AI Data Centers and Water: The Local Problem Everyone Is Getting Wrong

No, Data Centers Aren't Draining the Rivers. But Your Town Might Be Next.

TLDR: Nationally, AI data centers use well under 1% of US public water supply, versus agriculture's 42%. Locally, a single facility can hit 40% of a small city's water. The problem is siting and cooling, not a global water crisis.

The framing here cuts through a lot of noise in the AI environmental debate. The viral narrative about AI "draining rivers" and causing water crises at a civilizational scale is not supported by peer-reviewed research or USGS data. But the local impact story is real, documented, and worth taking seriously.

The Dalles, Oregon, where Google's data centers now consume around 40% of the city's municipal water, is the case study that keeps appearing for good reason. A city of about sixteen thousand people, one company, a significant share of the local water supply. The city spent two years in court on Google's side trying to keep the number secret as a trade secret. They lost; the figure became public record. That kind of data opacity around local environmental impact is its own problem separate from the water consumption itself.

Newton County, Georgia, where Meta's facility uses about 10% of county supply, is another documented case. These aren't small numbers for small communities. When a single building represents 10-40% of a municipality's peak water capacity, local infrastructure planning, drought resilience, and long-term water security all become legitimate concerns.

The pollution angle gets messier. Many of the viral images circulating about AI data centers causing water contamination turn out to be AI-generated fakes or misattributed to construction activity rather than operating facilities. No operating data center has been shown to be the primary cause of water-source contamination in a documented peer-reviewed case.

The honest conclusion here is that the right frame is siting and cooling technology. Where facilities are built (near water-stressed communities vs. not) and how they are cooled (water evaporation vs. air cooling vs. newer alternatives) determines the local impact. The national aggregate story is fine. The local story depends entirely on choices made by facility developers and permitted by local governments.

Key takeaways:

• AI data centers account for well under 1% of US public water supply nationally
• Individual facilities can reach 40% of a small city's water usage, as in The Dalles, Oregon
• Most viral water contamination stories about data centers are false or misattributed
• The problem is a siting and cooling technology question, not evidence of a global water crisis

Why do I care: For developers working on AI infrastructure or evaluating cloud providers, understanding the actual environmental impact story matters. The nuanced local-vs-global framing here is more useful than either the "AI is fine" or "AI is destroying the planet" narratives. For anyone involved in data center siting decisions or sustainability planning, this is foundational reading to distinguish real impact from manufactured outrage.

No, Data Centers Aren't Draining the Rivers. But Your Town Might Be Next.