(P11-3) From Grid to GPU: Rethinking Power Architectures for AI Data Centers

This presentation will examine how power delivery must evolve from medium-voltage distribution through intermediate conversion and point-of-load stages to support higher-density AI workloads. It will explore the growing importance of wide bandgap technologies, especially silicon carbide, in improving conversion efficiency, enabling higher switching performance, and reducing losses in increasingly power-constrained environments. It will also address the system-level tradeoffs among efficiency, thermal design, footprint, and reliability. Attendees will leave with a framework for evaluating future-ready AI data center power architectures and a clearer understanding of why optimizing individual devices is no longer enough.