Data centers could double their current share of U.S. power by 2030, consuming 9 to 17 percent of electricity generation by decade’s end as companies plan more and larger projects, according to a new analysis.
The findings released Thursday by the research organization Electric Power Research Institute reflect how rapidly the outlook for artificial intelligence is changing because of company announcements in the past year and a half. The new estimates on electricity demand are roughly 60 percent higher than EPRI anticipated two years ago.
We’re seeing an “unprecedented amount of investment in data center construction and project planning,” said Thomas Wilson, principal technical executive at EPRI, in an interview.
Data centers now use roughly 4.5 percent of U.S. electricity.
Projected growth in power use is slated to be uneven throughout the country. In Virginia, one of the world’s main data center hubs, the technology could consume 41 percent to 59 percent of electricity by 2030, up from 25 percent today, EPRI said.
In seven other states — Arizona, Indiana, Iowa, Nebraska, Nevada, Oregon and Wyoming — data centers could consume more than 20 percent of power by decade’s end, according to the report.
Natural gas is expected to dominate new U.S. power generation for AI in the near term. A separate study from research firm Cleanview this month found that a third of current planned data centers are behind the meter and of those, gas is dominant despite renewable energy pledges by many companies.
If more technology companies and data center developers commit to carbon-free energy targets, it could help push more renewable and battery projects online, “with new nuclear generation coming online when possible,” EPRI said.
The analysis is one of several projecting electricity demand, including a 2024 study from Lawrence Berkeley National Laboratory concluding that data centers could use 12 percent of U.S. power by 2028. The lab is expected to update its outlook this year.
While power demand forecasts vary depending on the time period studied and modeling assumptions, many have increased in the past year as more companies plan projects and advance massive AI complexes. Projects announced since 2024 include Colossus II, a gigawatt-scale complex from Elon Musk’s xAI in Tennessee, and plans from Fermi America — a company led by former Energy Secretary Rick Perry — to build a large AI complex powered by gas and nuclear reactors in Texas.
There are multiple factors that could swing forecasts, including the pace at which facilities ramp up, potential project delays and cancellations, and labor constraints.
“There’s a lot of uncertainty about how much of this actually gets completed,” Wilson said. A report this week from Sightline Climate found about one-third of data centers scheduled to open this year may be delayed or never get built because of challenges such as limitations on power supplies and grid equipment shortages.
Historically, efficiency gains have often helped offset growth in electricity demand. However, the scale and pace of video and image-intensive AI applications “are growing so quickly that efficiency improvements may not fully offset demand in the near term,” EPRI said in a fact sheet on the report.
Additionally, any breakthrough in server efficiency would take time to have an effect on electricity use, considering long-term planning with supply chains, said Wilson.
If a much more “efficient server were to come along, they would have to scale up really rapidly to make much of an impact [on power demand] through 2028,” he said.
EPRI examined state-level data and announced projects to generate three scenarios for electricity growth.
The least aggressive assumes that most data center projects under construction become operational by 2030, along with a quarter of initiatives in advanced planning. The most aggressive scenario — tied to the 17 percent forecast — considers that all under construction and advanced projects come online, along with 30 percent of those in the early planning stage.
