방대한 AI 데이터센터의 전력 수요가 150억 달러 규모의 PJM 용량 경매를 주도하고 있으며, 원자력 소형모듈원자로(SMR) 도입을 가속화하고 있다.

The rapid buildout of artificial intelligence infrastructure across the United States is straining the nation’s physical grid, which remains unprepared for the scale of development regardless of semiconductor advancements. Washington has effectively determined who will bear the cost of this expansion, forcing hyperscale developers to acknowledge their responsibility toward national infrastructure needs. Consequently, regional electricity markets, next-generation nuclear financing, and broader energy procurement strategies have all shifted, transforming energy taxes into major line items for the world’s largest technology firms. The epicenter of this shift is the PJM Interconnection Service Area, the largest regional grid operator in the U.S., spanning from Chicago to Washington, D.C. Over the past three years, surging data center development and rising user demand have driven a 38% increase in residential energy costs across the entire PJM footprint, including a 13% jump between 2020 and 2025 alone.

These pressures prompted intense scrutiny from the White House and state governments, culminating in an emergency electricity auction in January designed to compel major technology companies to underwrite new capacity supply. On March 4, seven leading hyperscale and artificial intelligence firms—Amazon (AMZN), Google (GOOGL), Meta (META), Microsoft (MSFT), OpenAI, Oracle (ORCL), and xAI—signed the Voluntary Taxpayer Protection Pledge. While not legally binding itself, the pledge establishes legally enforceable financial structures under U.S. law. Signatories agree to negotiate rate structures dedicated specifically to data center loads, ensuring operators cover all grid upgrade and new generation costs without passing them onto residential consumers or non-tech businesses. Crucially, data centers will pay for contracted capacity through PJM regardless of actual electricity consumption. PJM projects a 5.2% capacity shortfall by 2027–2028, necessitating approximately $15 billion in new plant investments. The emergency auction must conclude before September 2026, with long-term power purchase agreements expected to formally end the implicit public subsidy of AI infrastructure.

This auction has fundamentally elevated power procurement from an operational detail to a balance-sheet item capable of delaying project development. “As an example of this new burden,” wrote Wedbush analysts in a market note, “Financing methodologies used to build data centres now may consider to be reflected in the additional costs associated with modifying the grid, such as on transmission lines, re-decking substations and installing new connections for data centre projects that would have historically been included in a utility's rate base.” As a result, data center electricity costs could rise by 30% to 50% over 15-year contracts, expenses that will ultimately be passed through as higher prices for cloud and AI services. Time is critical. At GTC 2026, Nvidia CEO Jensen Huang identified power delivery—not silicon—as the primary bottleneck for chips, infrastructure, models, and applications, noting that land, electricity, and shell capacity are equally constrained. To address transmission gaps, the U.S. Department of Energy’s SPARK program aims to expand capacity by at least 50% through reconductoring and advanced technologies. However, final program funding will not arrive until late 2026 and early 2027, with core transmission projects typically requiring three to five years to construct. Without reliable new generation, the DOE warns that power outage frequency could increase 100-fold by 2030.

Facing grid inefficiencies, high auction costs, and tight timelines, hyperscalers have accelerated their response by securing energy supply directly on their own balance sheets—a strategic pivot that serves as the strongest catalyst for today’s nuclear renaissance. Rather than relying on competitive auctions, major AI players are writing direct checks to reactor developers, transforming small modular reactors (SMRs) from unregulated concepts into fully funded commercial pipelines. Meta has partnered with TerraPower to fund the development of two Natrium units capable of generating up to 690 MW of firm power, with delivery as early as 2032. Separately, Meta’s collaboration with Oklo advances an advanced nuclear campus in Pike County, Ohio, potentially online by 2030 and poised to deliver up to 1.2 GW of clean baseload power directly into the PJM market. Amazon and X-energy target bringing more than 5 GW online by 2039, marking the largest commercial SMR deployment goal to date. Meanwhile, Microsoft has layered nuclear agreements atop existing combined-cycle gas and renewable contracts. Collectively, these four companies have secured or will secure more than 10 GW of nuclear capacity to support operations by 2035, with additional announcements ongoing.

Beyond megawatt targets, corporate balance sheet involvement significantly alters the sector’s financial landscape. “Bringing a top tier corporate balance sheet into an industry that has historically been funded through a regulated, ratepayer backed revenue stream means a lot for the overall credit profile of that sector,” said Shioly Dong, a senior analyst at BMI. This improved credit profile is essential for attracting commercial lenders and institutional investors. “The modular design concepts have a smaller scale and shorter timeframe for construction; therefore, they have a significantly lower level of upfront capital risk,” noted Tim Winter, portfolio manager for the Gabelli Utilities Fund (GABUX). HSBC analysts added that the technology sector’s willingness to absorb cost overruns or schedule delays will be decisive in determining how quickly these agreements accelerate advanced reactor deployment, fuel supply chain expansion, and technology park development—projects previously considered unfinanceable under conventional nuclear models. Finally, while neither nuclear nor solar can rapidly ramp output to match AI workloads’ continuous service requirements, long-duration energy storage has become indispensable. Specifically, liquid-flow battery systems will likely operate alongside renewable generation and nuclear plants due to their favorable safety profiles and ability to stabilize intermittent supply.