In July, Reuters reported that America’s largest power grid was under strain as data centers and artificial intelligence (AI) chatbots increasingly consume power faster than new plants could be built. Within PJM Interconnection’s grid, which covers 13 states from Illinois to Tennessee, Virginia to New Jersey, there were more data centers than anywhere else in the world.
It was expected to spike energy costs by up to 20%.
According to a report from NPR, citing data from the International Energy Agency (IEA), the typical data center uses as much electricity as 100,000 households, and the most significant data centers now under development will consume 20 times more. In addition, these data centers require billions of gallons of water to keep the hardware cool.
“Training large models requires immense computational power delivered by advanced servers that pack in powerful computer chips such as GPUs,” explains Archit Lohokare, cybersecurity advisor and general manager of the security business at CyberArk.
“These servers and components, that can perform trillions of calculations per second, require many times as much energy as traditional servers to run,” Lohokare told ClearanceJobs. “A single AI-enhanced internet query can use roughly 10 times the electricity of a conventional search, so when you scale that to billions of queries, the impact on the grid becomes very real. In the U.S., data centers used an estimated 183TWh of electricity in 2024, about 4% of all power, and it is expected that this could more than double by 2030 as workloads expand.”
Strain in Key Areas
The states most impacted by the surge in data centers are, not surprisingly, those with open land and access to reliable power. A new study by the mineral rights brokerage firm Texas Royalty Brokers measured AI energy demand across America and reported that Indiana dedicates more electricity to AI than any other state, with its data centers consuming nearly half of the state’s energy output.
A single AI facility in Wisconsin was also found to use a quarter of the Badger State’s electricity.
Texas, which has a massive power grid, now hosts the most AI facilities, with 17 clusters that use 15% of the state’s power generation.
The AI and data center boom will present challenges, at least regionally, Dr. Jim Purtilo, associate professor of computer science at the University of Maryland, told ClearanceJobs.
“We have a national power grid, but each region makes its own projections of likely demand,” Purtilo explained. “They look years ahead to ensure they are building out enough capacity, and also to service or upgrade infrastructure, since there is not much point to new generation if the lines can’t get power to where it is used. Anyone who lives in hurricane country knows issues of the latter.”
However, the dramatic shifts in immigration patterns in recent years impacted the housing market, which in turn broke projections of demand to power those homes.
“As a result, some regions are operating more on the edge than others,” Purtilo added. “When you add high-demand data and AI centers to the mix, then there is a real danger that demand will outstrip capacity. This reality is already reflected in regionally increasing energy costs.”
However, he said we should not necessarily extrapolate demand by drawing a straight line pointing up, even as many entrepreneurs do when they invite investors to buy into their firms that capitalize on expansion.
“My guess is that the demand picture will become more nuanced,” Purtilo continued. “Computer scientists have done an amazing job with new AI technology, but mostly these are based on brute force techniques, hence the demand for so much computation. It was all about being the first with the most. Now forward-thinking scientists are working hard on ‘frugal AI,’ which is to say, finding AI techniques that don’t need the heavy computations.”
May Not Bring A Lot of Jobs
The issue in the short term is where the AI data centers are being built. The centers require large buildings yet employ a relatively small workforce. Unlike a factory that can provide jobs for hundreds of individuals, such facilities employ anywhere from a couple of dozen to around 100 full-time staff for operations, including technicians, engineers, and security personnel. While construction can temporarily involve thousands of workers, the permanent, on-site staff is relatively small for the massive size of the facilities.
At the same time, data centers require space, and unless there is an old shopping mall or shuttered factory, builders look to rural rather than urban options. This is done for other reasons as well.
“While putting data centers in open rural areas avoids building huge substations next to neighborhoods, many rural grids just aren’t designed for multi-hundred-megawatt loads,” warned Lohokare. “Without targeted upgrades, they risk congestion and higher local costs – which is why groups like the Electric Power Research Institute and World Resources Institute emphasize careful regional planning so AI growth doesn’t just shift today’s grid problems onto smaller communities.”
This is the latest disruption to some communities. It should be remembered that electricity to homes in cities and urban centers arrived long before it reached rural communities. It was only following the establishment of the Rural Electrification Administration (REA) in 1935 that efforts began, and it took until the early 1960s for it to be largely completed.
“Rural communities will indeed benefit to some extent from this trend – there will be some economic benefits from increased tax revenues, improved infrastructure and indirect jobs, such as services, construction, local supply chain; in addition to some long-term jobs in operating the data centers themselves,” added Lohokare.
Lessons From History
AI is still in its infancy, and the technology will undoubtedly evolve. What that means is still only coming into focus.
“A hundred years ago, access to telephone technology was spreading fast around the country, and scientists extrapolated to predict that inside a few decades, half the population would need to be employed as telephone operators to route the calls of the other half of the population,” said Purtilo.
“Ultimately, that became true! But not the way people projected,” Purtilo added. “It became true because of the further advance of automatic switching in calls. Each user became their own operator by dialing for themselves. In much the same way, AI will continue to spread rampantly, but how we do it will look very different in just a few years. The technology will grow into the power capacity we will have.”
