Key Questions on Energy and AI | Research in context

International Energy Agency · Published 16 April 2026 · Checked 5 October 2026

Data period: Periods vary by question. The data-centre demand cases show historical estimates from 2020 through 2025 and projections to 2035. Supporting market, project and supply-chain inputs include material available through early 2026.
Scope: Global and regional analysis of data-centre electricity demand, power supply, constraints, affordability and selected uses of AI in energy. It covers all data-centre workloads when modelling facility electricity, then separately analyses AI-focused infrastructure where the evidence permits.
Method: The electricity-demand model is a bottom-up equipment-stock model derived from annual shipments of servers, storage and network equipment, with equipment power and utilisation, facility type and power usage effectiveness assumptions. The broader report also uses project and satellite tracking, financial-market analysis, case studies and OECD economic modelling.
Assumptions: Four cases vary demand for digital and AI services, efficiency, financing and physical constraints. Regional allocation and electricity use depend on shipment forecasts, equipment lifetimes, idle and active power, utilisation, storage and network stocks, facility archetypes, power usage effectiveness and the timing of grid, chip and energy-equipment bottlenecks. Some technical inputs use United States estimates across regions.
Uncertainty: The report says uncertainty widens after 2030. Outcomes depend on AI adoption, model and hardware efficiency, market sentiment, speculative, early-stage, delayed, cancelled or otherwise unrealised project pipelines and the pace at which supply-chain and grid constraints ease. The underlying third-party datasets are not reproduced in enough detail in the public report for independent audit. Different sections have different evidence strength and cannot responsibly be combined into a single net impact.
Funding and interests: The work was designed and directed within the IEA. The acknowledgements list reviewers from governments, utilities, technology and energy companies, manufacturers, research organisations, civil society and consultancies. The report says its findings reflect the IEA Secretariat and not necessarily any funder, supporter or collaborator. No separate external funding amount or sponsor is identified in the acknowledgement pages reviewed. These disclosures do not establish whether the findings are correct.
Reuse: The report landing page labels the report CC BY 4.0. The PDF gives Annex A, including its data tables, a separate CC BY-NC-SA 4.0 notice. The standalone data product uses the IEA non-CC terms. The IEA excludes logos, cover images, photographs and third-party components from its CC grant. ComputeFinder links to the report and uses an original attributed summary only, without copying charts, tables, data, images or third-party material.
Attribution: IEA 2026; Key Questions on Energy and AI, https://www.iea.org/reports/key-questions-on-energy-and-ai, Licence: CC BY 4.0. Read the licence.
Derived-work notice: This is a work derived by ComputeFinder from IEA material and ComputeFinder is solely liable and responsible for this derived work. The derived work is not endorsed by the IEA or its Member countries in any manner.
What it cannot establish: It cannot establish the electricity or water demand, grid effect, planning merits, price effect, construction stage, community view or delivered benefit of a particular project. Its cases are not commitments, and global or regional results do not resolve a local decision.

Continue from research to a local decision

This report gives wider model context, not evidence about one proposal. Use the community questions guide to ask for local evidence, inspect the recorded projects, or read the publication method. Unknown local conditions stay unknown.

This record explains one source within its stated boundaries. It is not a verdict on a project, a forecast that a scenario will occur or evidence of community consent.