AI-driven PUE optimization in hyperscale data centers using real-time telemetry

Abstract

The rapid growth of hyperscale data centers has significantly increased energy consumption, making power usage effectiveness (PUE) one of the most critical metrics for evaluating operational efficiency and sustainability. This study presents a real-world case study of a hyperscale data center environment and investigates the potential of artificial intelligence (AI)-driven optimization using real-time telemetry data. Operational reports, rack-level power measurements, and facility energy consumption records were analyzed to evaluate current performance and identify opportunities for efficiency improvement. Historical operational analysis revealed a substantial reduction in PUE from approximately 2.70 under legacy operating conditions to 2.01 following infrastructure modernization and operational optimization initiatives. To support future efficiency improvements, a telemetry-driven AI optimization framework was developed as a decision-support architecture for adaptive cooling and energy management. The framework integrates real-time operational telemetry, including IT load, facility power consumption, environmental conditions, cooling demand, and historical PUE behavior, to support intelligent operational decision-making. A synthetic six-month time-series dataset with 5-minute intervals was generated based on observed operational patterns to enable long-term analytical evaluation. The primary contribution of this research is the development of a telemetry-driven AI optimization framework that combines operational data analysis, sustainability-oriented decision support, and adaptive cooling optimization within a hyperscale data center environment. Analytical evaluation and benchmark comparison indicate that AI-assisted optimization could potentially reduce PUE to a projected range between 1.4 and 1.6. These results represent optimization scenarios rather than experimentally validated production outcomes. The findings highlight the potential of AI-enabled infrastructure management to improve operational efficiency, sustainability performance, and strategic decision-making within hyperscale data center environments.