Just been diving into Leopold Aschenbrenner's latest portfolio moves and honestly, the guy is playing a completely different game than everyone else in this space.

So most people are still obsessing over nvidia stock like it's the only play in AI. But this 24-year-old just did something wild—he completely exited his nvidia position. Sold off nvidia, Broadcom, TSMC, Micron. All the usual suspects everyone's loading up on. And he made money doing it too, holding put options that paid off as chip stocks cooled.

The question everyone's asking: why would you walk away from nvidia stock when the whole market is still chasing chips?

His answer is actually pretty compelling. He's basically saying the GPU play is done. By end of 2025, early 2026, the market fully priced in what chips could do. The value's already reflected. So if you're still buying nvidia stock now, you're late to the party.

But here's where it gets interesting. He didn't just exit—he pivoted hard into something almost nobody's talking about. Energy infrastructure. Specifically, he's betting that power generation is the real bottleneck, not computing anymore.

Think about it. AI labs have more GPUs than they can use. But they can't power them. The existing grid was built for people, not for training massive language models. That's the constraint now. That's where the real money goes next.

His biggest move? Dumped 20% of his entire fund into Bloom Energy. Eight hundred fifty million dollars into a company most people haven't even heard of. This is a concentrated, high-conviction bet.

Bloom Energy makes these oxide fuel cells that convert natural gas into electricity on-site. Modular, deployable fast, no grid dependency. Their order backlog is sitting at twenty billion dollars. Revenue grew 34% last year, expecting 40% this year. That's not saturation—that's demand crushing supply.

The play is elegant: instead of relying on the power grid like everyone else, you install these fuel cells next to your data center and generate your own power. No waiting for grid upgrades. No competing with other data centers for limited capacity. You solve the energy problem yourself.

Then there's CoreWeave. He threw another three hundred million at them on top of his earlier five hundred million. CoreWeave basically handles the unglamorous stuff—racking GPUs, cooling systems, power delivery, technical support. They're like the infrastructure layer nobody thinks about but everyone needs.

But here's the clever part. He's also buying up Bitcoin mining companies. Sounds random, right? Crypto's not hot right now. But these companies own something valuable: land and power permits. They've already done the hard work of getting grid access, building out facilities, getting all the regulatory approvals.

So instead of waiting months or years to get licenses and grid connections for new data centers, he just acquires these mining companies, strips out the crypto stuff, and repurposes the infrastructure for AI. It's like taking over a bar that already has its liquor license instead of applying for a new one and waiting years. Genius shortcut.

He also shorted Infosys. Why? Because he sees AI agents replacing cheap outsourced labor. These IT outsourcing companies built their whole model on providing cheaper developers in India. But now Claude and GPT can handle complex coding tasks. That business model is becoming obsolete.

What's really striking about this whole approach is the philosophical shift. He went from betting on nvidia stock and chip abundance to betting on scarcity of the physical world. Energy, land, permits, manufacturing capacity. These are things you can't software your way out of.

Companies relying purely on software are going to struggle. The future is physical. Factories, infrastructure, power generation. These need human capital, licenses, legislation. AI can't build those.

Energy is the constraint nobody can escape. Google, Amazon, Nvidia—they just pledged 650 billion in capital expenditures in one earnings season alone. All that money flowing toward solving the power problem. And his portfolio is positioned right in the middle of it.

Look at the bigger picture. He started his fund at end of 2024 with 255 million. Six months later, five hundred million. Now it's at five and a half billion. In just over a year. He's outperformed the S&P 500 by eight times.

Most of that is because he saw the chip thesis playing out, rode it, then had the conviction to flip to infrastructure before everyone else caught on. While most investors are still arguing about nvidia stock, he's already moved to the next constraint.

Is it risky? Absolutely. It's concentrated. It's a single-theme bet on AI infrastructure demand continuing to accelerate. If that slows down or macro changes, everything takes pressure. And he's only 24—no long track record of weathering downturns.

But so far his simple ideas keep being right. Bitcoin mining companies have land and power. Check. Energy is the bottleneck. Check. Bloom Energy can scale fast. Check. AI agents will replace outsourcing. Increasingly looking correct.

The guy who got fired from OpenAI wrote a 165-page thesis predicting AGI by 2027. People thought he was crazy. But his predictions about GPU infrastructure proved right. Now he's making the same kind of contrarian call about energy infrastructure.

Whether you agree with his 2027 AGI timeline or not, the infrastructure thesis is worth paying attention to. He's basically saying the next five-year run isn't about nvidia stock anymore. It's about power generation, real estate, permits, and manufacturing capacity.

If he's right, Bloom Energy and companies like it could see the same kind of explosive growth nvidia stock saw over the last few years. That's the bet he's making with his entire fund.