Just caught myself down a rabbit hole on this 24-year-old Leopold Aschenbrenner guy everyone's been talking about. His fund went from $1B to $5.5B in about a year, which is absolutely wild. But what's more interesting than the numbers is what he actually did to get there.

So here's the thing that caught me: he basically exited his entire Nvidia position. Sold Broadcom, TSMC, Micron—basically cleared out all the major chip plays. When I first saw that, my reaction was the same as everyone else's: why would you walk away from Nvidia when the whole market is obsessed with it? But his logic actually makes sense. He published this 165-page thesis called "Situational Awareness" where he basically argues that by end of 2025 or early 2026, GPU value would be fully priced in. Looking at nvidia market cap and where chips are trading now, he might've been onto something.

What he pivoted to is way more interesting though. His largest position is now Bloom Energy—20% of the entire portfolio, around $855M. Never heard of them before? Yeah, neither had most people. They make these oxide fuel cells that convert natural gas directly into electricity for data centers. No grid dependency. Their order backlog is sitting at $20B, revenue grew 34% last year, and they're projecting 40% growth for 2026. That's the kind of growth trajectory you normally see in speculative tech stocks, except this is actual physical infrastructure.

The move makes sense once you think about it: AI labs are drowning in GPUs but starving for power. The existing grid was built for humans, not for training massive models. So instead of betting on who makes the chips, he's betting on who powers them. That's the real bottleneck now.

He also went hard on CoreWeave, pumping another $300M into their AI cloud services on top of his earlier $500M position. And here's the clever part—he started buying up Bitcoin mining companies. Sounds random until you realize these companies already have land, permits, and grid access. Took him months to acquire what would take years to build from scratch. It's like buying a bar that already has a liquor license instead of waiting years for one yourself.

Even shorted Infosys, betting that AI coding agents will replace cheap outsourced labor. That one feels like it's already playing out.

The whole thesis basically boils down to: software is easy now, so the real money is moving to the physical world. Energy, manufacturing, infrastructure, permits—the stuff AI can't just code into existence. The fund's concentrated bets on energy infrastructure and GPU hosting look directional if you buy into where the actual constraints are.

Worth keeping an eye on, especially if you're thinking about where infrastructure spending really needs to flow this year.