Or, for that matter, any incremental call Microsoft is making to any GPT-based model. The power consumption from these inferences are immense, and appearing everywhere in their interfaces.