AI Data Centers Now Consume 0.5% of Global Electricity—and Growing Rapidly

Summary

As artificial intelligence has been adopted at unprecedented speeds, concerns about its energy footprint have intensified. A new analysis finds that AI-focused data centers consumed approximately 0.5% of the world's electricity in 2025, with total data center electricity use reaching around 485 terawatt-hours (TWh)—equivalent to Germany's annual electricity generation and representing 1.5% of global electricity consumption. The article identifies three key concerns: environmental impact through increased carbon emissions and hampered decarbonization efforts, local grid strain and electricity price pressures in affected communities, and whether energy consumption could become a limiting factor for AI's continued expansion.

The analysis reveals that AI data centers currently represent about one-third of total data center electricity demand, with non-AI data centers (supporting email, streaming, banking, maps, and other digital services) consuming the remaining two-thirds. Energy figures include electricity for both training and running AI models—called inference—plus supporting infrastructure like cooling and lighting. Tech companies rarely publish detailed training energy data, but available estimates suggest inference dominates overall consumption. At the current global scale, this represents 0.1% to 0.2% of total primary energy consumption.

Projections show rapid growth ahead. According to the International Energy Agency's base-case scenario, data center electricity demand is expected to reach 3% of global electricity by 2030, with AI-focused facilities growing substantially and eventually matching non-AI data center consumption. However, some analysts argue that even the IEA's forecasts may be conservative, suggesting actual growth could exceed current projections. The article notes significant uncertainty around these estimates, particularly regarding how quickly AI demand will accelerate.

• Inference (running deployed models) rather than training likely accounts for the majority of AI's energy consumption, though tech companies rarely disclose detailed training energy figures

Editorial Opinion

This analysis reveals a critical blind spot in AI's rapid expansion: while the technology's capabilities dominate headlines, its infrastructure costs remain largely opaque to the public. The finding that AI currently consumes only 0.5% of global electricity may seem reassuring, but the trajectory toward 3% by 2030 deserves urgent attention. As AI companies race to build larger models and expand capacity, policymakers and industry leaders must prioritize transparency on energy consumption and invest aggressively in sustainable data center infrastructure—otherwise, the environmental and economic costs of AI could become prohibitive.