Sustainable AI Group Launches to Bridge Data Gap in AI Emissions
Amidst a US policy environment rolling back environmental protections, the new venture seeks to equip companies with the metrics needed to navigate growing regulatory and employee pressure for sustainable AI practices.
Sasha Luccioni, an artificial intelligence sustainability researcher formerly of Hugging Face, has co-founded the Sustainable AI Group alongside Boris Gamazaychikov, the former sustainability chief at Salesforce. The new venture aims to assist companies in reducing the environmental impact of artificial intelligence by improving transparency regarding energy consumption and emissions data. Luccioni argues that while the rush to build AI infrastructure continues, driven partly by US policy shifts, there is growing pressure from employees and international regulations, such as the EU AI Act, for better data on the carbon footprint of AI models.
Luccioni spent four years at Hugging Face, where she pioneered a leaderboard documenting the energy efficiency of open-source AI models. She has been a vocal critic of major AI companies for allegedly withholding energy and sustainability information from the public. The new venture will focus on identifying "levers" to make AI agents less environmentally damaging and will investigate the energy needs of specific AI tools, such as speech-to-text translation and photo-to-video generation, areas she notes have been understudied.
She advocates for detailed emissions reporting per query, suggesting that user interfaces for models like ChatGPT or Claude should display energy usage and greenhouse gas emissions at the end of each conversation. Luccioni believes this transparency could become a market competitive advantage, allowing providers to differentiate themselves in a crowded field. She argues that companies can reduce their carbon footprint by matching tasks to appropriate model sizes, such as using simpler classifiers for routine tasks rather than large language models, and by choosing data centres powered by renewable energy.
The push for accountability is being driven by internal corporate pressure as well as international regulation. Luccioni notes that employees and boards are demanding quantification of AI’s impact on ESG goals, particularly as AI becomes a core part of business offerings. While the US administration is reportedly rolling back environmental protections, other regions are moving forward; the EU AI Act includes sustainability clauses, and the International Energy Agency has highlighted a significant data gap regarding specific data centre energy use that hinders future capacity planning.
Luccioni distinguishes between large language models and other AI systems, noting that classifiers have historically driven significant productivity gains with lower resource demands. She suggests that greater transparency in token usage and query types would allow companies to optimise their model selection, reducing reliance on expensive, high-energy general-purpose models for simple tasks. The Sustainable AI Group aims to provide the actionable strategies and data clarity needed for organisations to navigate these complex environmental and operational challenges.
