U.S. Data Center Power Demand Report

By Laila Kearney

NEW YORK (Reuters) – U.S. data center power demand could nearly triple in the next three years, consuming as much as 12% of the country's electricity due to the industry's shift towards artificial intelligence, according to an unpublished report backed by the Department of Energy.

The report from the Lawrence Berkeley National Laboratory, expected to be released on Friday, aims to help the U.S. power industry and government understand the impact of Big Tech's data-center demand on electrical grids, power bills, and climate.

By 2028, annual energy use by data centers could reach between 74 and 132 gigawatts, representing 6.7% to 12% of total U.S. electricity consumption. Currently, data centers account for just over 4% of the country's power load.

Avi Shultz, director of the DOE's Industrial Efficiency and Decarbonization Office, noted that this indicates a significant frontier for growing energy demand in the U.S. Despite significant industry growth from 2010 to 2016, data center power demand remained flat, thanks to increased efficiency and a transition to larger cloud-based facilities.

However, the deployment of GPU-accelerated servers starting in 2017 led to more than a doubling of the sector's power usage over six years. AI's need for powerful chips and cooling systems is the main driver behind the projected growth in data centers.

When a similar report was last issued in 2016, AI servers represented only about 2% of total server energy use. Lead researcher Arman Shehabi emphasized the drastic changes seen in the industry since then and recommended annual or biannual reports to better track data center trends.

The estimates are based on calculations regarding electricity use from installed GPUs and other data-center IT equipment, complemented by data from market research firms and power-sector executives.

Shehabi and researchers advocate for greater transparency in the industry by collecting anonymous data about electricity usage, compute capabilities, and workloads. Understanding energy usage dynamics may uncover efficiency opportunities. Additionally, the report encourages further research and development of energy-efficiency strategies for the expanding AI data center sector.