Maximizing Efficiency for Generative AI