How AI Data Centers Cool Their Servers — and Why Water and Power Are at the Center of the Debate

AI data centers run thousands of GPUs at extreme temperatures, and keeping them cool requires large amounts of water and energy. This report visits operating facilities — including an Equinix site in San Jose — to examine how liquid cooling works, how much water different cooling methods actually consume, and why communities and utilities are pushing back on new projects.

■ Cooling Infrastructure and Liquid Cooling
- Inside a Grace Blackwell rack deployment and a closed-loop liquid cooling system
- How direct-to-chip cooling works, and why liquid is far more effective than air for GPUs
- Air cooling vs. evaporative cooling vs. hybrid approaches, and their water-energy tradeoffs

■ Water and Power Impact
- Reported water consumption figures from data center operators and local community concerns
- Why water-saving cooling methods increase electricity demand, and the role of on-site power generation

■ Industry and Chip-Level Solutions
- Equipment suppliers serving Nvidia and AMD, and corporate activity in the cooling sector
- Companies working on lower-heat AI chips and memory, and efforts to deliver coolant closer to the silicon

For anyone following AI infrastructure, data center policy, or energy and water resource issues, this video offers a clear picture of how cooling decisions shape the environmental footprint of AI — and what to watch as new facilities come online.