Leopold Aschenbrenner’s 24 Years

1/ Entered Columbia at 15, became class speaker at 19, managed ~$20 billion at 24.

The fund netted +439% in the first half of the year, then crashed 67% in July, ultimately selling most of its stock portfolio to Citadel.

Leopold Aschenbrenner’s life reads like AI mania and leverage risk compressed into 24 years. 👇 https://t.co/lG1vNTy7kv

2/ He was born in Germany to two physician parents.

At 15, he came to the U.S. alone to attend Columbia, majoring in economics and mathematical statistics, graduating as class speaker at 19.

During his time at school, he co-founded an Effective Altruism club, later joined Oxford’s Global Priorities Institute, and briefly worked at the FTX Future Fund.

3/ In 2023, he joined OpenAI’s Superalignment team, working alongside Ilya Sutskever and Jan Leike.

He submitted an internal memo warning that model secrets could be stolen by industrial espionage.

In April 2024, he was fired for allegedly leaking internal information externally. He denied any wrongdoing and believed the real reason was connected to his safety warnings.

4/ Two months later, he published a 165-page essay, Situational Awareness, predicting AGI would arrive around 2027, followed by a rapid transition to superintelligence.

The essay went viral and effectively became a public investment thesis:

If AI continues to scale, the truly scarce resources will be chips, memory, data centers, and electricity.

5/ In late 2024, he launched a fund under his own name, with early backers including Stripe founders the Collison brothers, Nat Friedman, and Daniel Gross.

The fund concentrated its bets on AI infrastructure and used leverage to amplify positions.

By the end of June 2026, year-to-date net returns had reached 439%, with assets under management briefly exceeding $20 billion.

6/ But in July, the trades suddenly reversed.

AI hardware stocks plunged while some software shorts rose simultaneously, and the fund lost roughly 67% in a single month. Faced with a choice between raising new capital and selling positions, it ultimately chose to deleverage, selling most of its public stock portfolio to Citadel.

According to Reuters, it remains unclear whether the fund formally faced a margin call.

7/ He may have been right about what AI would need in the coming years, but he couldn’t accurately predict how the market would move in that particular month.

Long-term logic can wait. Leverage cannot.

From entering Columbia at 15, to building a $20 billion fund off the back of a single essay, to being forced into a major retreat within a month—do you see this as a genius gamble, or a loss of risk control?

Original text:

That’s all.

If you enjoy this topic:

1. Follow me (@FinanceYF5)
2. Like + repost the first post below

Claude Accidentally Wandered Onto the Real Internet During Safety Evaluations

1/ Anthropic just disclosed a rare incident:

After a team reviewed 141,006 cybersecurity evaluations, they found that Claude had accidentally accessed the real internet during 6 runs and made unauthorized entry into 3 companies’ production systems.

This was not a simulation result—it was a real attack that actually occurred.