Version 7

AI electricity demand is reshaping US power infrastructure across three simultaneous fronts: a proposed $67 billion NextEra–Dominion utility merger [8][9] facing escalating opposition from consumer groups and Virginia clean energy advocates urging 'extreme caution' [16]; FERC's June 2026 deadline to rewrite large-load grid interconnection rules — a process the Department of Energy formally directed FERC to undertake [20], with federal-state boundary tensions now identified as the central jurisdictional question [17][22][21]; and a widening structural gap between AI compute demand growing at 30–40% annually and US grid generation growing at roughly 2–3% [4], with PJM wholesale prices up 76% year-over-year and capacity prices reportedly up tenfold [5][6].

The gap between AI power demand and grid capacity has moved from projection to market reality — visible in both wholesale and capacity prices, forcing hyperscalers to bypass the grid, and triggering layered regulatory responses at federal, state, and international levels. How FERC resolves federal-state jurisdictional boundaries in June 2026 will directly shape Virginia's authority to protect consumers in the merger review, making two seemingly separate proceedings deeply entangled.

The United States power sector is undergoing structural transformation driven by AI compute. The US Energy Information Administration has forecast that domestic power demand will reach record highs in 2026 and 2027, with AI data centers identified as the primary driver [1]. Those facilities already consume approximately 4% of US electricity [2], a share analysts project will grow substantially. SemiAnalysis has quantified the trajectory with unusual precision: US AI power demand has grown from roughly 3 GW in 2023 to a projected 28 GW by end-2026, with AI compute demand expanding at 30–40% annually while US grid generation grows at only about 2–3% [3][4]. Wholesale power prices on PJM — the US's largest electricity grid — surged 76% year-over-year in Q1 as AI-driven data center demand intensified [5], and IEEFA separately reports that projected data center growth has driven PJM capacity prices up by a factor of 10 [6]. Analysts estimate that $800 billion in announced AI capex implies 50–70 GW of new power demand [7].

The structural mismatch has produced three competing responses: grid consolidation, regulatory reform, and grid bypass. On consolidation, the proposed $67 billion NextEra–Dominion merger — the largest in US utility history, explicitly motivated by data center electricity demand concentrated in northern Virginia [8][9][10] — would create the world's largest regulated electric utility with a large-load pipeline exceeding 130 GW [11]. Legal analysts identify Virginia's State Corporation Commission as the single most consequential regulatory venue, with broader authority over Dominion's home territory than federal or other state regulators [12][13]. Opposition has escalated into antitrust territory: the American Economic Liberties Institute calls the deal a 'mega utility monopoly that makes families pay for the AI boom' [14], Consumer Reports has documented how data center electricity costs flow to residential bills [15], and Virginia's Clean Virginia advocacy group has called for 'extreme caution' citing risks to Virginia ratepayers [16]. On regulatory reform, FERC has committed to acting on a large-load interconnection rulemaking by June 2026 [17][18][19] — a proceeding that the Department of Energy formally directed the agency to undertake [20], adding executive-branch weight to what legal analysts describe as a potentially game-changing rewrite of rules for data centers and other large loads [21]. Snell & Wilmer's analysis frames the proceeding's central challenge as resolving federal-state boundaries: a more expansive FERC role in direct large-load interconnection could constrain state commission authority — a jurisdictional tension with direct implications for the NextEra–Dominion review [22]. Physical bottlenecks constrain both paths: transformer supply lead times have reached 128 weeks or more [23], creating a hard constraint that capital alone cannot overcome.

The third strategy — grid bypass — has quietly become the industry default. SemiAnalysis reports that onsite natural gas generation has shifted from a fringe option to the default planning assumption for the next wave of US AI training clusters, driven by the inability to obtain grid interconnection on the timelines hyperscalers require [24]. Bloom Energy has secured a $7.65 billion data center fuel cell contract pipeline [25], a 2.8 GW Oracle partnership [26], and a $2.65 billion deal with AEP for 1 GW of fuel cell capacity [27]. Small modular reactors have also entered the conversation, with approximately 22 GW of SMR development for data centers reported [28], though skeptical analysts contend deployment timelines are too long and uncertain to address near-term demand [29][30].

AI's energy demand is straining power infrastructure internationally as well. Denmark's grid operator Energinet has imposed a formal three-month moratorium on new grid connection requests after total pending applications reportedly reached 60 GW — a figure that dwarfs the country's actual generating capacity [31][32]. Microsoft's planned data center in Kenya has sparked local blackout fears [33], and India is targeting 10 GW of data center capacity by 2030 but confronts serious bottlenecks in power availability [34][35]. Community opposition to data center development — citing power draw, noise, light pollution, and rate impacts — continues to intensify in affected localities [36][37].