Dec. 29 (UPI) — The proliferation of data centers in the United States continued to accelerate in 2025 but experts, developers and utility companies are challenged with projecting the energy needs of the future.
Sen. Bernie Sanders, I-Vt., is calling for a moratorium on all data center construction for artificial intelligence, saying lawmakers and the public need a chance to “catch up.”
“There is a whole lot about AI and robotics that needs to be discussed,” Sanders. “This process is moving very, very quickly and we need to slow it down. We need all of our people involved in determining the future of AI and not just a handful of multibillionaires.”
The United States has more data centers than any other country in the world, amassing more than 5,000, with a strong uptick in new centers since 2021. Whether growth will continue at a high rate and how energy infrastructure can support that growth are the questions that stakeholders are trying to answer.
Uptime Institute told UPI it is difficult to answer just how many data centers there are or how many are coming but it projects the total planned power increased by 177% in over 2024.
Dr. Jonathan Koomey has been studying computing electricity use and data centers for 30 years. His research was part of a 2024 report by the Lawrence Berkeley National Laboratory on data center energy usage.
In the report, researchers found that data center energy usage in 2023 represented about 4.4% of total U.S. electricity consumption. This was an increase from about 1.9% of electricity consumption in 2018.
The data on energy consumption in the Lawrence Berkeley National Laboratory report includes compute data centers such as those used by companies like Google and Meta as well as communications data centers such as those operated by telecommunications companies.
“If you are following the news, you might have the impression that there is now explosive electricity demand growth and that it’s mostly being driven by data centers,” Koomey said. “The issue is that except for some specific states there is no evidence of explosive demand growth. The electricity used in the U.S. in 2023 was the same as in 2018.”
As data centers have sprouted up across the country, other factors have offset some of their demand on the power grid, such as gains in energy efficiency from things like household appliances.
Whether the growth of data centers will be a persistent boom, a shorter term bubble or something in between is a point of contention.
“There’s a whole industry around exaggerating IT electricity use. I’ve been at this long enough to see multiple cycles of this,” Koomey said. “Back in the year 2000 or so, the dot-com era, there are a couple guys running around talking about how the Internet will use half of all our electricity in 10 years. It was all a bunch of nonsense.”
Koomey added that belief in claims of explosive growth affects the behavior of investors and stakeholders. When the Internet experienced a period of rapid growth in the mid ’90s, there was a belief that the fiber optic communications network needed rapid growth to keep up or stay ahead of demand.
“There was this idea that the Internet, defined as data flows over the Internet, doubled every 100 days,” he said. “That was true for a six month period in 1995. But before that and after that it only doubled every year. The people in the tech industry, particularly the networking industry, basically used that fact to sell the street on spending essentially infinite money, all the money they could raise, on fiber. The problem is the demand wasn’t growing that fast, and by the year 2000, 97% of the fiber network was unused.”
Koomey said he believes this example is analogous to what is currently happening with data centers. There is an idea that data centers and the energy to support them can’t grow fast enough to keep up with demand.
Former Google CEO Eric Schmidt testified to the House Energy Energy and Commerce Committee in April that the “AI revolution is underhyped” and its energy demands will create an infrastructure crisis requiring investments in all sources of energy.
“It’s really not responsible, in my view, to claim that there’s an explosive demand because the data just doesn’t show it,” Koomeys said. “They’re creating a sense of urgency around building more fossil fuel power plants and keeping old fossil plants around and I think that’s a mistake. There is no urgency from society’s perspective to invest in electricity used for AI.”
Mark Mills, director of the National Center for Energy Analytics, has been involved in several aspects of data centers for decades, from construction and energy supply to policy. He served in the White House Science Office under President Ronald Reagan.
Mills argued that the impact of data centers on U.S. energy consumption has been moderated by “one-time gains” in efficiency that cannot be replicated in the future. Namely, data center cooling systems have become more efficient.
“The efficiency by which they were being cooled was quite poor in the old designs,” Mills said. “For the first 15 years of the data center world, roughly 2005 to 2020, that metric got better dramatically. The data centers’ power use for computers rose but the power-use for cooling didn’t. It actually went down. So that offset, so it gave the illusion that the compute load wasn’t going up, but it was. The effect of it was true, that it moderated the overall increased electrical demand.”
Mills emphasized AI has added a new level of demand that is energy intensive.
The discussion around data centers is deeply intertwined with the growing adoption of AI. Running AI models at a large scale is energy intensive itself. Mills likens AI and data centers to the jet engine and aviation.
“The development of the commercial-viable jet engine aircraft in 1958, the Boeing 707, changed the aviation industry profoundly. It incredibly expanded the utility, value and economics of flying,” he said. “There’s a difference between what AI does and regular computers. AI amplifies what computing functionally can do by making it more useful.”
To Mills, there is no question that the United States is in the midst of an AI boom. Beyond that, it is a guessing game over how much of a strain the boom will put on energy providers, if data centers connect to the grid at all. Some data centers have their own dedicated energy production.
“AI assist and search function can increase the energy use of doing the search by five or tenfold,” he said. “I’ve very energy intensive because it’s very compute intensive. It’s not a flaw. People mentally think it’s a flaw and somehow they’ll figure it out and fix the flaw. It’s not a flaw. It’s a feature. It’s a feature of the fact that the software is running all the time to solve a very difficult problem our brains do really easily. But computers don’t.”
Uptime Institute’s 2025 global survey notes that energy data collection dropped by 9% in North America. Energy collection data includes tracking of energy consumption and power usage efficiency — a metric used to calculate data center efficiency.
More than 60% of the electricity consumption of all data centers in 2024 came from servers, the International Energy Agency reports. About 20% of consumption was for cooling systems, about 5% was for storage systems and another 5% for networking equipment.
Research by Uptime Institute and IEA underline the uncertainty of what future infrastructure will be required.
Koomey said the two key uncertainties at the center of the problem are how service demand will grow and how the efficiency of delivering service will change
“Almost all the uncertainties are on the downside for what’s going to happen,” Koomey said “What I think is happening is we’re in another period of growth but people are forgetting that efficiency can offset that growth in service demand, at least to some degree. To me, the biggest uncertainties in these forecasts is that service demand may not grow as fast as people think and that efficiency may grow faster than people are assuming.”
Regardless of how energy demands grow, Mills said supplying the energy data centers requires will be a choice.
“Those who say we can’t make enough electricity don’t understand arithmetic,” Mills said. “We could choose not to supply the electricity as sort of a political or a policy decision but the ability to supply the electricity is clearly feasible. It’s difficult but not impossible by any means. Basically you have to build more engines and deliver more gas.”
This leads to the concerns raised by Sen. Sanders. The cost of expanding data centers and AI use goes beyond tech companies and utility providers. There are also concerns about the price consumers may pay to support data centers built in the jurisdictions they live in as well as their impacts on the environment locally.
Dr. Kelly Twomey Sanders, associate professor in the University of Southern California’s Sonny Astani Department of Civil and Environmental Engineering, told UPI more transparency is needed to help local governments make informed decisions about bringing or attracting data centers to their communities.
Sanders’ research focuses on reducing the environmental impact related to providing energy.
“I wish more local governments were asking for more transparency in terms of requiring data centers to report the water, electricity and pollution impacts of facilities,” she said. “It is difficult to manage what is not measured.”
Water use, in particular, is something that Sanders said there is little transparency about. California lawmakers proposed a law increasing requirements for data centers to report water and energy use but it was vetoed by Gov. Gavin Newsom.
“I don’t think we have a good sense of how much water the data center uses because companies are not required to report,” Sanders said. “There have been estimates for facilities and regions but companies in most cases are not required to report water or energy consumption. So researchers have very little actual water use data. Most estimates are based on models that have a lot of assumptions and uncertainty baked in.”
COMMENTS
Please let us know if you're having issues with commenting.