The race to build artificial general intelligence (AGI) isn't being won by the companies training the largest models or designing the fastest chips,it's being won by the companies that can keep the lights on. Leopold Aschenbrenner's Situational Awareness LP fund has grown from $225 million to $5.5 billion in just one year by betting on a contrarian thesis: the physical infrastructure powering AI data centers, not the AI companies themselves, represents the defining economic opportunity of the coming decade.

This shift in focus reveals something the broader tech industry has largely overlooked. While investors have poured capital into semiconductor companies like Nvidia and cloud giants like Microsoft and Amazon, Aschenbrenner's fund is deliberately avoiding those crowded trades. Instead, it's concentrating roughly 79 percent of its $5.5 billion portfolio in seven companies that solve what it sees as the true constraint on AGI development: reliable, abundant electrical power.

Why Is Power Suddenly More Valuable Than Computing Power?

The answer lies in a simple but urgent mathematical reality. Data centers required to train and run advanced AI systems consume hundreds of megawatts of electricity each, and the existing power grid cannot keep pace with demand. As AI capabilities scale exponentially with computing power, the bottleneck shifts from algorithms to the physical infrastructure that delivers electricity to the machines doing the computation.

Duke Energy, one of the largest utilities in the United States, has made this calculation explicit. The company unveiled a $103 billion five-year capital plan focused on building 14 gigawatts of new generation capacity, 4.5 gigawatts of battery storage, and exploring small modular reactors to power the data center buildout across its six-state territory. This represents the largest capital commitment any regulated utility has ever made, and it signals that data center electricity demand has moved from forecasting exercises into concrete infrastructure spending.

The scale of this shift is staggering. Data centers drove roughly half of all U.S. electricity demand growth in 2025, according to Duke Energy's investor materials. That single statistic explains why utilities are rewriting their capital plans and why investors are suddenly treating power generation as a scarcer commodity than computing chips.

How Are Hedge Funds and Utilities Positioning for AI's Power Crisis?

• Energy Generation: Bloom Energy, the largest holding in Aschenbrenner's fund at 20.6 percent of the portfolio, provides solid oxide fuel cells that deliver reliable on-site power for data centers, bypassing grid constraints entirely. The stock has risen 150 percent year-to-date.
• Computing Infrastructure: CoreWeave, a cloud GPU infrastructure provider, represents 18.2 percent of the portfolio and addresses the computing power pillar by providing efficient AI training and inference workloads. The fund holds both stock and call options on the company, which has risen 63 percent year-to-date.
• Optical Communications: Lumentum produces laser and photonic equipment that accelerates data transfer within and between data centers, solving the bandwidth constraint as AI clusters grow. The stock has risen 143 percent year-to-date.
• Data Center Operations: Core Scientific has pivoted from cryptocurrency mining to AI infrastructure, building facilities specifically for AI and high-performance computing workloads with backing from JPMorgan Chase and Morgan Stanley.
• Grid Infrastructure and Generation: Duke Energy is allocating roughly $33 billion of its $103 billion plan to grid infrastructure and modernization, with an additional $22 billion for new natural gas generation capacity and $18 billion for renewables and battery storage.

The strategy reflects a second-order beneficiary thesis: companies that supply the buildout will capture more value than those building the models themselves. Aschenbrenner's fund has deliberately exited positions in mainstream AI plays like Nvidia and Broadcom, signaling a judgment that the market has already priced in semiconductor demand while missing the asymmetric opportunity in the physical backbone.

Duke Energy's capital plan reveals how this plays out in practice. Of the $103 billion commitment, approximately 65 percent flows directly to grid infrastructure and new power generation, with the remainder funding nuclear license extensions, environmental compliance, and transmission expansion. This isn't a diversified bet on technology; it's a focused wager that the companies solving the power constraint will command pricing power unlike anything in the tech ecosystem.

What Does Duke Energy's $103 Billion Plan Tell Us About AI's Future?

Duke Energy's announcement in April 2026 reframed its infrastructure commitment around a single word: affordability. This pivot matters because it signals that Duke expects political turbulence as state utility commissions grapple with the cost of powering the AI buildout. The company is deploying more than $1 billion in capital every month, with acceleration expected in 2027 and 2028 as data centers connect to the grid.

The utility's capex allocation reveals a heavy tilt toward generation and transmission rather than customer-side technology. New natural gas combined-cycle plants account for 5 gigawatts of the 14-gigawatt generation surge, a choice that has drawn criticism from environmental groups but which Duke argues is necessary to reliably integrate the massive data center loads coming online. Without firming generation that can ramp on demand, the company contends, the regional grid cannot serve loads equivalent to the entire annual peak demand of mid-sized American cities while maintaining the reserve margins required by reliability standards.

One of the most consequential elements of Duke's plan is buried in a single line item: an early site permit application filed in December 2025 for a small modular reactor at the Belews Creek Steam Station in North Carolina. The existing coal plant site already has interconnection rights, water rights, and a trained workforce, making it an attractive location for next-generation nuclear capacity. This pivot signals that utilities are beginning to view small modular reactors not as speculative technology but as essential infrastructure for the AI era.

"The largest growth opportunity in our history" requires "balancing" with "the obligation to keep bills as flat as we can," stated Harry Sideris, CEO of Duke Energy, in comments to Fortune.

Harry Sideris, CEO at Duke Energy

The timing of Duke's announcement matters. It followed reporting that data centers drove roughly half of all U.S. electricity demand growth in 2025, a statistic that has reshaped how every regional transmission organization plans capacity. The interconnection queues managed by PJM, MISO, and SERC have ballooned past 2,000 gigawatts of pending requests, with Duke's territory being the single largest source of new connection requests in the SERC footprint.

For investors and infrastructure companies, the message is clear: the window to position for AGI's infrastructure demand is narrow because the scaling timeline is aggressive. Aschenbrenner's thesis predicts AGI by 2027, meaning the infrastructure buildout must precede that milestone, not follow it. The companies that solve the power constraint will capture the scarcity value that comes from being the limiting factor in the most important technological transition of the century.