Artificial Intelligence and Water Use: A Systems-Level Analysis of Assumptions, Accounting, and Impact
Major
Physics
Second Major
Art, BA
Submission Type
Poster
Area of Study or Work
Physics
Faculty Advisor
Narendra Jaggi
Location
CNS Atrium
Start Date
4-11-2026 11:15 AM
End Date
4-11-2026 12:30 PM
Abstract
Public discussion about artificial intelligence and water use often centers on striking claims about water consumed per prompt or per training run. This research project begins a critical examination of those claims by asking what is actually being counted as “water use.” It distinguishes among on-site cooling water, electricity-related water, and water embedded in hardware manufacturing, and examines how system boundaries, geography, climate, and cooling technology shape reported estimates. The aim is to clarify the assumptions behind widely cited numbers rather than accept or dismiss them. Focusing specifically on water, the project asks where withdrawals occur, when consumption happens, and under what local conditions impacts become significant. The same computational workload can look very different depending on whether it is cooled with evaporative systems in water‑stressed regions or with closed‑loop systems in water‑abundant areas. Instead of relying mainly on per-query comparisons, this research project is trying to develop a framework that situates water use within defined physical boundaries, regional hydrology, and explicit uncertainty ranges. The project also considers how current narratives around AI water use resemble earlier environmental debates, including early discussions around nuclear energy, where valid concerns were sometimes shaped by incomplete or mismatched system comparisons. Approaching the issue as a physics student means postponing judgment and focusing on clear definitions, transparent assumptions, and consistent magnitude comparisons. The presentation will report progress and highlight remaining technical questions.
Artificial Intelligence and Water Use: A Systems-Level Analysis of Assumptions, Accounting, and Impact
CNS Atrium
Public discussion about artificial intelligence and water use often centers on striking claims about water consumed per prompt or per training run. This research project begins a critical examination of those claims by asking what is actually being counted as “water use.” It distinguishes among on-site cooling water, electricity-related water, and water embedded in hardware manufacturing, and examines how system boundaries, geography, climate, and cooling technology shape reported estimates. The aim is to clarify the assumptions behind widely cited numbers rather than accept or dismiss them. Focusing specifically on water, the project asks where withdrawals occur, when consumption happens, and under what local conditions impacts become significant. The same computational workload can look very different depending on whether it is cooled with evaporative systems in water‑stressed regions or with closed‑loop systems in water‑abundant areas. Instead of relying mainly on per-query comparisons, this research project is trying to develop a framework that situates water use within defined physical boundaries, regional hydrology, and explicit uncertainty ranges. The project also considers how current narratives around AI water use resemble earlier environmental debates, including early discussions around nuclear energy, where valid concerns were sometimes shaped by incomplete or mismatched system comparisons. Approaching the issue as a physics student means postponing judgment and focusing on clear definitions, transparent assumptions, and consistent magnitude comparisons. The presentation will report progress and highlight remaining technical questions.