EIA's data-center pilot is a measurement story before it is a policy story, after the agency said it launched three voluntary pilot field studies in Texas, Washington state, and Northern Virginia/Washington, DC, identified 196 companies, and asked for energy sources, electricity consumption, site characteristics, server metrics, and cooling systems. [1]

That list is the news because data-center politics often begins with an accusation, either that AI companies are breaking the grid or that critics are blocking progress, while EIA's questionnaire begins somewhere less theatrical by asking what is in the building, how it is cooled, where the electricity comes from, and how the servers are counted. [1]

The agency's later analysis explains why the pilot matters, saying AEO2026 reports data-center server electricity separately from broader commercial computing and moves servers out of the haze of generic office load. [2]

The pilot does not yet answer the national question, since it is voluntary, covers three regional tests, and does not print a measured United States total for 2026 data-center electricity use, but that limitation is precisely why the source belongs in the paper: a fight about rates, grid upgrades, cooling water, and siting cannot be sharper than the instrument used to measure it. [1] [2]

The public argument wants a verdict, while EIA is still building a ruler, which is less satisfying than blame and more useful than another abstract sentence about AI electricity demand because before the country can decide who pays for AI factories, it needs to know what those factories consume. [1] [2]

-- DARA OSEI, London