Running AI models at scale uses significant electricity, drawing attention to efficiency. Engineers cut the energy per request through smaller and smarter models, specialised chips, and techniques that avoid unnecessary computation, while data centres improve cooling and tap cleaner power.
Lower energy per answer reduces both cost and environmental impact, and it widens access by letting more organisations run capable models affordably. Overall demand keeps rising as AI is used more, so efficiency gains are partly offset by growth, which is why reducing the energy of each individual request has become such a clear and practical priority.