Sustainability for generative AI is an overlooked yet important topic. Estimates from 2022 from an unpublished research paper from Hugging Face are here at this MIT news article. On the positive side, as of 2025 we have extremly efficient chips and more powerfull processors that can help mitigate the problems--see news from Forbes.
Here below there is a collection of resources that can help framing the problem.
The environmental Impact of AI: Measuring the impact
- Strubell et al (2019) Energy and Policy Considerations for Deep Learning in NLP (https://arxiv.org/pdf/1906.02243)
- Liu and Yin (2024)Green AI: Exploring Carbon Footprints, Mitigation Strategies, and Trade Offs in Large Language Model Training (https://doi.org/10.48550/arXiv.2404.01157)
- Luccioli et al. (2024) The Environmental Impacts of AI --Primer (https://huggingface.co/blog/sasha/ai-environment-primer)
- Massive list of papers from arxiv-sanity-lite: (https://www.arxiv-sanity-lite.com/?rank=pid&pid=2309.14393)
Research on how to build a framework to measure the environmental impact of genAI
- Lin, Yuan & Ghose, Debasish & Coates, David. (2024). Towards Sustainability of Large Language Models for Recommender Systems. (https://www.researchgate.net/publication/384104029_Towards_Sustainability_of_Large_Language_Models_for_Recommender_Systems)
Research on how to use LLMs to improve sustainability reports
- Zhou et al. (2024) Accessing the Capabilities of KGs and LLMs in Mapping Indicators within Sustainability Reporting Standards (https://mediatum.ub.tum.de/doc/1755295/1755295.pdf)
- Jonveaux J. (2024) Using Large Language Models for a standard assessment mapping for sustainable communities (https://arxiv.org/html/2411.00208v2)
Check the Hugging Face Spaces Carbon Compare and AI carbon.