A Water Research paper (vol 299, 2026) argues AI’s environmental footprint is assessed through the wrong lens. Electricity and emissions get the attention — but water is the “critical, less visible dimension.”
AI infrastructure draws freshwater three ways:
- Evaporative cooling in high-density data centers
- Indirect water use in electricity generation
- Water-intensive semiconductor manufacturing
Projections put the global footprint at 4.2–6.6 billion cubic meters annually by 2027 — with many facilities located in water-stressed regions.
The mitigation toolbox exists — cold-climate siting, natural water body cooling, waterless designs, waste heat recovery — but deployment is thin.
The paper’s contribution is a governance framework it calls “digital water sobriety”:
- Evaluate which AI applications actually justify freshwater consumption
- Water-conscious siting for new infrastructure
- Mandatory facility-level water-use transparency
The conclusion is pointed: water-sustainable AI demands policy reform that integrates water constraints into computational infrastructure planning — not just better cooling tech.
A useful complement to the energy-focused AI-sustainability conversation — and a reminder that the “just run it locally” answer has a water bill too.