AI is power-hungry: Will consumers pay the price with higher electric bills?

Artificial intelligence is seeping into daily routines, helping to craft emails, keep track of schedules and manage other tasks more efficiently. Much of AI’s potential might still be beyond imagination, but one thing’s clear — its power needs will be enormous.

Less certain is who will pay AI’s electric bills and just how much consumers will bear.

Such questions are open to debate as big tech firms — among them Microsoft, Meta, Google and Amazon — rush to lock in long-term, clean power sources for ever-larger data centers, making deals with utilities and power plant operators in Maryland and across the U.S.

Those include Baltimore-based Constellation Energy, the nation’s largest operator of nuclear reactors, which has power purchase deals with Microsoft and Meta, and Rockville-based X-energy Reactor Co., a developer of small nuclear reactors.  X-energy plans to develop 64 advanced modular reactors for Amazon, a key investor, by 2039.

Current and projected increased electric demand has spurred much-needed investment in new carbon-free energy generation, both conventional and new technologies, such as small modular reactors and geothermal, energy company officials say.

Increased demand means, “American businesses are succeeding,” Kathleen L. Barrón, Constellation’s chief strategy and growth officer, said in an interview. “We in the electric sector need to be able to meet that demand. The alternative is to say we don’t want growth, and we don’t think that’s good for anyone.”

The AI race has led to larger, more energy-intensive data centers

The race to develop AI has accelerated the scale and energy intensity of data centers, which require a consistent and steady electricity supply. A March study from Harvard University shows facilities that will consume hundreds of megawatts of power, as much as the city of Cleveland, have been under development since 2023. By the end of last year, companies were expanding to gigawatt-scale data center campuses. Even bigger centers, demanding more energy than the nation’s largest nuclear power plant, are envisioned,  the study said.

Within five years, it said, data centers may consume as much as 12% of all U.S. electricity and could be largely responsible for quintupling annual demand growth.

Since the release of OpenAI’s ChatGPT in late 2022, technology to create machines that can think like humans has been used in self-driving cars, customer service chatbots, programs that detect financial fraud and plenty more. AI is expected to contribute trillions to the global economy and play a role in addressing climate change, according to the University of Cincinnati, which specializes in AI business applications.

“It’s a paradigm shift, probably more so than the concept of the Internet or the gasoline engine,” said Jeff A. Shaffer, director of the business school’s Applied AI Lab. “We don’t yet know what that impact on society is. It’s going ubiquitous, it’s going to be in to be in everything we do.”

Proposed infrastructure faces opposition

The framework supporting the fast-changing technology is already sparking controversy. In three Maryland counties, landowners opposing a proposed 67-mile high voltage transmission line argue the Maryland Piedmont Reliability Project will cut across hundreds of properties, harming or destroying cropland, conserved land and waterways, only to feed the appetite of data centers in Virginia.

Demand from data centers sparked a nearly 10% annual jump in wholesale electricity prices set at a recent interstate “capacity market auction,” for a regional power grid serving 67 million people in Maryland and a dozen other states, Maryland People’s Counsel David S. Lapp said in a recent analysis. Data centers accounted for more than 5,400 megawatts of increased demand compared with last year’s level.

The Harvard study raised questions about who should shoulder AI’s energy costs.  Utilities are prioritizing the needs of a few, energy-intensive customers to satisfy the surge in computer chip-filled warehouses, says the study, by the law school’s Environmental & Energy Law Program. It argues that utilities are funding discounts to Big Tech by socializing their costs through electricity prices charged to the public.

That happens when utilities build infrastructure for new data centers but then spread the costs to everyone and when data center use impacts interstate wholesale electricity markets, trickling down to ratepayers in higher costs to ratepayers, said Ari Peskoe, director of the Harvard Electricity Law Initiative.

“As society’s demand for electricity grew, we all kind of paid for it under the theory that we all benefit from economic development and population growth,” Peskoe said. “But the growth of data centers is challenging the fundamental premise of utility regulation. Now we have massive cost increases that are being driven by just a handful of facilities being built by the world’s wealthiest corporations.”

A review of nearly 50 regulatory proceedings about data centers’ rates found ways in which existing and new rate structures and “secret contracts” are transferring Big Tech’s energy costs to the public. For instance, it found utility regulators frequently approved special contracts with just a cursory analysis, instead of gathering the lengthy evidence that typically comes from utilities and other parties in state regulators’ rate cases.

Peskoe argues that AI consumers should not be responsible for the technology’s power needs any more than grocery customers should pay store electric bills.

States compete for data centers

While there’s disagreement over how to allocate data center costs, most states have similar goals when it comes to attracting such facilities. At least 30 states, including Maryland, are competing with incentives to woo data center construction along with the associated tax base and jobs. Maryland, Texas and New York have taken steps recently to support new nuclear capacity and have turned to nuclear to add additional grid capacity.

Proponents of nuclear and next-generation nuclear believe the carbon-free, reliable form of energy is uniquely positioned to meet growing demand from data centers as well as electrification and onshoring of manufacturing.

But most of the nation’s nuclear fleet is aging, with only one new plant built since the 1980s, and about a dozen reactors shut down in the past decade, mostly for economic reasons. Some, including two owned by Constellation outside Maryland, have stayed open thanks to state and now federal energy production tax credits. Constellation operates power plants mainly in Maryland, Illinois, New York and Pennsylvania and supplies electricity on the competitive retail market.

X-energy, which plans to open a small nuclear reactor testing center in Frederick early next year, is working toward satisfying the power needs of “hyper-scalers” that has been rising about 100% every 18 months, said Steve Miller, executive vice president. Its customers include Dow Inc. and Amazon.

“It doesn’t show any signs of slowing down,” Miller said in an interview. “So really, in order to provide clean power for that large of a demand, I mean, nuclear is truly the only real answer.”

Constellation looks to expand nuclear power capacity

Constellation is working towards expanding its fleet through restarts, by adding to the output of existing resources through updates and by seeking to relicense plants. All are lengthy and costly processes that can be better financially justified by signing up long-term customers.

The owner or operator of 25 reactors across six states plans to add capacity at two of its plants through such deals. It will supply electricity to Microsoft’s Mid-Atlantic data centers by restarting Unit 1 at the former Three Mile Island, now called Crane Clean Energy Center. It will provide power to Meta, owner of Facebook, Instagram and WhatsApp, for the next two decades from a nuclear plant in Chicago.

“The large hyper-scalers, like most large corporate organizations, are looking to buy clean power to power their operations,” after decades of investing mostly in wind and solar power, Barrón said.  “What’s happened of late is that these companies have changed their perception of what’s clean power to include all sources that are zero carbon and not just wind and solar.”

That has led to deals such as Google buying hydropower from two dams in Pennsylvania to power AI operations in the PJM grid area.

In such “over-the-grid” deals, hyper-scalers get credit for matching consumption with clean power, Barrón said, “but the power is still going onto the grid every day. It’s still being used by families and businesses every day. It’s still helping to keep the lights on every day.”

In February, the company said it plans to invest about $100 million to boost future energy output at its Maryland nuclear power plant, Calvert Cliffs Clean Energy Center, formerly known as Calvert Cliffs Nuclear Power Plant, in Lusby. The company will be upgrading electrical systems and plant equipment to prepare for a potential renewal of operating licenses.

Barrón said both the electricity and data economy industries are looking for ways to meet demand efficiently in a way that takes advantage of the capacity that goes unused in systems built to serve peak loads, as a way to better control price increases for consumers.

Some, though, say AI’s energy demands are simply too difficult to predict. That’s because efficiencies are likely to occur as yet-to-be-developed AI models, training methods and computer chips advance, presumably cutting down on energy consumption.

“I’m not coming from the camp of we should ignore the energy aspect of it, but I think we just need to recognize we’re in a fluctuating period right now where people are trying to figure it out,” said Shaffer, of Cincinnati’s AI lab.  “When the models get more advanced, we’re going to get to the point where the models can improve upon themselves without the humans.”

Have a news tip? Contact Lorraine Mirabella at lmirabella@baltsun.com or (410) 332-6672.

Originally Published: August 4, 2025 at 5:00 AM EDT