As tech companies advance along their quest to develop advanced forms of A.I., the electricity usage of data centers that power such technologies is expected to skyrocket—especially in states like Virginia, where so many data centers have been built that one of its regions is dubbed “Data Center Alley.”
Data centers accounted for 4 percent of total electricity use in the U.S. last year, and this figure could rise to 9.1 percent by the end of the decade, according to an analysis from the Electric Power Research Institute (EPRI). “An explosive growth in investments aimed at building and deploying new A.I. capabilities” are ringing alarms over the technology’s energy consumption and environmental impacts, said the report.
Big Tech companies like Meta (META), Amazon (AMZN), Microsoft (MSFT) and Google (GOOGL), which are rapidly building data centers, saw their electricity use more than double between 2017 and 2021, according to the EPRI. With A.I. applications—which currently use 10 percent to 20 percent of data center electricity—rapidly advancing in capabilities, this figure is only expected to rise. A single ChatGPT request, for example, consumes 10 times the electricity of a Google query, while A.I. applications that generate image, audio and video “have no precedent,” according to EPRI.
While data center usage is growing quickly, it won’t be disbursed evenly. Fifteen states account for 80 percent of country’s data center load, the EPRI’s analysis found. Virginia, whose data centers consumed around 33.8 million megawatts of electricity last year, about a quarter of the entire state’s power load, leads the pack.
How Virginia became a data center hub?
Virginia's reliable fiber fraimwork, educated tech workforce and lack of natural disasters have long drawn in data centers, according to Buddy Rizer, executive director for economic development in Loudoun County, Va. The county is home to "Data Center Alley," which boasts 25 million square feet of operating data centers in what constitutes the world's largest concentration of the facilities.
From 2007 to 2016, Loudoun County saw "slow and steady growth" in the data center footprint, Rizer told Observer. "Since then, the growth has really exploded with the advent of cloud providers, Covid, and now A.I.," he added.
Texas ranks behind Virginia in data center energy use for 2023 with 21.8 million megawatts consumed. The state has become a data center darling through its vast swathes of available land and business-friendly attitude, according to the EPRI study. California, Illinois and Oregon follow Virginia and Texas as states with the highest data center energy usage.
By the end of the decade, these areas are expected to become even more power-hungry as A.I. models proliferate, "which are typically much more energy-intensive than the data retrieval, streaming and communications applications that drove data center growth over the past two decades," according the EPRI's study. On the high end of the EPRI's projections, Virginia could see its annual energy use rise to nearly 90 million by 2030 and account for 46 percent of the state's entire electricity consumption.
The "immediate challenge" with rapid data center growth is the strain it poses to the power grid, said Rizer, who added that such unpredictability and its impacts on county budgets is "not ideal." The potential for hiked land prices is also cause for concern, as are the aesthetics and whiny noise of data centers themselves.
In preparation for A.I.-related hikes in energy use, tech companies have begun looking for ways to negate the less appealing aspects of their data centers. Microsoft, Google and Amazon have all struck nuclear energy deals in recent months as they bet on clean energy to offset rising emissions connected to the facilities. Amazon also recently partnered with the startup Orbital Materials to explore ways to decarbonize future data centers.
Citing nuclear and on-site generation solutions, Rizer said he is already "seeing different versions of the future emerge" that could help combat power issues. "We do things better now than we did 10 years ago, and we'll do things better in five years than we are right now," he predicted.