Nobody Is Getting the Data-Center Water Question Right
4 hours ago
- #water usage
- #data centers
- #AI sustainability
- AI water use is debated: some view it as highly water-intensive, exacerbating droughts, while others dismiss concerns as exaggerated.
- Data center water consumption varies widely by location, design, and local water supply; national usage is low (<0.1% of U.S. farm water) but can strain local resources in drought-prone areas.
- Closed-loop cooling systems, used by many AI data centers, reuse water and avoid evaporation, reducing direct water demand but increasing electricity use (10–65% more), which indirectly consumes water via power generation.
- Indirect water use from electricity production for data centers can be much larger than direct use (e.g., Meta's indirect use was 23 times its direct consumption in 2024), though estimates are flawed due to grid averaging.
- Solutions exist to reduce both water and electricity use: leveraging cooler climates, evaporative cooling, higher chip temperatures, and renewable energy sources that require minimal water.
- Lack of transparency and accurate data on water use, combined with secretive negotiations by tech companies, fuels public confusion and concerns over environmental impacts and local control.