AI

2025 Electronics Design Trends: 800V HVDC Architecture Emerges for AI Data Center Power Delivery

Global data center electricity consumption reached 240-340 TWh in 2022. Projections indicate growth to 1,000 TWh by 2026, driven primarily by AI workloads. Rack power densities have increased from 30-40 kW per cabinet to 100 kW or more for AI systems. A 1 MW rack using traditional 48V/54V distribution requires over 200 kg of copper cabling due to I²R losses, making scaling unsustainable in weight, cost, and space.

AI-driven data center power demand could double by 2030 (IEA: 415 TWh in 2024 to 945 TWh). Racks jump from 10-30 kW to 300-800 kW. Enterprises counter rising costs/ESG pressure via high-efficiency hardware, HPR racks, liquid/immersion cooling, and HVDC cutting PUE and power use up to 40%.

NVIDIA has introduced an 800V high-voltage direct current (HVDC) architecture to address these constraints. This system converts medium-voltage AC (13.8-34.5 kV) directly to 800V DC using solid-state transformers (SSTs), then distributes it to racks for single-stage step-down to GPU voltages via DC-DC converters. Benefits include up to 45% reduction in copper thickness, 5% improvement in end-to-end efficiency, 70% lower maintenance costs from fewer power supply unit failures, and reduced cooling requirements.

To accelerate NVIDIA's 800 VDC data center adoption for megawatt-scale AI racks, key partners of Nvidia disclosed include:
Silicon providers...

This article requires a Silver or Gold membership
Silver+ members only
Continue reading with a subscription.
1 USD = Rs 95.46
  • Deep-technical articles & analysis
  • New-product comparisons
  • Premium online courses
  • 1 author article / year