquidity risk. In the AI investment wave of the mid-2020s, Situational Awareness LP became one of the clearest demonstrations of both the rewards and the hazards of that combination.
