Technology
Hyperscale data centers now account for 2% of US power consumption
A new analysis quantifies the electricity demand of large data centers, revealing they consume as much power annually as an entire state.
Ars Technica
The rapid expansion of hyperscale data centers to support artificial intelligence workloads has outpaced planning for clean energy supply and grid management. A recent study measuring energy consumption across these facilities from May 2024 through April 2025 provides the first concrete benchmark for tracking this growth.
Measuring the landscape#
Researchers began with 675 potential facilities identified through commercial databases, news reports, company announcements, and satellite imagery. After filtering for facilities with confirmed electrical capacity and verifiable infrastructure—cooling systems, power distribution equipment—they identified 403 operational data centers with an average capacity of roughly 40 megawatts. The analysis excluded 272 suspected large facilities due to lack of verification, making these findings a conservative estimate.
The uncertainty extends to actual power draw. Conventional data centers typically operate at approximately 48% of their maximum capacity, while AI training operations run hotter—closer to 70% utilization.
The scale of consumption#
Using these utilization rates, researchers calculated annual electricity use between 68 and 99 terawatt-hours. This range translates to 1.5–2.2% of all US electricity consumption—roughly equivalent to the total annual power usage of Wisconsin or Indiana.
The study acknowledges it cannot distinguish between AI-specific workloads and traditional data services within these facilities, introducing some ambiguity into the results. Nonetheless, the analysis offers a baseline for monitoring how data center energy demands evolve as the industry continues its AI pivot.
Why it matters#
The concentration of electricity demand in hyperscale facilities has outpaced regulatory and infrastructure planning. Without coordinated action on grid capacity and renewable energy deployment, rapid further growth could strain regional power systems and increase pressure on ratepayers.

