Nature Energy intl_tech D1

发布：2026-05-28 · 事件：2026-05-28

Subjects Energy infrastructure Energy modelling Access through your institution Buy or subscribe Understanding the energy used each time artificial intelligence (AI) systems generate a response is inc...

Subjects Energy infrastructure Energy modelling Access through your institution Buy or subscribe Understanding the energy used each time artificial intelligence (AI) systems generate a response is increasingly important for energy policy and grid planning. However, many estimates are based on non-production or small-scale settings that do not reflect how AI systems operate in real data centres, leading to systematic overestimation of real-world energy use. Now, Felipe Oviedo and colleagues at Microsoft report a bottom-up method that estimates energy use by adding up the main parts of the system, such as computing speed, hardware power use, and data centre overheads under realistic serving conditions. The researchers estimate an energy use of 0.31 Wh per query by modelling the performance of three open-source AI models that were on a similar scale to widely-used commercial chatbots. The approach also separates simple queries from longer queries, such as those used for programming and agentic models, showing that these longer queries require nearly 13 times more energy per query. Oviedo and colleagues also examined efficiency improvements reported in earlier research, including better hardware and more efficient inference methods and showed potential energy usage reductions of up to a factor of 20. However, they note that such reductions may be partially offset if improvements lead to greater usage or more compute-intensive applications. This is a preview of subscription content, access via your institution Access options Access through your institution Access Nature and 54 other Nature Portfolio journals Get Nature+, our best-value online-access subscription 27,99 € / 30 days cancel any time Learn more Subscribe to this journal Receive 12 digital issues and online access to articles 111,21 € per year only 9,27 € per issue Learn more Buy this article Purchase on SpringerLink Instant access to the full article PDF. 39,95 € Prices may be subject to local taxes which are calculated during checkout Author information Authors and Affiliations Nature Reviews Clean Technology https://www.nature.com/nrct/ Matthew Allinson Authors Matthew Allinson View author publications Search author on: PubMed Google Scholar Corresponding author Correspondence to Matthew Allinson . Rights and permissions Reprints and permissions About this article Cite this article Allinson, M. Energy use in AI responses. Nat Energy 11 , 650 (2026). https://doi.org/10.1038/s41560-026-02084-9 Download citation Published : 27 May 2026 Version of record : 27 May 2026 Issue date : May 2026 DOI : https://doi.org/10.1038/s41560-026-02084-9 Share this article Anyone you share the following link with will be able to read this content: Get shareable link Sorry, a shareable link is not currently available for this article. Copy shareable link to clipboard Provided by the Springer Nature SharedIt content-sharing initiative