The environmental cost of AI inference comes from the electricity, cooling water, hardware, and regional infrastructure used each time deployed models serve workloads at scale. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.