FC Health Index1,428.60+0.42%Novo NordiskDKK 812.4+1.10%Intuitive SurgicalUSD 546.9−0.30%EU AI Act — Art. 6in forceM+7FDA 510(k) AI clearances (YTD)312+18 w/wNHS AI Diagnostic Fund£123mcommittedKarolinska trials open48+2Reimbursement CPT codes (AI)17+1 QFC Health Index1,428.60+0.42%Novo NordiskDKK 812.4+1.10%Intuitive SurgicalUSD 546.9−0.30%EU AI Act — Art. 6in forceM+7FDA 510(k) AI clearances (YTD)312+18 w/wNHS AI Diagnostic Fund£123mcommittedKarolinska trials open48+2Reimbursement CPT codes (AI)17+1 Q
Tuesday, 17 November 2026 · Oslo · London · New York

Intelligence · Explainer

How AI Data Centers Are Changing America's Energy Future: What Consumers Should Know

The EIA now expects US electricity use to set records in 2026 and 2027, largely because of AI data centers. A plain-language guide to why these facilities need so much power, where that power comes from, and what the latest research says about who pays.

AI-generated editorial illustration of large data center buildings and transmission towers behind a row of suburban homes at dusk. Not a photograph of a real facility.
AI-generated editorial illustration of large data center buildings and transmission towers behind a row of suburban homes at dusk. Not a photograph of a real facility.

Independent coverage

S

By Shel Evergreen

Contributing Writer — Climate Tech / Science · Freelance

Published 17 November 2026

7 min read

Evidence: Analysis

On November 4, voters in dozens of American towns were asked a question that would have sounded strange five years ago: should a data center be allowed here? According to The Washington Post, residents chose to restrict data center development in 40 of 41 local ballot races, many of them in Ohio. Six days later, the US Energy Information Administration said national electricity consumption will hit record highs in 2026 and 2027, driven largely by AI data centers.

Those two facts frame the story. Artificial intelligence is no longer just software. It is warehouses full of chips, substations, transmission lines and power plants, and ordinary electricity customers are now part of the conversation about how it gets built.

Why AI companies are investing so heavily in data centers

Training and running large AI models takes enormous amounts of computing, and that computing has to live somewhere. Companies such as Microsoft, Google, Amazon and Meta are racing to secure the buildings, chips and, increasingly, the electricity needed to serve AI products to millions of users. Whoever has capacity can offer faster, cheaper AI services. Whoever lacks it waits years for a grid connection. That is why spending has shifted from software toward concrete and copper, as we covered in our report on the power, financing and construction challenges facing US AI data centers.

How AI workloads differ from traditional computing

A traditional data center stores files, hosts websites and runs business applications. Demand rises and falls through the day. AI workloads, especially model training, run thousands of graphics processors at near full power for weeks. Each rack of AI servers can draw several times the power of a conventional server rack, and the heat is concentrated in a smaller space. Running a trained model for users, known as inference, is lighter per request but happens billions of times a day.

Why so much electricity and cooling

Almost every watt a chip consumes turns into heat, and that heat has to be removed. Older facilities blow chilled air through server rooms. Many new AI facilities pipe liquid directly to the chips or use evaporative cooling towers, which are more efficient but can use large amounts of water.

The scale is growing quickly. A December 2024 report from Lawrence Berkeley National Laboratory, prepared for the Department of Energy, found that data centers used about 4.4 percent of US electricity in 2023 and could reach between 6.7 and 12 percent by 2028. The wide range reflects real uncertainty about chip efficiency and how fast AI adoption continues.

What it means for local grids and utilities

Demand is concentrated, not spread out

Data centers cluster where land, fiber and power are available: Northern Virginia, Texas, central Ohio, Georgia and Arizona among them. A single large campus can need as much power as a mid-sized city. The EIA's November outlook projects total US consumption rising from a record 4,216 billion kilowatt-hours in 2025 to 4,306 billion in 2026 and 4,370 billion in 2027.

Grid upgrades cost money

New demand requires new transmission lines and substations, and utilities recover those costs through rates. A Cornell study published November 9 in Nature Energy examined four major US grids from 2020 to 2025 and estimated that data centers paid about 38 percent of the infrastructure costs linked to their growth in 2025, or $3.12 billion of $8.25 billion, with much of the rest falling on other customers.

Where the power comes from

The answer is a mix, and it varies by region. The EIA expects natural gas to supply about 40 percent of US generation in 2026, renewables to rise from about 24 percent in 2025 to 28 percent in 2027, nuclear to hold near 18 percent and coal to slip to 15 percent.

Tech companies are signing long-term deals across all of these. The most prominent nuclear example is Microsoft's 20-year agreement with Constellation Energy to restart the undamaged Unit 1 reactor at Three Mile Island in Pennsylvania, renamed the Crane Clean Energy Center. Solar, wind and battery projects are faster to build, while gas plants provide around-the-clock supply. Researchers at Oregon State University and the National Laboratory of the Rockies found in a Texas modeling study that allowing new gas generation near data centers could reduce the residential cost impact, which shows how much local planning choices matter. That paper is a preprint and has not yet been peer reviewed.

Water, emissions and sustainability

Water is the less visible concern. The Berkeley Lab report estimated that US data centers directly consumed about 17 billion gallons of water in 2023, mostly for cooling, and that figure excludes water used by power plants supplying them. In dry regions such as Arizona and parts of Texas, that competes with homes and farms. Facilities using closed-loop liquid cooling use far less water, but they often need more electricity. Emissions depend on the grid: a data center powered mostly by gas has a very different footprint than one matched with new nuclear or wind.

How communities and consumers could be affected

This is where the evidence gets genuinely mixed, and it is worth being careful. A Harvard Business School and Wharton working paper summarized by HBS found that only about 4 percent of residential price increases from 2020 to 2025 could be attributed to data center demand, with inflation, fuel costs and grid upgrades driving most of the rest.

The Cornell study reaches a sharper conclusion about distribution. It found data centers pushed up wholesale prices by more than 10 percent in the most exposed market hubs by 2025, and CBS News reported its estimates of nearly 20 percent in California's southern Bay Area and close to 17 percent around Dallas-Fort Worth. The researchers found rural and lower-income households carried a disproportionate share of the burden.

Both can be true. Nationally, data centers are a modest factor in bills so far. Locally, in fast-growing hubs and under certain cost-sharing rules, the effect can be meaningful. Communities also weigh benefits such as construction jobs and property tax revenue against noise, land use and water concerns.

What AI infrastructure might look like by 2030

Forecasts beyond two or three years are uncertain, and readers should treat them that way. What can be said with confidence is the direction: more demand, more pressure on state regulators to decide who pays for grid upgrades, and more data center operators building or contracting their own generation. Several states are already considering special rate classes that require large new customers to cover more of their connection costs. Efficiency gains in chips and cooling could slow growth, as they did in the 2010s, when data center energy use stayed roughly flat despite rapid growth in computing.

For consumers, the practical takeaway is simple: watch your state utility commission. Decisions about rate design and cost allocation, more than the technology itself, will shape what AI means for your electric bill.

Frequently asked questions

Are AI data centers raising my electric bill?

Possibly, but usually modestly so far. National research attributes a small share of recent residential price increases to data centers, while studies of specific hubs find larger local effects. Your exposure depends on your region and your state's rate rules.

How much US electricity do data centers use?

Lawrence Berkeley National Laboratory estimated about 4.4 percent in 2023, with a projected range of 6.7 to 12 percent by 2028.

Do data centers use a lot of water?

Some do. US data centers directly consumed about 17 billion gallons in 2023, according to Berkeley Lab, mostly for cooling. Water use varies widely by cooling design and climate.

Will nuclear power run AI data centers?

It will play a growing role through deals such as the Three Mile Island restart, but natural gas and renewables are expected to supply most new power in the near term.

Limitations

This explainer relies on published government data, peer-reviewed research, a working paper, a preprint and news reporting. It does not include original measurements. Cost-attribution studies use different methods and are not directly comparable, and forecasts may change with chip efficiency, policy and energy prices.

Sources

Published 17 November 2026