On May 19, 2026, the U.S. Energy Information Administration (EIA) — the federal government’s independent energy statistics agency — published new commercial-buildings data showing that energy consumed by data center servers is growing across the nationwide commercial building stock. The finding lands in the middle of an intense public debate over how much electricity the AI build-out actually consumes.

The release matters less for any single number than for its source: this is federal survey data, not a vendor forecast, quantifying how server energy use has expanded within America’s offices, dedicated data centers, and the server rooms tucked inside ordinary commercial buildings.

Executive Summary

EIA’s announcement extends its commercial-buildings statistical program — best known through the Commercial Buildings Energy Consumption Survey (CBECS), the government’s long-running census-style study of how U.S. commercial buildings use energy — to document rising server energy consumption across the building stock. In plain terms: the computers doing the computing inside commercial buildings are drawing a growing share of those buildings’ electricity.

Why it matters: nearly every claim about the ‘AI power crunch’ to date has rested on private-sector estimates from consultancies, utilities, and technology vendors, each with its own methodology and, in some cases, its own commercial interest in the answer. A federal statistical agency measuring the same trend from building-level survey data gives regulators, utilities, and investors a common, disinterested baseline — the kind of number that ends up cited in rate cases, siting decisions, and congressional testimony.

For infrastructure operators, the direction of the data is unsurprising. The significance is that the growth is now visible across the commercial building stock — not only in purpose-built hyperscale campuses, but in the broader population of buildings that house servers.

Federal Numbers Change the Power Debate

Until now, the data center energy conversation has been dominated by projections — analyst decks, utility interconnection queues, and corporate sustainability reports. Projections are arguments; survey data is evidence. EIA’s commercial-buildings program measures what buildings actually consumed, which makes it the closest thing the industry has to a scoreboard. When a .gov dataset says server energy use is growing across the building stock, it becomes much harder for any side of the debate — boosters or critics — to dismiss the trend as hype or alarmism.

That cuts both ways. Utilities seeking rate recovery for grid upgrades, developers seeking permits, and efficiency advocates seeking standards will all now cite the same federal source. Expect this data to surface in state utility commission filings and local zoning fights, where the credibility of the underlying numbers is often the whole battle.

The Hidden Data Center Problem

The phrase ‘commercial building stock’ is doing important work in EIA’s framing. Public attention fixates on gigawatt-scale AI campuses, but a substantial slice of America’s server fleet has historically lived in less visible places: server rooms in office buildings, hospital basements, university closets, and small enterprise data centers. These embedded loads are dispersed, often inefficient, and poorly captured by headline hyperscale statistics.

Growth measured across the whole stock suggests the compute boom is not just a story of a few hundred giant facilities — it is diffused through the built environment. For the efficiency industry, that is a market signal: dispersed, aging server rooms are prime candidates for consolidation into professionally run colocation facilities, which typically achieve far better power usage effectiveness (PUE — the ratio of total facility power to the power that actually reaches computing equipment).

Winners, Losers, and the Grid in Between

The beneficiaries of officially documented demand growth are the companies positioned to serve it: colocation and cloud operators with contracted power in hand, transmission developers, and equipment suppliers across the cooling and electrical chain. Utilities gain justification for capital programs, though they also inherit the political risk of rising rates being blamed on data centers.

The exposed parties are energy buyers competing for the same electrons — manufacturers, electrified transport, and ordinary ratepayers — and any data center developer whose business case assumes cheap, quickly available power. Federal confirmation of demand growth strengthens the hand of grid planners who argue for building ahead of load, but it equally strengthens critics who ask whether that growth should pay its own way. The honest reading of EIA’s data is that it quantifies the trend without settling the policy argument.

Background

EIA has surveyed U.S. commercial buildings for decades through CBECS, producing the government’s authoritative picture of how offices, schools, hospitals, and other non-residential buildings consume energy. Data centers historically registered as a small but disproportionately energy-intensive slice of that stock — buildings that consume many times more electricity per square foot than a typical office.

The context shifted sharply after 2023, when large-scale AI training and inference drove a wave of data center construction and record utility interconnection requests, making data center electricity demand a national policy issue. Against that backdrop, federal measurement of server energy use across the building stock arrives as a reference point both industry and its critics have lacked.

Source: Data center server energy use grows across the commercial building stock — U.S. Energy Information Administration announcement of new commercial-buildings energy data, published May 19, 2026.