United States data centers consumed approximately 176 terawatt-hours of electricity in 2023, according to International Energy Agency data, representing roughly 4% of total US electricity consumption. The IEA projects that data center electricity demand in the US could reach between 325 and 400 terawatt-hours annually by 2030, driven primarily by the rapid expansion of artificial intelligence training and inference workloads.
More than 4,500 data centers currently operate in the United States, making it the most data-center-dense country in the world by a significant margin. The Northern Virginia corridor alone hosts the largest concentration of data center capacity globally, consuming a substantial portion of the state's grid capacity, which has driven utility providers to accelerate transmission infrastructure planning.
Goldman Sachs Research has warned that the pace of data center power demand growth presents near-term inflation risks for electricity consumers in affected regions. In areas where large hyperscale operators have secured power purchase agreements ahead of other commercial and residential users, grid operators face the challenge of building new generation and transmission capacity quickly enough to avoid supply constraints.
Power purchase agreements between hyperscale data center operators and utilities have become more complex and longer-term in response to the sustained demand signal. Operators including Google, Microsoft, Meta, and Amazon have signed 10-to-20-year PPAs for nuclear, solar, and wind generation specifically to power data center campuses.
The shift toward liquid cooling in AI-optimized data centers is changing power density calculations. Rack-level power densities that previously ran at 15 to 20 kilowatts per rack are now reaching 100 to 200 kilowatts per rack in AI training clusters, requiring both more power infrastructure and different cooling architectures.
Source: International Energy Agency -- https://www.iea.org/reports/electricity-2024