A report compared xAI's Colossus 2 water usage to 2.5 In-N-Out burger restaurants. The math is defensible for one specific datacenter, at one specific point in time. But datacenter water consumption changes dramatically based on how you cool the chips and how you generate the power.
This report walks through the thermodynamics, engineering tradeoffs, and real-world case studies behind that shift, from Brayton and Rankine cycles to what actually happened at xAI's Memphis facility to why nuclear energy brings water consumption right back into the conversation.
What you'll walk away with
A clear understanding of why two datacenters with identical compute capacity can have wildly different water footprints
The ability to evaluate carbon vs. water tradeoffs across different power generation technologies
Technical context on behind-the-meter power (gas turbines, fuel cells, solar) and why datacenters use them
A realistic picture of nuclear power as the long-term datacenter energy solution, including its water implications
A framework for questioning oversimplified infrastructure metrics, whether they come from analysts, executives, or policymakers