A 24-year-old fund manager's annual return is 24 times higher! His AI investment portfolio targets the "most scarce resource"

Previously, former OpenAI researcher Leopold Aschenbrenner doubled $225 million to $5.5 billion through his fund. He said the bottleneck in AI development lies in electricity, not chips or models.

Previously, a former OpenAI security researcher and a German man just 24 years old, Leopold Aschenbrenner managed to turn $225 million of the fund Situational Awareness he oversaw into $5.5 billion in less than a year. While Wall Street money was flooding into AI models and chipmakers, he spotted a blind spot that the market was ignoring: electricity. By precisely betting on the infrastructure to solve AI’s power-hungry problem, he created astonishing investment returns.

After leaving OpenAI, he turned around and threw himself into the AI investment market

After leaving OpenAI, Aschenbrenner wrote a 165-page report, claiming that general artificial intelligence (AGI) would arrive faster than everyone thinks, and that the final winners would not be companies with the strongest AI models, but rather companies that “control electricity.” For this, he set up a hedge fund called “Situational Awareness LP,” and poured $875 million into buying shares of fuel cell company Bloom Energy.

This week, Bloom Energy announced that it had signed a major fuel cell order of 2.8 billion watts (GW) with Oracle, causing the stock price to surge 15% after-hours, and the book value of Aschenbrenner’s holdings also instantly jumped to nearly $2 billion.

Portfolio revealed: go long on infrastructure, short traditional IT industries

Reports say his investments all follow the logic of “electricity first”:

• Bloom Energy (BE): Invested $875 million to buy this fuel cell company. The technology enables data centers to generate power on-site directly, without relying on an outdated power grid. Benefiting from the 2.8 GW deal signed with Oracle, the stock surged, and the book value of his holdings has already climbed to nearly $2 billion.
• CoreWeave (CRWV): Invested $700 million in this top AI cloud computing infrastructure provider, locking in scarce infrastructure resources.
• Infosys (INFY): Aggressively shorted this Indian IT outsourcing giant; he expects AI coding agents will completely wipe out traditional IT outsourcing businesses.
• Intel (INTC): Used Intel call options to leverage his position, earning several times more returns as the stock rebounded strongly by 53%.
• Core Scientific (CORZ): Holds a 10% stake. This former bitcoin mining company is pivoting to convert its existing power facilities into AI data center colocation sites.

The power-hungry monster behind computing power: electricity use doubles year after year

Aschenbrenner said that looking back to 2022, the compute cluster used to train GPT-4 consumed about 10 megawatts (MW) of electricity, costing about $500 million. However, AI compute demand is expanding at roughly a half-order of magnitude every year, which means the power demand of the largest training clusters will double every 12 to 18 months.

By 2024, the power consumption of the largest compute clusters has reached 100 MW, equivalent to 100,000 high-end graphics processing units (GPUs) running simultaneously. Now in 2026, the leading training clusters require as much as 1 GW of continuous power, equivalent to the power generation output of a large nuclear reactor.

Whoever controls electricity controls the future of AI

He estimates that by 2028, power consumption for AI training will soar to 10 GW—bigger than the electricity generation output of entire many U.S. states. By 2030, it will reach 100 GW, consuming as much as 20% of the United States’ current total electricity generation at once. This is only the electricity used for “training” models; if you add the “inference” compute that is actually used by the public, the electricity consumption becomes even harder to imagine.

However, over the past decade, the United States’ total electricity generation has grown only slightly by 5%. Now reports from across the country about major shortages of transformers and data centers that can’t get built are proof that the power grid can’t support the demand. That’s also why he dares to heavily bet on Bloom Energy: the real bottleneck for AI development isn’t chips or software—it’s whether human beings can produce enough electricity.

• This article is reproduced with authorization from: 《Chain News》
• Original title: 《24-fold annual return! A 24-year-old fund manager’s AI portfolio targets the “most scarce resource”》
• Original author: Co2