Rethinking Power Conversion and Distribution for the AI Era
Ehsan Nasr of Microsoft explores how data centers must rethink power systems for the AI era, focusing on the adoption of solid-state transformers, centralized high-voltage DC delivery, and innovative architectures to enable high-density, AI-ready infrastructure.
Rethinking Power Conversion and Distribution for the AI Era
By Ehsan Nasr, Director, Energy Technology, Microsoft
Introduction
As the adoption of AI accelerates, data centers are facing dramatic increases in IT equipment density—especially with large-scale GPU cluster deployments. This surge places substantial new demands on power conversion and distribution systems, challenging legacy electrical architectures and necessitating a fresh approach.
The Inflection Point
Traditional data center power design, centered on AC transformers and distributed conversion, struggles to keep pace with the performance requirements of high-voltage GPU racks. Questions arise: Can current infrastructure support growing rack loads, or is a full-stack architectural rethink essential for both performance and energy efficiency?
Microsoft’s Approach: Disaggregated Power Architecture
Microsoft has initiated a paradigm shift by moving power conversion out of individual servers and into centralized rack-level modules. The Mt Diablo disaggregated power specification, announced in partnership with Meta and Google, enables:
- Centralized power delivery at ±400VDC and 800VDC
- More efficient, scalable high-capacity power distribution across racks
- Enhanced support for AI and future high-density workloads
Solid-State Transformers (SSTs): Core Technology Enabler
Solid-state transformers represent a new class of technology that operates at much higher frequencies than conventional units (tens to hundreds of kHz vs. 50/60 Hz). Benefits include:
- Direct high-voltage DC to racks: SSTs can handle medium-voltage AC (<35kV) and deliver DC (<1500VDC) straight to IT racks
- Infrastructure simplification: Reduces need for MV-to-LV transformers and associated equipment, shrinking footprint
- Improved energy efficiency: Lowers conversion losses, potentially reduces costs, boosts ROI
- Flexibility: Supports integration with energy storage and backup for enhanced reliability
- Compactness: Much smaller and lighter than traditional chains, ideal for constrained spaces
Looking Ahead
The adoption of SSTs is just one component in a broader architectural transformation. By reimagining everything from power distribution to protection and conversion, Microsoft aims to build data centers with the resilience, efficiency, and flexibility required for ongoing AI-driven progress and extreme rack density.
Next Steps
- Explore further innovations such as SSTs for data center modernization
- Engage with the broader ecosystem driving power technology advancement
- Watch for case studies, open specifications, and benchmarks as AI-scale requirements mature
For further reading, see Microsoft’s Mt Diablo Disaggregated Power Platform and detailed technical spec. CONNECT with leaders shaping the future of AI infrastructure.
This post appeared first on “Microsoft Tech Community”. Read the entire article here