U.S. data-center
energy use through 2028

Why the report was produced

The Energy Act of 2020 called for the U.S. Department of Energy to make available an update to Berkeley Lab’s earlier national data-center energy study. Berkeley Lab published the new report on December 19, 2024, with Arman Shehabi as lead author and a multidisciplinary author team.

The report reconstructs historical electricity consumption back to 2014 using prior research, equipment shipment data and other sources. It then provides a scenario range through 2028 based on recent trends and uncertainty in future deployment.

Sources for this sectionBerkeley Lab report page ↗

The estimates most often cited

DOE’s release reports that data centers used about 4.4 percent of U.S. electricity in 2023, compared with 1.9 percent in 2018. For 2028, the report provides a range of approximately 6.7 percent to 12 percent of total U.S. electricity consumption.

The width of the 2028 range is not a weakness to be hidden. It communicates uncertainty about AI adoption, new construction, equipment efficiency, utilization, cooling systems and the pace at which announced capacity becomes operational.

Sources for this sectionBerkeley Lab report ↗U.S. Department of Energy release ↗

Estimate, scenario and facility claim are different

Historical national estimates combine multiple data sources because no single public registry measures every U.S. data center. Future values are scenarios, not a promise that a particular percentage will occur. A facility announcement is further removed: planned megawatts, connected capacity and actual load can differ substantially.

For that reason, a national percentage should not be allocated to a Virginia facility without local evidence. Utility resource plans, transmission proceedings, interconnection records, local approvals and company disclosures are more appropriate for project-level load questions.

• National demand does not identify a facility’s load.
• Air permits usually cover on-site emission sources, not total grid consumption.
• A scenario range should remain a range when cited.

The institutional authority is stronger than a borrowed endorsement

The report lists Arman Shehabi, Sarah Smith, Alex Hubbard, Alexander Newkirk, Nuoa Lei, Md AbuBakar Siddik, Billie Holecek, Jonathan Koomey, Eric Masanet and Dale Sartor as authors. Their names should be presented as authorship of the cited report, not as participation in Public Evidence Project.

The strongest credibility statement is therefore factual: this page independently summarizes a Berkeley Lab report prepared in response to a Congressional mandate. No university logo, laboratory mark or implied partnership is needed.

How the national report connects to Virginia

Virginia is one of the clearest places to observe the local infrastructure consequences behind the national trend. But the Berkeley Lab report and the Virginia JLARC review answer different questions: one estimates U.S. demand; the other examines state-level development, grid requirements, water, land use and local impacts.

A source-linked evidence page should connect the two without merging them. National growth establishes context. Virginia documents establish what has been proposed, permitted, built or forecast in the Commonwealth.