AI Energy Consumption Index: Q2 2025 Edition
The hidden electricity footprint of leading AI services, benchmarked for UK decision-makers. An independent, evidence-based quarterly index covering ChatGPT, Claude, Gemini, Copilot and more — with UK demand, grid impact and forecasts to 2035.
What this report covers
Global data centre electricity reached an estimated 485 TWh in 2025, with AI workloads driving 20–25% of that total and growing 50% year-on-year. This edition of the AI Energy Consumption Index sets out transparent, evidence-based estimates for the leading AI models and services, benchmarks UK demand against international peers, and projects the electricity trajectory to 2035 under three scenarios.
Headline conclusions
- Global data centre electricity reached ~485 TWh in 2025 (1.6% of global demand); AI is now 20–25% of that total.
- Under the IEA Base Case, global data centre demand could reach 945–950 TWh by 2030 — roughly Japan's annual consumption.
- Inference has overtaken training as the dominant AI electricity driver, led by ~2.5 billion daily ChatGPT queries.
- UK data centres used ~5–6 TWh in 2023–2024 (~2% of demand); Oxford Economics projects 26.2 TWh by 2030 (8.8%).
- UK grid-connection offers for demand-side projects jumped from 41 GW to 125 GW between late 2024 and mid-2025 — network capacity, not capital, is now the binding constraint.
Intended audience
- Policymakers, regulators and DESNZ / Ofgem analysts
- Data centre operators, hyperscalers and grid planners
- AI companies, cloud providers and semiconductor investors
- Sustainability, ESG and net-zero leads at UK enterprises
- Journalists and researchers covering AI's energy footprint
Inside the 44-page report
- 011. Executive summary and key findings — Q2 2025
- 022. Methodology — how AI electricity estimates are calculated
- 033. Executive dashboard — regional demand and service benchmarks
- 044. State of AI — adoption timeline and market structure
- 055. Electricity profiles — ChatGPT, Claude, Gemini, Copilot, Meta AI, Perplexity, Grok, DeepSeek
- 066. Image generation — Midjourney and Stable Diffusion
- 077. Model comparison — efficiency by class, open-source vs proprietary
- 088. Data centres — AI server deployments, cooling and grid impacts
- 099. Electricity trends 2020–2025 and quarterly peak demand
- 1010. Water consumption — cooling systems and regional impacts
- 1111. Carbon footprint — Scope 2, renewables and net-zero commitments
- 1212. International comparison — UK, US, EU, China, Japan, Singapore, Middle East
- 1313. Electricity demand forecast to 2035 — three scenarios
- 1414. Industry analysis — semiconductors, nuclear, renewables and grid flexibility
- 1515. Key risks and opportunities
- 1616. Recommendations — government, energy companies, AI developers, cloud providers and investors
- 1717. Glossary, references and detailed methodology appendices
Sample pages
Frequently asked questions
Is the index free to download?+
Yes — the full 44-page Q2 2025 edition is free. We ask for an email so we can send you future quarterly updates.
How are the electricity estimates produced?+
From disclosed hardware configurations, published user and query metrics, academic modelling and infrastructure filings. Methodology and assumptions are documented in full in the appendices.
How often is the index updated?+
Quarterly. Estimates are revised as new operator disclosures, academic research and infrastructure data become available.
Download the full report
Enter your email to get the full PDF plus the weekly UK AI energy briefing.