thousands of tons per model. Operating data center cooling systems also consumes large quantities of freshwater, while the production and disposal of hardware such as GPUs involve the extraction of rare earth minerals and contribute to the growing problem of electronic waste. Rising demand on power grids to support data center operations has also strengthened reliance on fossil fuels and nuclear power. At the same time, some Big Tech companies promoting their AI-related performance have begun to retreat from environmental, social, and governance (ESG) commitments, including targets for reducing carbon emissions. Some argue that the environmental costs of generative AI are overstated or that they will be mitigated through technological advances. Even if this proves true, it is difficult to treat the issue lightly given the rapid expansion in the scale of use—where generative AI is being integrated into an increasing number of services and, in some cases, operates continuously at the operating-system level, such as with Microsoft Copilot. While technical solutions such as transitioning data centers to renewable energy or developing more efficient algorithms (including model compression and lightweight architectures) are being explored, these approaches remain limited as they rely primarily on voluntary corporate efforts. There are also claims that investing even more resources into advancing AI technology could help solve major challenges such as climate change and ultimately offset current environmental costs. However, such arguments seem closer to romantic optimism than to scientifically grounded projections. Generative AI Guide for Civil Society