Hugging Face AI climate researcher
Sasha Luccioni
Profile
Sasha Luccioni is the AI & Climate Lead at Hugging Face, and probably the single most-cited voice on the energy cost of running machine learning models. She has a PhD in AI, spent a few years working alongside Yoshua Bengio at Mila on AI-for-good projects, and is a founding member of Climate Change AI. If you’ve ever wondered how much carbon your fine-tuning run actually emits — you’re using numbers she helped put on the map.
Her most cited work is the BLOOM carbon footprint paper, the first serious accounting of emissions from training a 176B-parameter model end-to-end. She followed it with Power Hungry Processing, which measured inference costs and showed that general-purpose generative models burn orders of magnitude more energy than task-specific ones for the same job — a finding every engineer picking between a small fine-tuned model and a big API call should internalize. She also co-created CodeCarbon, the Python library that’s been installed over a million times to log emissions from training runs.
In 2025 she launched the AI Energy Score at the Paris AI Action Summit — an energy-star-style rating for models across ten common tasks. The December 2025 refresh added reasoning as a benchmarked task and found reasoning models use roughly 30x more energy than non-reasoning ones. That’s the kind of measurement that matters if you’re shipping products: “chain-of-thought everything” has a real cost attached.
Why she matters to developers: she’s neither a doomer nor a cheerleader. She doesn’t argue against building AI — she argues for knowing what it costs. That work earned her spots on the TIME 100 AI 2024/2025 and BBC 100 Women 2024 lists. She does the actual measurement work that lets the rest of the field make honest tradeoffs.
Key Articles & Papers
Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model Power Hungry Processing: Watts Driving the Cost of AI Deployment? Counting Carbon: A Survey of Factors Influencing the Emissions of Machine Learning Announcing AI Energy Score Ratings AI Energy Score v2: Refreshed Leaderboard, now with Reasoning Making an image with generative AI uses as much energy as charging your phone AI Models Hiding Their Energy Footprint? Here's What You Can Do The Environmental Impacts of AI — Policy PrimerVideos
Spotify Podcasts
