IEA Estimates AI Used 155 TWh in 2025 as Grid Strain Risks Rise
Updated
Updated · Our World in Data · Jul 20
IEA Estimates AI Used 155 TWh in 2025 as Grid Strain Risks Rise
3 articles · Updated · Our World in Data · Jul 20
Summary
AI-focused data centers consumed about 155 TWh in 2025—roughly 0.5% of global electricity—within total data-center demand of 485 TWh, or 1.5% of world generation, the IEA estimates.
That global share masks sharp local concentration: data centers use about 5% of U.S. electricity, more than 20% in Ireland, and over one-quarter in parts of Virginia, raising risks to grids and local power prices.
Individual AI use remains comparatively small. Google put a median Gemini text query at 0.24 Wh, while other estimates place a typical chatbot query near 0.3 Wh; agentic reasoning tasks can reach about 50 Wh.
By 2030, the IEA base case sees data centers rising to 3% of global electricity, with most growth coming from AI facilities, though future demand will depend heavily on chip efficiency and infrastructure buildout.
The report says climate impact will hinge less on electricity volume alone than on how clean the supplying grids are, since coal-heavy power makes the same AI workload far more emissions-intensive.
Beyond its carbon footprint, is AI's massive water and e-waste problem the real environmental threat?
Are we choosing between AI progress and affordable electricity as communities reject new data centers?
With power grids and supply chains at their limit, is the AI boom heading for a physical collapse?
AI’s Energy Surge: How Data Centers Will Double U.S. Electricity Demand by 2030—and What It Means for Grids, Communities, and Climate
Overview
Between 2025 and 2030, the rapid growth of Artificial Intelligence is driving a dramatic increase in energy demand, as major technology companies invest heavily in AI and accelerate data center construction worldwide. This surge is led by key players like AWS, Microsoft, Google, Meta, and Oracle, who together have deployed hundreds of data centers. The large-scale rollout of advanced AI systems, especially Agentic AI, is creating new energy bottlenecks in both computation and communication. As a result, the industry is entering a costly era where energy efficiency, infrastructure upgrades, and sustainable solutions become critical challenges for continued AI expansion.