Home › Directories › Green AI research
Directory
Green AI research
Research projects, labs, networks and open tools working on green and efficient AI.
About this directory
25 research projects, labs, networks and open tools working on green and sustainable AI, from measurement tools to efficient algorithms and hardware.
GreenCode (ITEA 4 project 23016)
Digital Tactics Ltd (project leader), with partners in eight countries · international
Eureka ITEA 4 project (October 2024 to October 2027) using AI and machine learning to optimise and certify software for energy efficiency. Partners span Austria, Denmark, Germany, Portugal, Romania, Spain, Türkiye and the UK; project site at greencode.ai.
AI-driven software decarbonisation
Visit →Green Software Foundation
Green Software Foundation (Linux Foundation) · international
Non-profit building an ecosystem of people, standards, tooling and best practice for green software. Home of the Software Carbon Intensity (SCI) specification and related guidance.
Standards and practice for green software
Visit →Climate Change AI
Climate Change AI (volunteer-run non-profit) · international
Volunteer community from academia and industry promoting machine learning where it genuinely helps tackle the climate crisis. Runs workshops, summer schools, grants and an interactive research library.
Machine learning for climate action
Visit →Green RSE Special Interest Group
Society of Research Software Engineering · United Kingdom
Community of research software engineers working to cut the environmental impact of research software through greener coding, efficiency and policy advocacy. Promotes the GREENER principles for sustainable computing.
Lower-impact research software
Visit →ML.ENERGY Initiative
University of Michigan (SymbioticLab), with collaborators at MIT, Columbia, CMU and University of Washington · United States
Academic effort to make modern machine learning energy-efficient. Publishes the ML.ENERGY Leaderboard and Benchmark on the time and energy cost of generative AI inference.
Measuring and reducing ML energy
Visit →Zeus
ML.ENERGY Initiative (University of Michigan) · United States
Open-source framework for measuring GPU energy use in deep learning and tuning workloads to use less of it. Part of the ML.ENERGY family of tools.
Deep learning energy measurement and optimisation
Visit →CodeCarbon
CodeCarbon open-source community · international
Open-source Python tooling that estimates the energy use and emissions of code running on local or cloud hardware. Now also hosts EcoLogits for generative-AI API calls.
Carbon tracking for compute and AI
Visit →EcoLogits
GenAI Impact (now part of CodeCarbon) · France
Library estimating the energy and environmental impacts of calls to generative AI model APIs. GenAI Impact, which created it, has joined CodeCarbon and continues the work there.
Footprint of generative AI APIs
Visit →Green Algorithms Initiative
Cambridge Sustainable Computing Lab, University of Cambridge · United Kingdom
Builds tools to estimate and monitor the carbon footprint of computation, including an online calculator and HPC dashboards, and studies whether they change behaviour. Led by Dr Loïc Lannelongue.
Carbon footprint of computing
Visit →Green DiSC (Digital Sustainability Certification)
Software Sustainability Institute, managed from the University of Cambridge · United Kingdom
Free, open certification scheme helping research groups and support teams reduce the environmental impact of digital and computational research. Tiered criteria start at Bronze.
Sustainability certification for computational research
Visit →AI Energy Score
Hugging Face and Salesforce, with Cohere, Carnegie Mellon University and others · international
Benchmarks AI models on a fixed GPU setup across ten common tasks and gives relative star ratings on a public leaderboard, in the spirit of appliance energy labels. Launched at the Paris AI Action Summit in February 2025.
Energy ratings for AI models
Visit →MLCommons Power Working Group
MLCommons · international
Develops methods for measuring the energy use of machine-learning systems so that power can be reported and compared alongside MLPerf performance results.
Power measurement for ML benchmarks
Visit →ML CO2 Impact Calculator
Schmidt, Luccioni, Lacoste and Dandres (Mila-affiliated researchers) · international
Web calculator that estimates the carbon emissions of machine-learning work from hardware, runtime, cloud provider and region. An early and widely cited tool for reporting ML emissions.
Estimating ML training emissions
Visit →Institute of Computing for Climate Science (ICCS)
University of Cambridge (Cambridge Zero, DAMTP, Computer Science and Technology, UIS) · United Kingdom
Develops computational science, data science and machine-learning tools to speed up climate research, with code clinics, training and internships. Part of the Schmidt Sciences Virtual Institute for Scientific Software and engaged with Green DiSC on sustainable computing.
Computational and ML tools for climate science
Visit →ADViCE: AI for Decarbonisation Virtual Centre of Excellence
The Alan Turing Institute and partners · United Kingdom
Virtual centre applying AI to help high-emitting industries move towards net zero. Also publishes material on reducing the electricity use and carbon footprint of AI itself.
AI for industrial decarbonisation
Visit →AEGIS: Adaptive Grid-Interactive Edge Datacenter Fleets
Technical University of Denmark (coordinator Prof. Paul Pop) · international
Marie Skłodowska-Curie Doctoral Network (2026-2030) funding 15 PhDs to coordinate edge datacentres as a virtual power plant using batteries, cooling and adaptable computing. Covers safe control, carbon-aware intelligence and fleet orchestration.
Carbon-aware, grid-flexible edge datacentres
Visit →MANOLO: Trustworthy Efficient AI for Cloud-Edge Computing
18-partner Horizon Europe consortium (CeADAR, Ireland, named as coordinator by partners) · international
Horizon Europe project (2024-2026) building algorithms and tools for compressed, frugal and energy-aware AI and its allocation across cloud and edge. Validated in health, manufacturing and telecoms.
Energy-efficient, trustworthy AI on the cloud-edge continuum
Visit →Boavizta
Boavizta (association) · France
Association producing open-source tools, methodologies and data for assessing and reducing the environmental impact of digital technology, including hardware embodied impacts.
Open data and methods for digital footprints
Visit →Coalition for Sustainable AI
Initiated by France with UNEP and ITU · international
Launched at the Paris AI Action Summit in February 2025 to stop AI worsening environmental harm and to use it to accelerate decarbonisation and protect ecosystems.
Multi-stakeholder action on AI and environment
Visit →Bonn Sustainable AI Lab
University of Bonn, Institute for Science and Ethics (Prof. Aimee van Wynsberghe) · Germany
Interdisciplinary lab, funded by the Alexander von Humboldt Foundation, studying the environmental, social and economic costs of designing, developing and using AI. Runs a recurring Sustainable AI conference.
Environmental and social costs of AI
Visit →Green Metrics Tool (Green Coding Solutions)
Green Coding Solutions (green-coding.io) · Germany
Open framework for measuring the energy use and carbon emissions of software applications. Developed by the green-coding.io team.
Measuring software energy and CO2
Visit →ARIA Scaling Compute
Advanced Research and Invention Agency (ARIA) · United Kingdom
Programme of nearly £100m in ARIA's Nature-Inspired Computation space, aiming to cut the cost of AI hardware by more than 1000x by drawing on natural information processing.
Radically cheaper, more efficient AI hardware
Visit →MIT HAN Lab
Massachusetts Institute of Technology (Prof. Song Han) · United States
Co-designs algorithms and systems (e.g. compression, quantisation, sparsity) to reduce the training and inference cost of large models, from reasoning and multimodal models to embodied AI.
Efficient ML through algorithm-system co-design
Visit →PULP Platform
ETH Zürich and University of Bologna · international
Open-source, energy-efficient RISC-V computing platform used widely for ultra-low-power and edge AI hardware research.
Open-source energy-efficient RISC-V hardware
Visit →Phase-Level Examination of the Energy Efficiency of LLM Inference
University of Twente, Formal Methods and Tools group · Netherlands
PhD project (2024-2028) measuring how prompt processing and token generation each drive LLM inference energy, and proposing software-level optimisations that balance the two.
Energy of LLM prefill versus decoding
Visit →Frequently asked questions
Can I list my research group or project?
Yes, for free. Submit your research with a link to its public page and a short description.