The buildout of artificial intelligence has quietly become one of the biggest forces reshaping the United States’ electricity system in decades, and the strain is starting to show up well beyond the data centers themselves. Utilities, regional grid operators, and federal regulators are now racing to keep pace with a category of electricity customer that barely existed a decade ago: a single campus of AI computing hardware that can draw as much power as a mid-size city.
How Much Power AI Data Centers Actually Use
The U.S. Energy Information Administration has tracked data center electricity use climbing steadily across the country’s commercial building stock, driven overwhelmingly by the growth of standalone data center campuses rather than the smaller server rooms tucked inside offices and hospitals. The agency’s separate Short-Term Energy Outlook projects total U.S. electricity consumption reaching record highs of roughly 4,193 billion kilowatt-hours in 2025 and 4,283 billion kilowatt-hours in 2026, a jump the agency attributes in large part to data center demand after more than a decade of relatively flat nationwide electricity use.
Why AI Workloads Draw So Much More Than Ordinary Computing
Not all data center demand is created equal. Training and running large AI models requires racks of specialized processors running at sustained, near-maximum power draw for long stretches, a very different load profile than the traditional cloud computing that powers email or web browsing. That density means an AI-focused data center campus can require tens or even hundreds of megawatts of continuous power, comparable to a small city, concentrated on a single parcel of land, which is a very different planning problem for a utility than the same amount of demand spread across thousands of homes.
Grid Bottlenecks Show Up as Delays, Not Blackouts
For most of the country, the strain has shown up less as rolling blackouts and more as long waits. Developers proposing new data centers in high-demand regions are increasingly told that connecting to the transmission grid could take years, because existing interconnection queues were never designed to process requests of this size and frequency. Some companies have responded by contracting power directly from independent producers or installing their own natural-gas generators on-site so a project can move forward without waiting on a utility’s standard connection timeline, a workaround that regulators are watching closely because it can shift costs and reliability questions outside the normal planning process.
Where the Strain Is Showing Up First
The pressure is not evenly distributed across the country. Northern Virginia, home to the densest concentration of data centers in the world in what is often called Data Center Alley, has become a focal point for utility planning because so much new demand is arriving in one relatively small service territory. Texas presents a different version of the same problem: its grid operates largely independent of neighboring states, so a wave of new data centers and cryptocurrency-mining facilities adds load to a system that already has to manage extreme swings between winter cold snaps and summer heat waves. Georgia, Ohio, and parts of the broader Midwest have seen similar clusters of proposed projects, each one forcing local utilities to rethink long-term generation and transmission plans that were built around much slower, steadier demand growth.
Regulators Are Racing to Speed Up Grid Connections
The Federal Energy Regulatory Commission moved in 2026 to address the backlog directly, issuing orders to the country’s regional grid operators requiring them to justify or revise their interconnection rules for what the agency calls “large loads,” defined as facilities with a peak demand above 50 megawatts connecting at higher transmission voltages. The commission’s action followed a request from the Department of Energy to accelerate how quickly data centers and other major electricity users can be connected, and it reflects how much the interconnection process itself, not just generation capacity, has become a bottleneck. A separate Department of Energy report examined the scale of the projected demand increase and what it could mean for grid planning nationwide.
The Cost That Flows Back to Ordinary Ratepayers
The Energy Information Administration’s longer-range projections suggest the trend has years left to run: data centers could account for somewhere between 22 percent and 33 percent of commercial building electricity use by 2050, up from a much smaller share today. That kind of sustained demand growth affects everyone connected to the same grid, since new transmission lines, power plants, and grid upgrades needed to serve data centers are typically paid for, at least in part, through the rates charged to all customers on a utility’s system. How data centers are planned, sited, and billed for their share of grid upgrades is quickly becoming one of the more consequential regulatory fights in the electricity sector, precisely because the AI buildout shows no sign of slowing down.
This article was produced with the assistance of AI and reviewed by Morning Overview editors.
More from Morning Overview
- More than 60,000 people flee the Spokane area as complex fires overrun 600 structures
- 11 engines built to run well past 200,000 miles
- Card skimmers hidden on gas pumps and ATMs are draining accounts, and here’s the tell
- The FBI says hackers are hijacking outdated home routers, and it named the models to check
