Think of all the stuff you use electricity for that you didn't 20 or 25 years ago — all those devices, maybe even your car — and yet electricity use has barely budged this century. In 2000, the country used about 4 million gigawatt-hours of electricity, according to the International Energy Agency; in 2022, it used about 4.5 million GWh, a growth rate of about 0.5%.
In some ways, the purpose of current U.S. climate policy is to reverse this trend. Only about a fifth of all energy produced in the United States is electrical. Removing carbon emissions from transportation, heating and industry will require first converting all of those industries from running on combusted hydrocarbons to running on electricity — while at the same time, of course, working to make electricity generation carbon-free.
All that is to say, we’re definitely going to be using more electricity. Today, if you ask any utility, electricity market organization, or anyone working on energy generation and transmission, they’ll tell you we’re in for an era of load growth.
“For a long period of time, we could balance out additional demand with efficiency improvements,” Xan Fishman, energy policy director at the Bipartisan Policy Center, told me. “Recent forecast are showing we’re going to need a lot more electricity.”
When GridStrategies LLC looked at documents grid planners filed with federal regulators, it found that their aggregate five-year load growth forecasts had gone up from 2.6% in 2022 to 4.7% last year, while their forecast for peak demand, i.e. the maximum amount grids plan on having to be able to provide, had shot up by 18 GW. That’s the equivalent of about 35 gas-fired power plants running on full blast.
In New England, for example, ISO-NE is forecasting 2.4% annual growth over the next 10 years, while its winter peak demand will grow by 3% per year thanks largely to electrifying transportation and heating; that, in turn, is largely thanks to aggressive decarbonization mandates in the region’s constituent states.
When PJM Interconnection, the 13-state East Coast and Midwest electricity market, was making its load forecast, it specifically called out Intel’s CHIPS Act-funded facility under construction outside Columbus, Ohio; the electrification of New Jersey ports funded by the Inflation Reduction Act; and planned data centers in Maryland and Virginia as notable examples of increased load generation. For AEP, the utility serving Columbus, the forecast peak summer load in 2030 has gone from about 23.5 GW to 26 GW, compared to around 21 GW in 2023. Dominion, the utility serving Virginia and the booming Loudon County datacenter complex, forecast annual load growth of around 5% over the next decade.
To get a sense of how tremendous that is, when the energy system researchers with Princeton University’s REPEAT project wanted to project how much electricity consumption would have to increase annually to reach net zero by 2050, it turned out to be “only” 2.4%. Virginia is planning load growth at twice that rate just to feed electrons to its data centers.
“When you’re talking about a data center or a three-shift, seven-day-a-week manufacturing process, that’s far less manageable” than, say, electric cars, David Porter, vice president of electrification and sustainable energy strategy at the Electric Power Research Institute, told me. EVs can be powered at specific times based on demand for electricity across the grid, or by a distributed energy resource like residential solar and batteries. To power energy-hungry manufacturing processes, though, requires the kind of consistency that only fossil fuels and nuclear (or naturally limited renewables like hydropower) have historically been able to provide.
There’s no better example of the tension between electrification and emission reductions than in Georgia, where the state’s main utility Georgia Power has said that its estimates for load growth between 2023 and 2031 had jumped up from less than 400 megawatts to 6,600, a 17-times increase. The utility attributed this forecasting hike to “rapid economic expansion and an unprecedented increase in the demand for energy to the state,” including electric vehicle and battery manufacturing facilities, which the Biden administration has done so much to boost demand for and encourage their construction in the United States.
The utility also said that to serve this load growth, it would have to add new renewable resources, acquire power from other utilities and generators, and build new gas power plants, which immediately raised the ire and suspicion of green groups. The Sierra Club described the request as “shocking.”
But proponents of climate action shouldn’t necessarily despair at this new load, Fishman told me. “It’s really easy to decarbonize if you stop building stuff,” he said. “But [Americans] would likely keep buying stuff, and that stuff would be built elsewhere, quite likely with greater emissions intensity.”
In other words, “a resurgence of American manufacturing might lead to more U.S. emissions than in a scenario where we aren’t increasing our manufacturing base,” Fishman told me, but it’s “highly likely to reduce global emissions.” That’s because even now, U.S. electricity is cleaner than electricity in, for example, China, which is still heavily reliant on coal. (According to the IEA, 63% of China's electricity comes from coal burning, compared to 20% in the United States.)
Data centers, meanwhile, are expected to account for 6% of total electricity demand in the U.S. by 2026, according to the IEA, up from about 4% in 2022. And the AI ones will eat up even more: A ChatGPT query is about nine times as energy intensive as a Google search, according to the IEA. If generative artificial intelligence grows at anywhere near the rate that its proponents expect, it will lead to hefty increases in electricity demand, both from manufacturing the chips needed to power the systems and the electricity to power them. One example is Silicon Valley Power, a utility serving, well, Silicon Valley, which forecast load to double by 2035, “primarily” due to data centers’ demand for electricity.
But there may be some reason for skepticism about these load growth projections from data centers, Jon Koomey, a veteran information technology and energy researcher, told me. The particularly energy intensive large language models may not win out as a business, which would slow the growth in data center electricity demand, he said. And even if data centers continue to grow, they could also get far more efficient in how they use electricity — and might just end up using less than what they ask for from utilities.
“You don’t want to get caught short,” Koomey said, explaining why requests for power will be biased on the high end. “There’s an incentive for everyone to request more.”
But still, it’s no surprise that the companies at the heart of the data center boom — Google, Microsoft, and OpenAI — have shown an interest in finding ways to match that constant electricity demand with non-carbon-emitting power. Their facilities need to be powered 24/7, which existing renewable sources largely struggle to provide. (It’s neither windy nor sunny 100% of the time.) This has led to a flurry of investment and dealmaking by these companies to develop and procure “clean firm” resources. Google has a deal with Fervo, the enhanced geothermal startup, to purchase power generated by its operation in Nevada, while Microsoft signed an agreement with Constellation to purchase nuclear-generated electricity for its Virginia data centers to complement its existing renewable power. Silicon Valley Power also said in its planning documents that it’s looking to acquire more geothermal resources. And OpenAI’s Sam Altman has invested in a fusion company.
“If we want to grow our manufacturing base we need the energy to make that work, we need to get that energy to those new manufacturing plants,” Fishman said. “It would be bad if we had a bunch of companies who said, ‘We want to build a factory,’ and can’t because they don’t get enough electricity.”
REPEAT Project
SEIA and Wood Mackenzie