← Home
Field guide

AI and the environment

Training runs need electrons; the footprint is grids, hardware, water, and minerals—not just a cloud icon.

Overview

Artificial intelligence depends on electricity for data centres and networks, and on hardware supply chains that use energy and materials. The International Energy Agency (2025) synthesises global and regional modelling on electricity demand linked to data centres and AI, including which generation sources meet incremental demand in its scenarios. IPCC AR6 WGIII frames energy systems as portfolios where infrastructure, flexibility, and emissions interact; critical minerals for electronics and storage connect AI hardware to mining and recycling debates.

By the numbers

Evidence highlights

3 beats
  1. 01

    International Energy Agency (2025) reports analyse how electricity demand from data centres and AI-related workloads could evolve over the next decade, including regional differences and the generation mix that meets incremental demand in modelled scenarios.

  2. 02

    IPCC AR6 WGIII Chapter 6 stresses that integrating low-carbon electricity requires networks, storage, flexible demand, and institutions—not generation alone.

  3. 03

    International Energy Agency (2021) work on critical minerals highlights supply concentration and environmental pressures along chains that also underpin chips, batteries, and digital hardware.

Chart

Explore the data

Hardware purchase and electricity costs estimated to train notable AI systems—compare landmark models as training scale grew. Definitions (what counts as “training”), inflation treatment, and which systems are included are on the grapher page; this chart is not a full datacentre or lifecycle footprint.

Chart: Our World in Data (CC BY). Each grapher page lists the underlying datasets, units, and processing notes—use it when citing numbers.

Open on Our World in Data

Trade-offs

Where it helps—and where it hurts

Strengths

  • AI can support forecasting, grid operation, and materials discovery when deployed with transparency and sound data governance.
  • Policy mixes—efficiency standards, clean procurement, and grid planning—can steer new compute demand toward lower-emissions electricity where institutions are credible.
  • Improved metering and reporting of facility-level energy use can reduce information gaps compared with opaque growth.

Limits & trade-offs

  • Rapid demand growth can stress grids and emissions if incremental electricity is met disproportionately with unabated fossil generation; outcomes are scenario- and region-dependent.
  • Hardware lifecycles (manufacturing, water use for cooling, e-waste) carry environmental burdens beyond operational electricity.
  • Projections of AI-specific electricity use remain uncertain; media narratives can outrun verified public data and methods.

References

Sources for this page

These entries are starting points for verification. Prefer the original report or dataset when checking numbers and figures.

  1. IEA Energy and AIInternational Energy Agency. (2025). Energy and AI. IEA. https://www.iea.org/reports/energy-and-ai
  2. IPCC AR6 WGIII Ch. 6Clarke, L., Wei, Y.-M., De La Vega Navarro, A., Garg, A., Hahmann, A. N., Khennas, S., Azevedo, I. M. L., Loschel, A., Singh, A. K., Steg, L., Strbac, G., & Wada, K. (2022). Energy systems. In P. R. Shukla et al. (Eds.), Climate Change 2022: Mitigation of Climate Change (IPCC AR6 WGIII, Chapter 6). Cambridge University Press. https://doi.org/10.1017/9781009157926.008
  3. IEA critical mineralsInternational Energy Agency. (2021). The role of critical minerals in clean energy transitions (World Energy Outlook Special Report; revised March 2022). IEA. https://www.iea.org/reports/the-role-of-critical-minerals-in-clean-energy-transitions
  4. Our World in DataRitchie, H., & Rosado, P. (2020). Electricity mix. Our World in Data. https://ourworldindata.org/electricity-mix (underlying grapher datasets include Ember and Energy Institute series, cited per chart metadata).