Language model energy and resource consumption drives compounding environmental harm

Executive Summary

Large language models impose material environmental costs across training, inference, water consumption, and hardware resource extraction, with secondary and behavioural emissions hardest to measure. Boards deploying AI at scale face unquantified carbon liability and growing regulatory exposure under sustainability reporting obligations.