As artificial intelligence becomes embedded in everyday life, its impact on water resources remains largely invisible. Compared with energy and carbon emissions, the freshwater demand of data‑center cooling and electricity generation is more concealed. To address this gap, we propose SustainAI, a water‑aware, closed‑loop framework for environmental accountability. SustainAI integrates real‑time water metering, a hallucination‑aware penalty model, and a water‑aware routing algorithm that accounts for regional water stress. Experiments using small language models (SLMs) to extract health misinformation across geographically distributed data centers show that the water footprint per inference varies from 0.0477 mL to 0.5360 mL, an eleven‑fold difference. Across 1,335 inference runs the system consumed about 399 mL of water but produced only 240 correct outputs, indicating substantial resources spent on inaccurate responses. The framework further adopts a Care by Design perspective, framing AI sustainability around relational ethics, regional equity, and ecological stewardship. By combining water monitoring, adaptive accountability, and care‑by‑design principles, SustainAI offers a practical foundation for embedding ethical care and environmental responsibility into AI infrastructure design and lifecycle management.
Review