Vanderbauwhede study warns AI data centre expansion will spike CO2 emissions
New arXiv paper quantifies both operational and embodied emissions from AI servers...
Wim Vanderbauwhede's latest arXiv paper, "Estimating the growth in emissions from AI data centres" (2608.14421), tackles the climate cost of the AI boom. Using state-of-the-art life cycle assessment models, it calculates both operational emissions (electricity used by servers) and embodied emissions (carbon produced during manufacturing and disposal) from AI data centre expansion. The analysis runs projections based on growth scenarios from the International Energy Agency (IEA) and McKinsey, offering a rigorous, quantitative framework for understanding the environmental footprint of the industry's rapid build-out.
The findings are stark: the AI industry's own growth scenarios would lead to a dramatic rise in overall CO2 emissions, and embodied carbon is far from negligible. This complicates the narrative that AI will drive efficiency gains that offset its carbon footprint. The paper, accepted at the 2nd International Workshop on Low Carbon Computing (LOCO 2026) in Lancaster, UK, adds a timely data point for policymakers and data centre operators weighing sustainability against AI's resource demands.
- Uses life cycle assessment to model both operational and embodied emissions from AI servers
- Projects emissions under IEA and McKinsey scenarios for AI data centre expansion
- Finds industry-promoted growth would cause a dramatic rise in emissions, with embodied carbon a significant factor
Why It Matters
AI's sustainability promises could be undermined as data centre construction and embodied carbon become major emission drivers.