AI's Electricity Footprint: Data Centers Consume 0.5% of Global Power in 2025
Key Takeaways
- ▸AI data centers consumed approximately 0.5% of global electricity in 2025, or 0.1–0.2% of total primary energy consumption
- ▸AI-focused facilities currently use one-third of total data center electricity, but are projected to match non-AI data centers by 2030 due to exponential growth in inference workloads
- ▸Energy consumption includes server operation plus cooling, lighting, and infrastructure; excludes end-user device power and cryptocurrency mining
Summary
According to analysis based on International Energy Agency (IEA) data, artificial intelligence consumed approximately 0.5% of the world's electricity in 2025—equivalent to roughly one-third of all data center consumption. Global data centers used around 485 terawatt-hours (TWh) of electricity, comparable to Germany's annual generation. The article breaks down consumption between non-AI data centers (two-thirds of total) and AI-focused facilities (one-third), examining both training and inference workloads as the primary drivers of energy demand.
The analysis reveals that AI data centers, while currently consuming less than traditional data centers in aggregate, are experiencing the fastest growth. IEA base-case projections expect data center electricity demand to reach 3% of global supply by 2030, with AI facilities accounting for roughly half of that consumption. The projections carry significant uncertainty, with some analysts arguing that the IEA remains conservative in its AI demand growth forecasts. On a per-query basis, the energy cost of individual AI operations remains modest, though aggregate consumption continues to climb rapidly as usage scales.
- IEA projects data center electricity demand will reach 3% of global supply by 2030, with significant uncertainty around actual AI growth rates
Editorial Opinion
As AI's energy appetite accelerates, the industry stands at a critical juncture between growth and sustainability. While current consumption remains a modest fraction of global electricity, the trajectory is steep—and unlike other infrastructure buildouts, AI's energy demands could concentrate in geographic regions with limited grid capacity. Tech companies and policymakers must move beyond tracking consumption metrics to embedding efficiency into model architecture and data center operations; treating energy as a second-order concern risks creating artificial bottlenecks that slow AI development and exacerbate climate pressures.
