AI Energy and Environmental Impact

What AI Energy and Environmental Impact is

Training a frontier model consumes substantial energy once; serving it consumes energy continuously, and at scale inference dominates lifetime footprint. Data centres also draw significant water for cooling.

How it works

Reduction levers include efficient architectures, quantisation, smaller task-specific models, better utilisation, carbon-aware scheduling and locating capacity where clean power is available. Published figures vary widely in methodology, so specific numbers should be checked against primary sources.

Why it matters

Energy availability is becoming a real constraint on AI deployment and a growing part of corporate sustainability reporting.

Common uses

• →Sustainability reporting
• →Model selection on efficiency grounds
• →Data centre siting