US electricity consumption by data center servers is projected to surge through 2050, driven by rising demand for artificial intelligence and cloud computing, the US Energy Information Administration said on Tuesday.
EIA said electricity consumption by servers could range from 446 to 818 billion kWh by 2050, depending on assumptions about electricity demand and server efficiency. The upper end of the range reflects stronger growth in server installations and higher power consumption in the agency's High Electricity Demand scenario.
Standalone data centers are expected to account for most of the increase. In the high-demand case, servers in standalone facilities alone consume 581 billion kWh of electricity by 2050.
Servers accounted for an estimated 7% of commercial-sector electricity consumption in 2025, according to the report. By 2050, that share is projected to range from 22% to 33% across the agency's scenarios.
The rapid growth in data centers is also expected to push commercial building electricity intensity to its highest level since 2003. The commercial sector's electricity intensity is measured in kWh per square foot. The EIA projects that electricity consumption per square foot in commercial buildings will surpass the prior high of 14.9 kWh between 2031 and 2032.
Beyond computing power, cooling systems are expected to drive significant additional electricity demand. The agency estimates that cooling requirements in data center space are up to 2.9 times more energy-intensive than in non-data center commercial space. Under the High Electricity Demand case, electricity use for space cooling is projected to be 84 billion kWh higher in 2050 than in the agency's baseline scenario.
The EIA said it assumes server electricity demand remains relatively constant throughout the day because data center operations run continuously.
In its baseline outlook, the agency assumes server efficiency improves after 2040, reducing average operational power draw by 10% every three years beyond historical trends. No such efficiency gains are assumed in the high-demand case, which also projects a larger share of energy-intensive AI servers.