It takes energy to generate the LLM response. If you read the already generated response instead of generating your own, we burn less coal.