Chapter 08 · Finance, data & evidenceData, Technology & Verification Systems
Artificial intelligence (AI)
Definition
The development and use of computer systems able to perform tasks that typically require human intelligence — such as prediction, pattern recognition, language and decision-making. In sustainability contexts, AI spans machine-learning models for forecasting and optimisation and, more recently, large generative models whose training and operation drive data-centre electricity demand.
References
415 TWh (2024, ~1.5% of global electricity); base case ~945 TWh by 2030; sensitivity ranges; accelerated-server growth.
2025 consumption 485 TWh rising to ~950 TWh by 2030; per-query efficiency gains vs energy-intensive new use cases; disclosure recommendations.
Overview
What it means
AI's sustainability relevance runs both ways. As a tool, it supports emissions monitoring from satellites, grid balancing for variable renewables, climate and biodiversity modelling, and supply-chain traceability. As a footprint, the data centres behind AI consumed about 415 TWh of electricity in 2024 (1.
5% of global use), and the IEA projects this to roughly double to ~950 TWh by 2030, with AI-focused facilities growing fastest.
How it is used
Companies apply AI to energy management, precision agriculture and ESG data processing; regulators and standard-setters are beginning to require disclosure of data-centre energy and water use. The IEA now publishes dedicated energy-and-AI analysis, and grid operators plan around data-centre load growth.
Why it matters
AI could accelerate decarbonisation and environmental monitoring, or add a major new electricity load that strains grids and prolongs fossil generation — most likely both. Managing that trade-off (efficiency gains vs surging use, disclosure vs opacity) is now a live energy and sustainability policy question.
Definitions and controversy
Energy-use estimates carry wide uncertainty because operators disclose little; the IEA publishes scenario ranges (about 670–1,260 TWh by 2030) rather than a single forecast.