Liquid cooling redefines data center efficiency beyond PUE

Schneider Electric promotes shifting data center efficiency metrics from PUE to PCE and WUE, advocating liquid cooling as the default for AI infrastructure.

• Main announcement/action: Schneider Electric argues that data center efficiency measurement must move beyond PUE to include Power Compute Effectiveness (PCE) and Water Usage Effectiveness (WUE), and it promotes liquid cooling as the preferred solution for AI workloads. Key facts: Goldman Sachs estimates 76% of AI servers will be liquid-cooled by end of 2026, liquid cooling can enable PUE ~1.05–1.15, and it supports rack densities >100 kW. Publication date: May 14, 2026.
• Background and details: The article explains how PCE measures useful computational work from provisioned power and highlights outcome metrics tokens per watt and cost per token; it cites AWS fleet-wide PUE near 1.15, discusses single-phase direct-to-chip (DTC) cooling shifting water use to heat-rejection layers, and emphasizes system-level design (cooling + power + compute) rather than treating cooling as a standalone product.