A physical world
Compute depends on semiconductors, buildings, water, cooling and a grid. Demand can grow even as chips become more efficient because new applications spread faster than efficiency gains reduce consumption.
Start with the right number
The International Energy Agency’s updated 2026 forecast projects roughly 950 TWh of global data-centre electricity use in 2030, up from approximately 485 TWh in 2025. The estimate is for data centres overall, not AI alone; IEA expects AI-focused facilities to grow faster.
Local costs are uneven
A data centre in a region with clean surplus electricity creates a different challenge from one competing for scarce power or water. Local grid congestion, households’ bills and delayed industrial connections can matter more than global averages.
Policy can keep up
Publish site-level energy and water accounting where safe and appropriate, link development permission to grid planning, and compare genuine climate applications against the resources they consume. AI innovation and environmental accountability can — and should — proceed together.
Demand local, auditable resource accounting instead of slogans about “green AI”.