The Incompatibility of AI and Decarbonization (www.lemonde.fr)

A new report from the Shift Project (chaired by engineer Jean‑Marc Jancovici) warns that AI’s growth is on a collision course with decarbonization. The think tank estimates electricity use by data centers will triple by 2030, and that AI workloads could make up 33–50% of that demand—up from about 15% today. The increase is driven by the massive, always‑on server farms that support model training and inference at scale, and it risks boosting greenhouse‑gas emissions just as other sectors try to cut them to meet Paris targets. The findings matter because much grid electricity—especially in the U.S.—still comes from fossil fuels, and even regions with low‑carbon power (e.g., France) could face allocation conflicts as data centers absorb more clean capacity, pushing prices up and slowing electrification elsewhere. The report dismisses “techno‑solutionism”: efficiency gains are unlikely to outrun exponential demand. It calls for measures like caps on electricity use and prioritizing essential applications, but those recommendations clash with fierce global competition for AI investment and strategic national agendas, making regulatory restraint politically fraught. In short, the Shift Project reframes AI as not just a computational and policy challenge, but a core energy and climate dilemma requiring urgent cross‑sector planning.