Author: Curry , Deep Tide TechFlow

Whenever someone makes a fortune in the US stock market, the first thing onlookers always does is the same: check their portfolio report and find the next stock to buy.

The report that has been most frequently reviewed recently belongs to a 24-year-old German named Leopold Aschenbrenner.

In March of this year, domestic media outlets gave him extensive coverage, with similar headlines; for example, the genius fired from OpenAI who wrote a 165-page paper predicting AI trends, started a hedge fund managing $5.5 billion...

But labels are just labels. What truly makes this fund remarkable is that it doesn't buy Nvidia, doesn't buy OpenAI, and doesn't buy any companies that make AI models. It only buys things that AI cannot function without: power generation, chip manufacturing, optical communications, data centers...

In his own words from his paper, the bottleneck of AI is not in algorithms, but in electricity and computing power. The entire fund is betting that this statement is true.

A social media investment blogger calls himself "the son of the US stock market in the AI ​​era," or "the AI ​​version of Warren Buffett." This title has recently resurfaced because his predictions have become increasingly accurate.

According to data released by the copy trading platform Autopilot on May 1, his simulated portfolio rose by 61% in two months. Based on this, his fund size is approaching $9 billion.

Where did the money come from? Mainly from two heavily invested stocks. Bloom Energy, a fuel cell company that provides off-grid power to AI data centers, has seen its stock price rise by 239% year-to-date.

According to a holdings report released at the end of last year, he held $875 million worth of stock and options in the company, which has now ballooned to nearly $3 billion.

And then there's Intel. The same holdings report shows that he bought 20.2 million Intel call options in the first quarter of 2025, when Intel's stock price was around $20, and the mainstream opinion on Wall Street was that Intel was not doing well.

Last week, Intel shares rose to $113, a 25-year high. In less than a year, it has nearly quintupled, making the young company's option returns far more dramatic than the stock returns.

I can understand the impulsiveness of onlookers. The American investment website Motley Fool published four articles in one day dissecting his holdings, and the overseas Reddit investment forum was discussing whether to copy his strategies. Everyone was trying to find the next Intel from his portfolio report.

However, you should know that position reports generally have a 45-day delay. By the time you see what he bought, the market has already moved halfway through.

More importantly, even if you know his holdings in real time, you can't replicate why he keeps betting correctly.

Circles are the greatest Alphas.

First and foremost, what makes Leopold Aschenbrenner so remarkable is his 24-year-old paper on AI, which almost perfectly predicted the current direction of AI development and investment trends.

The core argument can be summarized in one sentence: the computing power for training AI models is increasing by about half an order of magnitude every year. At this rate, artificial general intelligence (AGI) with capabilities close to those of humans will emerge around 2027.

However, maintaining this growth rate hinges not on the algorithm itself, but on electricity, chip production capacity, and physical space. The power consumption of a single training cluster will jump from megawatts to gigawatts, approaching the output of a large nuclear power plant.

This is the underlying logic of his entire fund. The speed of AI development is determined by physical bottlenecks, so you should invest in the bottlenecks themselves.

This judgment sounds like a conclusion drawn by a smart person after doing a lot of research in their study; but in reality, I think it was the circle of people he formed that led him to this judgment.

Before writing his paper, he spent a year working on OpenAI's Superalignment team. This team specializes in controlling AI that is smarter than humans and reports directly to the lead scientist, Ilya Sutskever.

During that year, he witnessed the internal training plans, the actual computing power consumption, and the specific power and chip requirements of the next-generation models. When he wrote the judgment of "gigawatt-level power consumption" in the paper, it was probably based on the internal roadmap in the laboratory.

He was fired from OpenAI in April 2024, triggered by an internal memo he wrote to the OpenAI board warning of inadequate security measures and the potential risk of infiltration by foreign intelligence agencies.

This memo sparked tensions between management and the board, and OpenAI subsequently fired him for "leaking information."

Two months later, the paper was published. Rather than being an independent study, this paper should be understood as a public version of his internal understanding of OpenAI.

AI research papers have solved the problem of "which direction to look at." But when it comes to investing, simply knowing the direction is far from enough.

AI needs more electricity; many analysts were already saying this in 2024. What's truly valuable is timing and position sizing—for example, would you dare to invest 20 million call options when Intel's stock was at $20?

This confidence stems not only from believing in the general trend of AI, but also from knowing specifically which companies are signing large power purchase contracts, which data centers are expanding, and the actual scale of demand.

The investors in Situational Awareness, the fund founded by Leopold Aschenbrenner, happen to be sitting in the front row of these decisions.

The fund's LPs include the two founders of Stripe, a company that handles payment transactions for most of Silicon Valley's tech companies and has direct access to the accelerated infrastructure spending.

Another investor is Nat Friedman, the former CEO of GitHub and current product manager of Meta AI, who is involved in the decision-making process for computing power procurement every day.

In addition to initial capital, they bring to the fund a continuously updated information pipeline.

In addition, his fund's research director is also a key figure in this chain. Carl Shulman, a veteran in the field of AI security, previously worked at Peter Thiel's hedge fund Clarium Capital, where he was responsible for translating AI insights into actionable trading strategies.

There's another easily overlooked, encrypted corner in his portfolio.

His holdings report at the end of last year showed that he had established new positions in CleanSpark and Bitfarms, both of which are Bitcoin mining companies that are transforming their BTC mining facilities into AI computing centers.

Crypto mining farms naturally possess large-scale power access and cooling systems, which happen to be the scarcest resources for AI data centers.

Interestingly, he is no stranger to the crypto industry. In 2022, he worked for nine months at Future Fund, the FTX charitable foundation founded by SBF, and left just before FTX collapsed.

Whether this experience directly influenced his judgment of mining companies is unknown to outsiders. However, it is certain that he is one of the very few people who have had in-depth contact with both the crypto industry and cutting-edge AI labs. This intersection itself represents a rare opportunity for both intellectual engagement and networking.

Another detail is that his fiancée, Avital Balwit, is the chief of staff to Anthropic CEO Dario Amodei. Anthropic is Claude's parent company and OpenAI's most direct competitor.

He worked at OpenAI, and his fiancée works alongside the CEO of Anthropic. He has practical experience at one of the two leading companies in the AGI competition, and regular contact with the other.

Last year, Fortune magazine interviewed more than a dozen people in the industry who had contact with him, and the conclusion was that he was very good at "packaging ideas that were brewing in Silicon Valley labs into narratives".

I think that's too polite. What he did was more direct: he was betting on the public market the knowledge he gained in his private circle. The published AI papers were declassified versions; his investment fund was the full version.

A positive feedback loop that outsiders cannot enter

Looking back, Leopold Aschenbrenner's fund opted for a less common structure.

Most AI funding follows a venture capital route, investing in early-stage companies and betting on who will become the next OpenAI. He didn't take that path. According to Fortune, he explicitly rejected the VC model when he founded his fund, arguing that AGI's influence was too great and that investment decisions could only be fully expressed in the most liquid public markets.

This choice itself reveals a consensus in his circle: the biggest investment opportunities in the AI ​​era may lie in established companies that already possess physical infrastructure.

This could be a fuel cell company with readily available power access, a chip giant with wafer foundry production lines, or a Bitcoin mining company with mining farms and cooling systems. These companies have been listed for years and have good liquidity, but most analysts are still using old valuation frameworks to price them and haven't seriously incorporated the variable of "essential AI infrastructure needs" into their models.

This is his arbitrage opportunity.

Those in the industry already know the pace and scale of AI infrastructure expansion, but the public market is still pricing based on outdated logic. The price difference is where the profits come from.

This information advantage has another characteristic: it is self-reinforcing.

The better the fund's returns, the more people at the core of the industry will be willing to become limited partners (LPs). The more LPs there are, the more concentrated the information the fund has access to from decision-makers. The more concentrated the information, the higher the accuracy of the bets. This is a positive feedback loop, and for outsiders, the barrier to entry into this loop will only become higher and higher.

Of course, this cycle also has its vulnerabilities. Highly concentrated holdings coupled with significant leverage mean that the entire fund is extremely reliant on a single narrative. As long as the premise of "continuous expansion of AI infrastructure" holds true, everything goes smoothly.

However, if the pace of AI development slows down, or if the energy bottleneck is circumvented by a technological breakthrough, the pullback of concentrated positions will be much faster than the speed of position building. He's betting not only on the direction, but also on the timing. Once the timing is off, the consensus within the circle may become a collective blind spot.

Let's go back to the original question.

Everyone is studying his holdings, trying to replicate his operations. But behind his god-like returns lie structural conditions.

His papers are public, his portfolio reports are public, and his investment logic is clearly explained in podcasts and interviews. But even if you fully understand every single one of his judgments, you cannot replicate the position he was in when he made those judgments.

Positions can be traced back, and returns are enviable, but the source of knowledge cannot be shared. This is perhaps the most expensive asymmetry of our time.