Data Center water scare
As you know, Data Center water use has blown up as an explosive NIMBY political issue. When I first heard about it, I thought it odd as we’d solved most of these problems years ago. 200mm fabs were the first to solve this in the 90s, back when 80s-era 6-inch fabs were projected to create a water shortage in Arizona. It’s no secret that we’re still building fabs in Arizona.
The Chip Insider®
Never miss an edition
Strategic and timely insights tailored for senior semiconductor executives.
October 21, 2026
From Dr. Jekyll to Mr. Hyde
Today’s 300mm fabs sip water compared to the fabs of old on a per-wafer basis. Chip fabs still use plenty of water. For example, TSMC’s Fab 21, Phase 1, in Phoenix is spec’d at 4.75M gallons per day… When TSMC gets to 90% reclamation levels it will be equal to half a local golf course.
Data Centers have a Jekyll and Hyde character problem in fact, and with the public. In the public’s mind, they loved the Dr. Jekyll data center that came to town for all the jobs and opportunity it promised. When Mr. Hyde showed to steal their water, there was a fast transition from happy to angry. You know all this. But in fact, the Jekyll and Hyde data centers are two separate classes, because the first Jekyll was kind of an unknowing con artist. The real Dr. Jekyll came with AI.
Two Data Center Classes
The non-fiction Mr. Hyde one is really the Cloud Computing Data Center. The Dr. Jekyll one is the newer AI Data Centers. The difference is due to differing compute densities. The real surprise is once AI-class data center is up and running, it uses zero water for cooling.
Cloud Data Centers
In a cloud computing data center the water is taken from the community water shed and evaporated up into the clouds, to be returned in rain or snow somewhere else. The political Mr. Hyde comes from the companies and community authorities bringing in cloud data centers. They either didn’t know, didn’t plan for, or didn’t care what the impact on the community’s water supply would be. Now there is a huge populist backlash.
AI Data Centers/AI
After 2024, you see a shift to direct-to-chip radiator cooling, similar to how a car’s engine is cooled with water. Once the water is put in, you don’t fill it again unless it needs to be flushed… Water Use Intensity (WUI)2 for Microsoft’s Fairwater AI Data Center is under 20 gallons/1K-square-feet/day. That’s about the same as the EPA’s average for a church and half that of an office building or a K-12 school. Nvidia’s Omniverse AI Data Center design blueprint is zero-water for cooling and it’s been made publicly available. Examples of zero water for cooling data centers include: OpenAI/Oracle Stargate in Texas, and xAI’s Colossus in Memphis.
So, I expect the political issues to eventually die down as the failure stories die down and are replaced with success stories. Yes…
About The Chip Insider ®
Gain a competitive edge with The Chip Insider®, the trusted source for strategic and tactical insights tailored for senior semiconductor executives. Stay ahead with edgy, blunt industry analysis, exclusive email reports, and expert perspectives on market trends, strategies, and key challenges.
Dan is Vice Chair of TechInsights Inc. He is a recognized authority on the semiconductor industry, winning SEMI’s Sales and Marketing Excellence Award in 2012 for “empowering executives with tremendous strategic and tactical marketing value" through his e-letter, The Chip Insider®; his book Maxims of Hi-Tech, and his many interviews of executives.
Strategic and tactical insights tailored for senior semiconductor executives.
Sign in or create a free TechInsights Platform account.
