AI's Hidden Thirst

Understanding AI Data Center Water Demand

As generative AI expands, the demand for digital services is driving a rapid construction boom of hyperscale data centers. A hyperscale data center handling billions of daily requests can use one million to five million gallons of water daily, comparable to the total daily residential water consumption of a mid-sized city in Minnesota. Along with information gaps between developers and local municipalities, attempts to save water consumption through internal recycling (circularity) can concentrate heavy metals, biocides, corrosion products, and treatment chemicals in centers’ wastewater, shifting part of the burden from water quantity to water quality.

To address the data gaps in water demand for hyperscale data centers, this project builds upon the foundational IonE Mini Grant that modeled the micro-level water consumption of graphic processing unit (GPU)-based AI workloads. Co-designed with the Freshwater Society and the UMN Morris Clifford J. Benson Center for Community Partnerships, the project extends a proof-of-concept from GPU-level modeling to a facility-level, community-driven predictive decision support dashboard. Freshwater Society and Benson Center will help design stakeholder metrics, community impact reports and educational materials.

Project Goals

• Co-design the decision-support dashboard’s metrics and compliance thresholds with the Freshwater Society and the UMN Morris Benson Center for Community Partnerships
• Develop a deterministic, micro-to-macro water quantity model that translates algorithmic GPU workloads into facility-level water consumption and evaporation rates
• Develop chemical concentration and thermal models to predict water quality impacts and inform the protection of local aquatic ecosystems
• Design and prototype a decision-support dashboard that integrates the deterministic water quantity and quality models, which utilizes Large Language Models (LLMs) strictly as a secure translation layer, to provide private optimization tools for data center operators, alongside transparent compliance metrics for regulators and communities
• Develop a community report on data transparency to empower civic groups to educate residents
• Design high school educational modules deployed in rural Minnesota