A recent forecast predicts AI will use over half of data center electricity by 2028. Research shows that over 80% of AI compute is now used for inference, with AI energy consumption dominated by scale, deployment patterns, and system inefficiencies rather than individual model runs. MIT’s research indicates that per-task energy consumption varies across text, image, and video tasks. MIT Technology Review shows that task-specific models are less carbon- and energy-intensive than general-purpose generative models used for narrow tasks. Infrastructure studies show that servers account for about 60% of data center electricity consumption. Recommendations include prioritizing inference efficiency, measuring per-task energy metrics, using task-specialized models, reducing system-level waste through batching, caching, and smarter scheduling, and treating hardware efficiency and Power Usage Effectiveness as software concerns.