Search for a command to run...
Data centers' power and water use have surged substantially as demand for artificial intelligence (AI) accelerates. The International Energy Agency projects that both electricity and water consumption will more than double by 2030, creating significant pressure on stressed regional resources. Almost two-thirds of new U.S. data centers are located in water-stressed areas such as Texas, Arizona, and California, underscoring the importance of efficiently managing power and water resources. In response, this study creates a predictive control framework for the combined management of water and energy systems in an AI data center. The initial stage involves day-ahead optimization to generate an optimal control sequence for both systems. Subsequently, a real-time control approach (5-minute interval) grounded in nonlinear model predictive control (NMPC) is developed to follow the day-ahead plan while accounting for dynamic pricing and operational constraints. The initial findings indicated that data center workloads can be considered a flexible load to reduce water and energy use. Moreover, real-time control effectively satisfied demand while managing multiple resources, despite fluctuations in renewable energy availability and electricity prices. Future work will extend this approach by embedding a resilience index into the NMPC formulation, incorporating state estimation to infer unobservable states, and deploying distributed NMPC across geographically dispersed AI data centers, enabling coordinated workload routing under regional water constraints and renewable uncertainty.