As companies quickly integrate Generative AI (GenAI) into their core operations, there is a fundamental contradiction between technical advancement and Environmental, Social, and Governance (ESG) duties.With an emphasis on training versus inference energy consumption, this article examines the extent of AI's carbon footprint and assesses the adequacy of current carbon-neutrality measures. We anticipate that daily AI inference currently accounts for 88% of total AI-related energy use, requiring a change from training-focused mitigation to real-time operational optimization, based on an analysis of simulated energy consumption data from 50 top tech and finance organizations (2023-2026). According to our research, businesses that used "Carbon-Aware Scheduling" decreased their Scope 2 emissions linked to AI decreased by an average of 22% during the same time period, but those who merely used carbon offsets experienced an effective reduction of 5%. Our "Green AI" architecture emphasizes algorithmic and infrastructure efficiency for sustainable implementation.