Setting a New Bar for AI Infrastructure

Google’s Water Pledge

A typical mid‑sized data center uses something like 300,000 gallons of water a day. That’s roughly equivalent to the amount used by 1,000 U.S. households. And large facilities—the kind we are told will be built in significant numbers—can use millions of gallons per day.

AI is already the most rapidly adopted general-purpose technology in history. The data centers are rapidly following suit. Since realistically, usage will not abate anytime any time soon, we can expect the amount of water usage to soar and quickly.

That’s, of course, cause for concern.

For serious people in the tech and NGO community who prioritize making meaningful progress above ideological purity, the important question now isn’t whether to build new data centers. It’s how we should be building them. In some places with extreme water stress or very weak governance or critically important ecosystems, the right answer may be not to build at all. But for most projects that cross the desks of corporate deal teams and NGOs, the more productive question is: “Under what terms do we say yes?”