How Is AI Impacting Energy Production And Prices?

This series explores the impacts of data centers on water supply, energy use, air quality and stormwater runoff in the Chesapeake Bay watershed. Data centers house the computer systems that enable internet activity and, increasingly, artificial intelligence. Northern Virginia, in the middle of the Bay watershed, is the global epicenter of these warehouse-like facilities. Their footprint is now spreading into Maryland and Pennsylvania.

ANNAPOLIS, Md. – Electrical energy may seem like an unlimited resource in the United States, where grid operators are federally mandated to ensure the lights stay on. But the insatiable appetite of artificial intelligence has begun to strain that power system in a way that is also threatening environmental goals.

In some parts of the region, fossil fuel-powered plants that were scheduled for decommissioning have been kept online to power the grid’s growing needs. And solar and wind power sources that had been ramping up are now viewed by some as unreliable in a data-driven landscape that demands 24-7 access to power.

Many AI users have yet to grasp the massive energy consumption associated with everyday use of the technology, even as it’s been integrated into a growing number of routine tasks. The focus of technology companies and the federal government on winning a global “AI arms race” is showing no signs of slowing, even as the U.S. Department of Energy acknowledges the industry’s contribution to an “energy emergency facing the U.S. power grid.”

These changes are occurring even more rapidly in data center hotspots like Northern Virginia and the broader Chesapeake Bay region that supplies energy to the facilities. A report in late 2024 found that supplying the energy to meet even half of the industry’s projected future demands would require Virginia to purchase it from outside the state. But that may be more difficult as other states also work to attract and supply power to growing numbers of data centers.

Data centers are the warehouse-like buildings clustered near power lines and fiberoptic cables that have for decades housed the infrastructure enabling the internet. But the newer data centers being built to train and run AI models use far more processing power than their predecessors, resulting in exponentially greater energy and cooling demands.

A U.S. Department of Energy report found the amount of energy consumed by data centers tripled in the decade leading up to 2024. This increase in the country’s total energy use followed about 15 years of relatively flat electricity demand, from the mid-2000s to the early 2020s. That, coupled with the rapid pace of the AI ramp up, has left power grid operators and suppliers scrambling to keep up with infrastructure demands.

The DOE report also predicted that data center energy consumption alone would double or triple again by 2028 to consume as much as 12% of the country’s electricity. Some industry officials say it could be even more. But predicting how much energy the still nascent AI industry will consume is also inherently risky.

A July 7 report prepared by London Economics International LLC for the Southern Environmental Law Center found that many regional projections reflect a bias toward overstating future demand. That’s in part because data center developers have an incentive to say they will build a given project in more than one jurisdiction to get “in a queue” for future power supplies. This results in some loads being counted more than once in demand projections at both the regional and national level.

To determine how many projected data centers may actually come to fruition in the near term, LEI’s analysis considered the global availability of the semiconductor chips that AI data centers require. For all the data centers projected in the U.S. from 2025 to 2030 to go forward, the report found it would require 90% of the global chip supply for that period be directed to the U.S. market.

LEI energy economist Marie Fagan said that “just isn’t realistic,” since the U.S. represents only half of the global demand for chips.

“It’s a sanity check,” Fagan said of the chip availability. “Even people in the industry who would love to see demand grow are saying it won’t all show up.”

The Trump administration also issued several emergency orders under the Federal Power Act to prevent aging infrastructure from retiring. In the case of the Eddystone Generating Station near Philadelphia, PA, the order directed the plant’s natural gas and oil-fueled generators to continue running a day before its planned retirement at the end of May, which had been scheduled for nearly two years.

Both the regional grid operator PJM and the plant’s owner Constellation had agreed that it was “uneconomic” to continue running the aging units. The Natural Resources Defense Council is suing the Department of Energy over the mandate to keep the plant running anyway, insisting that the environmental and economic costs of doing so will fall on ratepayers.

The California-based nonprofit GridLab also released an analysis of the report driving the DEO’s energy emergency. It found that the department both overstated future demand and understated the amount of new capacity that would be added to the grid in the coming years to justify delaying the retirement of old plants.

“It’s a manufactured emergency,” said Clara Summers, campaign manager for Consumers for a Better Grid. “The experts on this all agree that it was fine for these plants to retire. Having this abuse of emergency power is really concerning because who pays for it? Consumers.”

In places like Virginia, which is home to about half of all data centers in the U.S., the outsized energy appetite of AI data centers isn’t a future prospect. It’s the present.

The data center industry already accounts for more than a quarter of the state’s electricity use, according to a report by EPRI, a California-based research institute. And its demands for electricity had been doubling every year, then every six months. By the end of 2024, data centers in the state were consuming about 40 gigawatts of power, according to Dominion Energy. That’s enough to power about 10 million homes in a state with 8.8 million residents.

This steady and increasing growth in energy demand has led to a flurry of new power generation projects and transmission lines to fuel them, many of them cutting through tree-lined backyards and near national battlefield parks.

“Growth is a good thing, but that means that we have a big-time growing demand in power,” Virginia Gov. Glenn Youngkin said at the Virginia Energy Summit in Richmond in June. “In order to meet that demand in power — let me just be clear — we need to nearly double our adjusted power generation.”

To that end, Youngkin, a Republican, has taken an increasingly aggressive stance against the 2020 Virginia Clean Economy Act, which created a blueprint for the state to source 100% of its energy from renewable sources by midcentury. Youngkin has said he now favors an all-of-the-above approach to power generation that includes adding offshore wind, nuclear fusion and, he said at the summit, “clean coal.”

“Bottom line, we don’t have enough [power] and VCEA doesn’t work,” Youngkin said.

Amazon announced in June its plans to spend $20 billion on two data center complexes in Pennsylvania, including one that would siphon power directly from an existing nuclear power plant. In July, President Donald Trump appeared at the Pennsylvania Energy and Innovation Summit to announce that AI companies would be investing $92 billion in energy and related infrastructure in the state, including new natural gas-power stations.

The White House also released an “AI action plan” in late July aimed at accelerating the industry’s growth nationwide. Maryland has sought to get in on the data center game too, with mixed results.

The Maryland’s Office of the People’s Counsel, an independent state agency that represents residential customers, filed comments with the Federal Energy Regulatory Commission in July declaring that, regardless of how many data centers are in the state, its residents are already footing the bill for the industry’s regional growth. Data center growth drove an increase in the grid’s capacity investments last summer, for example, contributing to price increases that will not hit the bills of most customers until after the summer, the OPC said.

“In future months and years,” Maryland People’s Counsel David S. Lapp said, “customers will see their utility bills rise significantly if regulators don’t fix current policies.”