NVIDIA Introduces 800 V HVDC Architecture for Future AI Data Centers

The exponential growth of artificial intelligence (AI) workloads is driving a significant increase in data center power requirements. Traditional power distribution methods, such as the 54 V in-rack systems, are becoming inadequate for the upcoming megawatt-scale racks designed for modern AI factories. To address this challenge, NVIDIA has announced its transition to an 800 V high-voltage direct current (HVDC) architecture, which aims to support data centers with 1 MW IT racks and beyond, starting in 2027.

Challenges of Current Power Systems

Existing AI data centers rely heavily on 54 V DC power distribution, which involves using large copper busbars to transport electricity. This method is becoming less viable as racks surpass 200 kilowatts due to space constraints and the excessive amount of copper required. A single 1 MW rack demands up to 200 kg of copper, highlighting the inefficiency and unsustainability of current systems.

800 V HVDC: The Future of AI Factories

NVIDIA’s 800 V HVDC architecture provides a streamlined solution by minimizing energy conversions and reducing the number of power supply units needed. This new system enhances reliability, lowers heat dissipation, and improves energy efficiency. The architecture also allows for better voltage stability and reduces material costs by decreasing copper usage.

Collaborative Efforts for Innovation

NVIDIA is collaborating with industry leaders such as Infineon, STMicroelectronics, and Schneider Electric to drive the adoption of this innovative power infrastructure. The initiative aims to establish high-efficiency, scalable power delivery systems tailored for the next generation of AI workloads, ensuring greater reliability and reduced infrastructure complexity.

Efficiency and Scalability

The 800 V HVDC system supports racks ranging from 100 kW to over 1 MW, allowing for seamless expansion. It improves end-to-end efficiency by up to 10% compared to current systems and significantly reduces copper usage and thermal losses. This architecture is designed to meet the demands of evolving data centers, ensuring future-proof scalability.

Addressing Deployment Challenges

Although higher-voltage DC architectures have faced adoption challenges in the past, the convergence of AI-driven rack densities and advancements in power conversion is changing the scenario. NVIDIA and its partners are actively addressing safety, standards, and workforce training to facilitate the deployment of 800 V HVDC at the facility level.

According to [NVIDIA](https://developer.nvidia.com/blog/nvidia-800-v-hvdc-architecture-will-power-the-next-generation-of-ai-factories/), the full-scale production of 800 V HVDC data centers is expected to coincide with the release of NVIDIA Kyber rack-scale systems in 2027. This development is set to support increasingly demanding AI models and ensure sustainable growth in the sector.

Image source: Shutterstock