Generative AI is an emerging emission source. Training large models is very energy-intensive and day-to-day use is less so per query, but volume becomes significant when a company rolls out an assistant to 500 to 5,000 users. Emission factors are still being developed (by ADEME, Arcep and providers). The sound approach is to track usage volumes (monthly queries, models called) and apply the first estimates available, with a data quality flag to be reviewed every 6 months.
Read the source article: IT carbon footprint of a French company: where to act first