The International Energy Agency projects global data centre electricity consumption will reach approximately 945 terawatt-hours by 2026, nearly double the 2022 level of roughly 460 TWh, driven by rapid AI workload deployment that is far more energy-intensive per computation than traditional cloud computing. In the United States, data centres account for approximately 4% of total national electricity consumption today, with that share projected to rise to 6%-8% by 2030 as AI model training and inference infrastructure continues to scale.

The IEA data shows that the power usage effectiveness of the hyperscaler-class facilities being built in 2025 and 2026 has improved significantly compared to legacy enterprise data centers, with best-in-class US facilities achieving PUE ratios near 1.1. However, the sheer scale of new construction is more than offsetting efficiency gains. A single large AI training cluster can require 100 megawatts or more of continuous power draw, equivalent to the peak electricity consumption of a small US city.

The US holds the largest concentration of data centre capacity globally, with significant clusters in Northern Virginia, Dallas-Fort Worth, Phoenix, and increasingly the Southeast US including South Carolina, Georgia, and Tennessee. The IEA notes that the combination of hyperscaler demand and stricter grid connection timelines means data centre operators increasingly need to plan power infrastructure 3-5 years ahead of operational needs, which is driving the nuclear and long-term renewable contracting activity visible in 2025-2026 deal announcements.

Source: International Energy Agency -- https://www.iea.org/energy-system/buildings/data-centres