Energy and AI

Why it matters

The first authoritative attempt by a major energy body to quantify AI's global electricity footprint under multiple scenarios — moving the debate out of anecdote and into modelled ranges that policy can act on.

Key findings

• Global datacenter electricity consumption is projected to roughly double by 2030, with AI workloads the dominant marginal driver.
• The largest demand growth is concentrated in a small number of jurisdictions, creating localized grid stress even where national averages look manageable.
• AI is a plausible net-positive contributor to emissions reduction only under scenarios that also constrain the carbon intensity of its own infrastructure — an outcome not on the current trajectory.

Relevance to AI and sustainability

Establishes an evidence base for whole-system accounting of AI: whatever efficiency gains AI delivers downstream must be netted against the infrastructure load documented here. Without that netting, ‘AI for climate’ claims remain unaudited.