A 24-Year-Old Is Betting $5.5 Billion on Electricity — Inside Aschenbrenner's AI Infrastructure Fund

A 24-Year-Old Is Betting $5.5 Billion on Electricity. Here's the Logic.

You probably haven't heard of Leo Aschenbrenner. Twenty-four years old, four years out of Columbia, formerly part of OpenAI's early team. His fund, Situational Awareness LP, was already running over $1.5 billion by 2025. The latest filings show roughly $5.5 billion in US equity exposure across about 30 holdings.

The size alone is worth talking about, but what made me pay attention is that he has put nearly his entire personal net worth into the fund. That kind of skin in the game is unusual at any age, let alone twenty-four.

The Picks-and-Shovels Trade, Two Layers Deeper

Look at the 13Fs of most large investors and you keep seeing the same five names: Microsoft, Nvidia, Meta, Amazon, Google. That's the safest way to bet on AI, but those are companies in the $2 to $4 trillion range. The math doesn't allow for 5x or 10x returns from here.

Aschenbrenner's answer is a sharper version of "buy the picks and shovels." Most picks-and-shovels framing stops at semiconductor equipment, packaging, and optics. He goes one layer underneath that. Electricity itself, and the data center capacity that consumes it.

Breaking Down the Book

Power producers. Vistra and Constellation Energy. Both run nuclear and natural gas generation in the US. Both have been re-rated heavily over the past 18 months, driven specifically by data center power purchase agreements.

Fuel cells. Bloom Energy is the fund's single largest position. Bloom builds solid oxide fuel cells that turn natural gas into electricity, deployable in modular installations next to data centers. With grid interconnection wait times stretching past five years in many regions, the value of "I can give you power now" has gone vertical.

Semiconductors. Intel, Broadcom, and the VanEck Semiconductor ETF (SMH). Notice Nvidia is not the largest position. Aschenbrenner is taking sector exposure rather than doubling down on the most expensive single name in the cohort.

Crypto miners pivoting to AI hosting. Core Scientific, Iren, Applied Digital, and Cipher Mining. These are companies that originally built high-density compute facilities for Bitcoin mining and are now repurposing them for AI workloads.

This last category is the central thesis. A Bitcoin mining facility owns two things — long-term power contracts and a physical site that can absorb that power. In the AI compute era, both of those have suddenly become some of the most expensive assets you can own. The market is repricing these companies based on AI hosting revenue, not hash rate.

AI cloud. CoreWeave. An Nvidia-GPU-based AI-specific cloud infrastructure provider, with multi-year Microsoft contracts forming a meaningful share of revenue.

Why "Situational Awareness"?

The fund name comes from a 165-page monograph Aschenbrenner published while at OpenAI. Its central claim is that only a few hundred people in the world — most of them inside San Francisco's AI labs — actually understand what's coming next, and the rest of us have, in his phrasing, "not the faintest glimmer."

Counting himself among that few hundred is self-promotional. But his fund operations back the framing. To make Bloom Energy your top position out of 30 names, you need very strong conviction that AI compute demand will outrun the US grid for the next five to ten years.

The Risks Worth Naming

The thesis is compelling, but I want to be honest about what could go wrong.

First, the Bitcoin-miner-to-AI-hosting pivot is unproven. Repurposing a mining facility for AI training requires very different cooling, network topology, and reliability profiles. There's no guarantee this transition is smooth or fully economic.

Second, sizing a small-cap like Bloom Energy as your number one position carries real liquidity risk. When markets sell off fast, the exit door is narrow.

Third, the demand estimate could be too high. If the Big Tech four guide capex lower for even one quarter, the entire power-and-fuel-cell complex could correct together.

For my own framing, I agree with Aschenbrenner's underlying logic. AI is fundamentally a fight over electricity, data center capacity, and chips. But I'd warn against copying his 30-name list directly. For most retail investors, exposure to the theme through the larger power producers and a diversified semiconductor ETF is the more risk-adjusted entry point.