On April 15, 2026, Wired reported that the US Energy Information Administration (EIA) has told two senators it plans to develop the first mandatory nationwide survey of data center energy use. The commitment came in a letter dated April 9 from EIA administrator Tristan Abbey, responding to an unusual bipartisan pair: Democrats’ Elizabeth Warren and Republicans’ Josh Hawley, who had asked the agency in March to start quantifying the industry’s power footprint. If the survey goes through, it will be the first time the federal government compels data center operators to report how much energy they consume.
Warren’s framing was blunt: “Americans deserve to know how much energy data centers are sucking up and what that’s doing to their utility bills,” calling the survey “an important first step towards holding data centers accountable.” For the AI industry, the message is that the electricity behind model training and inference has officially entered federal statistics — and federal accountability.
The Timeline: From a Senator Letter to a Mandatory Survey
The actual schedule is more gradual than the headlines suggest. In late March, almost in step with the senators’ letter, the EIA announced a first pilot phase covering Texas, Washington state, and the Washington, DC–Northern Virginia metro area, with 196 companies identified. A second pilot will add at least three more states, names still unannounced. Both pilots are slated to wrap up by late September 2026, after which development of the mandatory nationwide questionnaire begins. No date is set for the final rollout. Abbey’s letter describes the pilots as “a necessary step in the methodical development of a nationwide mandatory survey.”
Read that as a track being laid, not a rule taking effect tomorrow: pilots validate the questionnaire design and data quality, then the national version takes over. For the 196 companies already identified, joining the pilot is effectively a dress rehearsal for future reporting obligations.
What the Pilot Surveys Collect
Each participating company reports metrics for just one facility, but the field design is thorough: annual electricity use, behind-the-meter power generation, data center classifications, cooling systems, square footage, and IT efficiency metrics, with questions tailored by location. The intent behind the fields is clear — not just “how much power,” but “where it comes from and how efficiently it is used.” The behind-the-meter item points directly at the trend of developers building their own gas-fired generation on site.
Notably, the EIA declined to explain how it selected the 196 companies, and would not give a timeline for the national rollout. The statistical agency’s reticence reflects how sensitive this data is: for a data center operator, energy consumption is close to a proxy for capacity and customer mix.
Why Now: AI Power Demand Meets Utility Bills
Three pressures are peaking at once. First, AI training and inference demand has ballooned data center electricity use, and developers chasing speed increasingly build on-site gas generation. Second, rising residential utility bills have triggered community pushback, stalling new projects in several states. Third, the litigation has started: the NAACP just sued xAI over allegedly unpermitted gas turbines at a Mississippi facility. Without official numbers, these policy fights run on competing estimates. The survey’s first job is to turn “how much power data centers actually use” from a trade secret into a public statistic.
What It Means for Operators and Policymakers
For operators, this is a compliance signal worth acting on early: the pilots may be statistical exercises, but the national version is mandatory reporting, and internal energy-data governance — measurement, retention, submission pipelines — becomes a real operating cost, feeding numbers that end up in public statistics. For policymakers, the first federal-grade dataset changes the evidence base for rate reviews, siting permits, and subsidy design, and gives state regulators a citable baseline. For developers and product teams, the practical takeaway is simpler: capacity cost models should start accounting for energy reporting and local grid constraints. Power is graduating from an infrastructure footnote to a product-level decision variable.
Sources
- The US Government Is About to Ask Data Centers How Much Power They Use — Wired
- Feds will require data centers to show their power bills — TechCrunch
- Data centers, get ready: the Senate wants to see your power bills — TechCrunch
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.