A single agentic task can fan out into dozens–or hundreds–of model calls, and each call spends power. In the agentic era, tokens per watt is the metric that governs what an AI data center can produce for the power it draws.

This panel brings together perspectives from across the rack to discuss how rethinking the rack as one system, backed by open standards and common designs, can drive real efficiency gains across every layer.