Mistral AI has unveiled an environmental impact assessment report for its flagship model, Mistral Large 2, offering unprecedented insight into the carbon emissions and water consumption associated with both the training and operational phases of the model. According to the report, during the training phase alone, the model emitted 204,000 tons of carbon dioxide and utilized 281,000 tons of water. Notably, while a single text generation task may seem negligible with just 1.14 grams of carbon dioxide and 45 milliliters of water, repetitive use can lead to substantial cumulative environmental effects. Furthermore, the study highlights that larger models exert a greater environmental toll, potentially steering the AI industry towards miniaturization and niche specialization. As environmental regulations tighten, transparency in environmental data is poised to become a new norm for the AI sector.