Are you curious about the secret behind the $5.5 billion fund created by a 24-year-old investor? Over the past few months, Leopold Aschenbrenner’s portfolio changes have been truly fascinating. The days when he was betting heavily on Nvidia’s stock are already behind him.

Last year, he focused on buying up the beneficiaries of the GPU boom. Companies like Nvidia, Broadcom, TSMC, and Micron, the chip manufacturers. But looking at his recent 13F filings, the situation has completely shifted. He’s almost completely divested from those stocks. His judgment is that Nvidia’s stock price has already been fully reflected in the market.

So where did that money go? He shifted his focus to energy and infrastructure. In his 165-page paper titled “Situational Awareness,” he predicted reaching AGI by 2027, but now he seems to have identified the next bottleneck precisely.

His current largest holding is Bloom Energy. It’s an $855 million investment, making up 20% of his portfolio. What does this company do? It makes fuel cells that convert natural gas directly into electricity usable in data centers. The key point is that it eliminates the need to rely on the existing power grid. Their order backlog alone is $20 billion, with revenues expected to grow 34% in 2025 and another 40% in 2026, indicating demand far exceeds supply.

Why the sudden focus on energy? Because the current power grid is designed for human energy needs, not the enormous power demands of AI data centers. The recent quarterly reports from Google, Meta, and Amazon, promising $650 billion in capital expenditures, are ultimately driven by this issue.

Investing $800 million in CorWeave is also in this context. Buying GPUs alone isn’t enough. Installing them in racks, powering them, and maintaining cooling systems are entirely different challenges. That’s why he invested in cloud service providers specializing in AI infrastructure.

The most creative move is his investment in Bitcoin mining companies. It may seem odd on the surface, but if you think about what they own—land and power, along with already secured permits and grid access—it's clear. Usually, obtaining such permits takes months or years. So he’s acquiring companies that already have permits and converting them into AI data centers. It’s like buying an already licensed bar instead of applying for a new liquor license and waiting.

His short selling of Infosys is also intriguing. This Indian company has relied on cheap labor for IT outsourcing, but now models like Claude or GPT have become powerful enough to handle not just simple tasks but critical IT processes. He believes their business model is about to collapse.

Ultimately, his investment philosophy captures a major shift from software to hardware. Software can now be created by anyone, but energy, manufacturing, permits, and infrastructure cannot be built by AI. And all of this converges into one question: who can most efficiently supply the energy needed for the future? His conclusion is that semiconductor companies like Nvidia, whose stock prices have already risen significantly, are no longer the best bets.

Looking at how he turned $1 billion into $5.5 billion in a year and a half, it’s clear how accurate his analysis has been. Of course, this isn’t investment advice; it’s just one investor’s portfolio. But it’s worth paying attention to how the energy and AI infrastructure sectors might move this year.