Just spent the last few hours digging into Leopold Aschenbrenner's latest portfolio moves, and honestly? This 24-year-old's thesis makes way more sense than most people realize.

So here's what caught my attention. His fund went from $1B to $5.5B in roughly a year, right? But the interesting part isn't that he made money—it's what he actually sold and why. He completely exited his nvidia stock position. Sold NVIDIA, Broadcom, TSMC, Micron. All the usual AI infrastructure plays that everyone's been loading up on.

His reasoning is deceptively simple: by end of 2025, the market had already priced in the GPU value story. The chip narrative was done. So instead of sitting in nvidia stock like everyone else, he pivoted hard into something almost nobody was watching—energy infrastructure.

Think about it. AI labs have more GPUs than they know what to do with. But they can't actually power them because the grid was built for humans, not for 50-megawatt data centers. That's the real bottleneck now.

His biggest bet? Bloom Energy. 20% of his entire portfolio, about $855M. Most people had never even heard of this company three months ago. They make oxide fuel cells that convert natural gas directly into electricity for data centers. No grid dependency. Modular deployment. $20B order backlog. Revenue up 34% in 2025, projecting 40% growth for 2026.

Then there's CoreWeave—another $300M on top of his existing position. They're basically the infrastructure layer that actually deploys and maintains all those GPUs. Added another $800M-ish across multiple positions.

But here's where it gets clever. He also started buying up Bitcoin mining companies. Seems random, right? Except these companies already have something that takes years to acquire: land, power permits, grid access. He's essentially using them as infrastructure acquisition vehicles, stripping out the crypto element and repurposing everything for AI data centers. It's like buying a bar just for the liquor license.

He's also shorting Infosys because he believes AI agents will replace the entire outsourced labor model. That's a pretty bold call, but given what we're seeing with Claude and GPT, it's not crazy.

The whole philosophy shift is fascinating: from software to atoms. From chips to power generation. From betting on what companies build to betting on what they need to operate.

Energy is the only resource nobody has enough of right now. Every major tech company is suddenly desperate for it. Google, Amazon, Microsoft—they're all throwing billions at data center infrastructure. And this guy positioned his fund right at that intersection months ago.

Whether his AGI by 2027 prediction comes true or not, his infrastructure thesis is tracking real constraints in the market. He's essentially building a portfolio around the actual bottleneck, not the narrative everyone's still talking about.

Worth watching where this goes. The simplicity of the logic—identify the constraint, invest in the solution—is almost refreshing compared to most hedge fund positioning.