Just been deep diving into Leopold Aschenbrenner's latest 13F and honestly, the guy's investment thesis is way more interesting than most people realize. Everyone's obsessing over whether he'll hit AGI by 2027, but his actual portfolio moves are telling a completely different story.

So here's what caught my attention: dude completely exited Nvidia. Like, sold his entire position. I know, wild right? When I talk to people on Wall Street, Nvidia is literally the only thing anyone wants to own. But Leopold saw it differently—by end of 2025, he basically called that the GPU trade was already fully priced in. He's not wrong either, looking at how things have played out.

But the real move? He dumped all that capital into something almost nobody's talking about: Bloom Energy. This is his 20% portfolio concentration play—we're talking $855 million into a company most retail investors probably haven't even heard of. Their oxide fuel cells convert natural gas directly into electricity for data centers, completely bypassing the grid. Revenue up 34% in 2025, projecting 40% growth for 2026, with a $20B order backlog sitting there.

Think about it logically: AI labs are drowning in GPUs but starving for power. The grid can't handle it. So instead of betting on chip manufacturers anymore, he's betting that whoever controls power generation controls the entire AI infrastructure play. It's kind of genius in its simplicity, honestly.

He also poured $300M more into CoreWeave—basically the infrastructure layer that makes GPUs actually usable—and here's the part that really shows his thinking: he's been buying up Bitcoin mining companies. Not for Bitcoin. For their land, their electrical permits, their grid access. These licenses take years to get normally. He's just acquiring companies that already have them, stripping out the crypto operations, and converting them into AI data center infrastructure. It's like buying a bar that already has a liquor license instead of waiting three years for your own.

Then he shorted Infosys hard. His thesis: AI coding assistants are now powerful enough to replace cheap outsourced labor. That one's already looking prescient.

The whole shift basically comes down to one observation: energy is the new bottleneck. Not chips anymore. The physical world—manufacturing, real estate, power generation—that's where the actual constraints are. Software's become too easy to build. Hardware's what matters now.

What I find fascinating is how concentrated this is. Most hedge funds would never put 20% into a single holding like Bloom Energy, especially something outside their core competency. But that's exactly why he's up 5.5x in 18 months while everyone else is still arguing about GPU stocks. He saw the next move before the crowd even finished processing the last one.

Could it blow up? Sure. Single-theme portfolio is risky if the macro shifts. But right now, every signal's pointing toward this thesis playing out exactly as he mapped it. Guy's literally just following the bottlenecks in real time and putting money there. Sounds obvious when you say it out loud, which is probably why most people miss it.