You ever notice how everyone tries to copy the winning portfolio the moment it gets published? Well, there's this 24-year-old German guy named Leopold Aschenbrenner who's basically broken that game. His fund is up 61% in two months, and people are going absolutely nuts trying to figure out his next move.

The wild part? He's not buying Nvidia. Not buying OpenAI. Not touching any of the obvious AI plays everyone else is chasing. Instead, Leopold Aschenbrenner is betting everything on the stuff AI actually needs to exist: power generation, chip manufacturing, data centers. The whole thesis is simple - the real bottleneck isn't algorithms, it's electricity and computing power.

His two biggest wins show how this plays out. Bloom Energy, a fuel cell company providing off-grid power to AI data centers, jumped 239% year-to-date. His position there went from $875 million to nearly $3 billion. Then Intel - he bought 20 million call options when the stock was around $20 and everyone on Wall Street was bearish. Intel hit $113 last week, a 25-year high. That's roughly a 5x return on the options alone.

Obviously, Reddit and investment blogs immediately started asking the same question: which stock is he buying next? The Motley Fool published four articles in one day breaking down his holdings. But here's what everyone misses - position reports have a 45-day delay. By the time you see what he bought, the market already moved halfway through. More importantly, you can't actually replicate why he keeps getting it right.

The real edge isn't the thesis. Plenty of people figured out AI needs more power. The edge is the circle. Leopold Aschenbrenner spent a year at OpenAI's Superalignment team working directly under Ilya Sutskever. He saw the actual roadmaps, the real power consumption numbers, the specific chip requirements for next-gen models. When he wrote about gigawatt-level power demands in his paper, that wasn't speculation - that was based on internal lab knowledge.

His LPs are basically Silicon Valley's infrastructure gatekeepers. The Stripe founders have direct visibility into which companies are signing massive power contracts. Nat Friedman, former GitHub CEO and now Meta AI's product lead, is literally involved in computing power procurement decisions every single day. His research director, Carl Shulman, spent years at Peter Thiel's hedge fund translating AI insights into actual trading strategies.

There's also the detail that his fiancée is chief of staff to Dario Amodei at Anthropic. So Leopold Aschenbrenner has practical experience at OpenAI and regular contact with Anthropic's leadership. He's one of the few people with deep connections to both sides of the AGI race.

Even weirder - he held positions in CleanSpark and Bitfarms, both Bitcoin mining companies converting their operations into AI computing centers. Turns out he spent nine months at FTX's charitable foundation before it collapsed. He's basically one of the few people who understands both crypto infrastructure and cutting-edge AI labs. That intersection alone is worth something.

The structure of his fund is telling too. He explicitly rejected the VC route because he wanted to bet on established companies with existing physical infrastructure - fuel cells, chip manufacturers, mining farms. These have been public for years but analysts are still using old valuation models. They haven't seriously priced in the variable of "essential AI infrastructure." That's the arbitrage.

But here's the thing that actually matters - this creates a self-reinforcing loop. Better returns attract more industry insiders as LPs. More LPs means access to more concentrated information from decision-makers. Better information means more accurate bets. Higher returns attract even more connected LPs. For outsiders, the barrier to entry only gets higher.

Everyone sees his portfolio and wants to copy it. But you're not copying a position, you're copying a moment in time when someone had information you didn't have. His papers are public. His holdings reports are public. His investment logic is explained in podcasts. But even if you understand every single judgment perfectly, you can't replicate the position he was in when he made those judgments. Knowledge can't be shared the same way positions can be traced. That's the real asymmetry. Leopold Aschenbrenner's returns aren't about being smarter - they're about being in the right circle at the right time. And that's the thing you actually can't buy.