Every AI headline about a new model or a new data center campus is secretly an energy story. Training and running large models requires enormous, always-on electricity — a single hyperscale AI campus can draw as much power as a mid-sized city — and the U.S. grid was not built for this kind of demand growth. For the first time in nearly two decades, utility planners are revising load forecasts sharply upward, and the constraint on AI buildout is shifting from chips to megawatts.
That shift makes Washington newly relevant to a sector investors used to think of as boring. Federal Energy Regulatory Commission (FERC) rules on transmission interconnection, Department of Energy loan guarantees and permitting reform for nuclear and gas, state public utility commission rate cases, and even tariff policy on grid components (transformers, steel, solar panels) now directly determine who can build power fast enough to capture AI demand — and who gets left waiting in an interconnection queue that in some regions stretches past 2030.
This guide is a durable reference for that mechanism: how policy decisions translate into which companies profit from the buildout, which real tickers sit in the crosshairs, and how to keep tabs on the story as it evolves over years, not news cycles.
The mechanism: why Washington controls the electrons
Electricity in the U.S. is not a free market the way software is — it runs through a heavily regulated system of federal transmission oversight, state-approved utility monopolies, and multi-year permitting for anything from a new gas plant to a transmission line. FERC sets the rules for how new power sources connect to the grid ("interconnection queues"), state public utility commissions decide whether a utility can raise rates to pay for new capacity, and federal agencies (DOE, NRC) control permitting timelines for nuclear plants and loan guarantees for advanced generation.
This means the pace of the data-center power buildout is gated by policy decisions, not just engineering. A FERC order that speeds up interconnection reviews, a state commission that approves a utility's multi-billion-dollar capital plan, or federal permitting reform for nuclear reactors can each unlock billions in construction spending almost overnight. Conversely, a slow queue or a rate case denial can strand a data-center project for years even if the land, chips, and capital are ready. Reading these policy signals — not just corporate press releases about new data centers — is how you get ahead of where the money flows.
Who profits: regulated utilities with the load growth
The most direct beneficiaries are regulated utilities in regions where hyperscalers are concentrating data centers, because state-approved rate cases let them earn a guaranteed return on the capital they spend building new generation and transmission to serve that load. Watch NextEra Energy (NEE), the largest U.S. utility owner with heavy renewable and grid investment; Southern Company (SO), whose Georgia territory has become a major data-center hub around Atlanta; Dominion Energy (D), whose Virginia service territory includes "Data Center Alley" in Loudoun County, the densest concentration of data centers on earth; American Electric Power (AEP), a major transmission owner across the PJM grid region; and Duke Energy (DUK), building out capacity across the Carolinas and Southeast.
The mechanism here is straightforward: these companies file integrated resource plans with state regulators, get approval to spend on new plants and lines, and earn a regulated rate of return on that capital base for decades. Each favorable rate-case ruling or transmission-cost-recovery decision is a direct, trackable event that expands the utility's earnings base.
Who profits: the hardware that turns policy into megawatts
Behind every utility capital plan sits equipment makers who actually build the new capacity. On the generation side, GE Vernova (GEV) makes the gas turbines that are the fastest way to add dispatchable baseload power, and its turbine order backlog is one of the cleanest real-time gauges of how fast utilities are committing to new gas-fired capacity. On the nuclear side, Constellation Energy (CEG) operates the largest U.S. nuclear fleet and has signed long-term power agreements directly with hyperscalers, while Vistra (VST) owns a mixed nuclear-and-gas fleet that benefits similarly from data-center power contracts.
On the grid-equipment side, transformers and switchgear have become a genuine bottleneck — lead times stretched to multiple years — benefiting makers like Eaton (ETN), whose electrical products division supplies data-center power infrastructure directly, and Vertiv (VRT), which specializes in the power and cooling systems inside data centers themselves. Federal tariff policy on imported steel, aluminum, and electrical components can move input costs for all of these suppliers, making Section 232/301 tariff actions and Commerce Department trade rulings worth tracking as a secondary lever.
The nuclear angle: permitting reform as a catalyst
Nuclear power has re-emerged as a preferred long-term solution for AI's constant, high-density power needs, because reactors run at near-100% capacity factor without the intermittency of wind or solar. This has made federal nuclear policy — Nuclear Regulatory Commission licensing reform, DOE loan guarantee programs, and legislative permitting streamlining — a genuine catalyst class. Constellation Energy (CEG) has already moved to restart previously retired reactor capacity under long-term contracts with technology companies, an early template for how idle nuclear assets can be revived for AI demand.
Small modular reactor (SMR) developers remain earlier-stage and speculative, so investors should treat any specific SMR name with caution and focus instead on the policy mechanism: every NRC rule that shortens licensing timelines or every DOE program that de-risks reactor financing expands the addressable set of nuclear capacity that can plausibly come online this decade, which is what ultimately supports contracted revenue for existing nuclear operators.
How to spot it as it happens
Track the interconnection queue data that grid operators like PJM, ERCOT, and MISO publish, since a sudden clearing of stalled projects signals real construction spending is about to follow. Watch state public utility commission dockets in data-center-heavy states — Virginia, Georgia, Texas, Ohio — for rate case filings and integrated resource plan approvals; these are public filings, not rumors, and they name the exact capital figures utilities plan to spend. Follow FERC's docket activity on transmission planning and interconnection reform, since FERC Order changes have historically been the single biggest unlock or bottleneck for new generation nationally.
On the corporate side, quarterly earnings calls from the utilities and equipment makers above increasingly disclose data-center-specific load growth and backlog figures directly — GE Vernova's turbine backlog and Vertiv's data-center order book are now standard line items analysts ask about. Treat any announcement of a new gigawatt-scale data center as a demand signal, but treat the follow-on utility rate case or interconnection approval as the actual investable event, since that is when regulated capital spending — and the associated earnings growth — gets locked in.
The risks that cut the other way
This mechanism can reverse. If state regulators push back on rate increases needed to fund data-center-driven grid buildout — a real political risk as residential ratepayers object to subsidizing hyperscaler demand — utility capital plans can be trimmed or delayed. Local and state permitting fights over new gas plants, transmission lines, or water use for cooling can stall specific projects for years regardless of federal policy, since siting authority is often local.
Demand forecasts themselves are a risk: if AI compute demand growth slows, or if efficiency gains in chips and cooling reduce power draw per unit of compute faster than expected, the aggressive load-growth forecasts utilities are currently building capital plans around could prove overstated, leaving overbuilt capacity and weaker-than-expected returns. Readers should treat data-center power as a real, durable structural theme, but one that plays out through years of regulatory dockets and rate cases rather than a single clean catalyst.
Bottom line
The data-center power crunch turns AI's electricity appetite into a decade-long capital cycle: utilities that can build generation and transmission, the reactor and turbine makers who supply new baseload capacity, and the grid-equipment suppliers who connect it all are set to earn on a multi-year backlog that Washington's permitting and interconnection decisions can accelerate or choke off — track the policy signals, not the headlines about any single data center.