Sustainable AI infrastructure: Forecast of global water footprint
EarthArXiv
· April 24, 2025
· ✓ verified
- A global assessment quantifies the water footprint of AI-driven data centers distinguishing between operational water, off-site electricity generation water, and embodied water in hardware manufacture.
- Scenario-based probabilistic forecasting until 2050 projects global water use could increase more than sevenfold without mitigation, with majority from cooling-related operational use.
- Mitigation includes improving cooling efficiency, adopting alternative technologies like liquid and immersion cooling, and regional infrastructure planning based on water availability.
- Focus on efficiency and infrastructure planning aligns AI expansion with long-term sustainability goals and addresses water stress in vulnerable regions.
- The rapid rise in AI computational demand follows an exponential trend increasing water consumption considerably, highlighting the urgency for sustainable digital infrastructure policies.