The shift
AI services are often described as weightless. Their computers are not. Data centres use electricity, cooling systems, land and capital-intensive hardware. The IEA’s 2026 central projection puts global data-centre electricity use at around 950 TWh in 2030, approximately double 2025 levels. This projection covers data centres broadly, not AI alone.
The opportunity
AI may optimize grids, accelerate materials research, and help forecast weather and energy demand. Efficiency per task is also improving. Neither benefit eliminates the need to count the energy used by more tasks and expanding infrastructure.
What to monitor
Track actual electricity demand, the carbon intensity of the marginal power used, water stress in host communities, supply-chain impacts, and new generation and transmission investment. Compare forecasts to real measurements as they arrive.
The public-interest response
Make large-compute impacts transparent by region, price grid upgrades fairly, protect water-stressed communities, and require credible accounting across the lifecycle — while supporting AI applications with measurable climate benefits.