Is Power Usage Effectiveness (PUE) Still Fit for Purpose? An Integrative Review of Data Center Metrics for Assessing the Environmental Impact of Modern AI Infrastructure

Résumé fourni par la source

Current U.S. policy directives indicate a "full steam ahead" approach to maintaining global dominance in Artificial Intelligence (AI). This strategic posture accelerates both technological innovation and the rapid expansion of data centers across the country. However, the proliferation of AI is driving a fundamental shift in data center design, introducing extreme power densities and specialized computing hardware. These advancements directly challenge traditional efficiency benchmarks, placing the long-standing industry standard—Power Usage Effectiveness (PUE)—under growing scrutiny regarding its relevance in the AI era. This paper presents an integrative review of current literature to critically assess PUE’s suitability for modern AI infrastructure and to synthesize emerging alternative metrics.This paper presents a review of current literature to critically assess the suitability of PUE for modern AI infrastructure and to identify and synthesize emerging alternative efficiency metrics. The review identifies a suite of complementary metrics—including those for water usage (WUE), carbon usage (CUE), and IT equipment utilization—that could offer a more holistic evaluation of data centers’ impact on the environment. The goal of this review is to ensure that U.S. leadership in AI does not come at the cost of ecological overshoot or expense of public good but instead models sustainable, practical, and equitable innovation that can endure across changing dynamics under different political administrations.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.