Technical Details of 800V HVDC Systems

On September 4, 2026, seven Chinese government departments, led by the Cyberspace Administration, unveiled a comprehensive plan to accelerate the green and low-carbon transformation of computing infrastructure. A cornerstone of this initiative is the exploration of an 800V High-Voltage Direct Current (HVDC) power supply architecture. Traditional data centers rely on 12V or 48V DC systems, which suffer from significant transmission losses at high power levels. By upgrading to 800V HVDC, operators can drastically reduce copper usage and conversion losses. This architecture is specifically designed to support next-generation AI clusters, facilitating the deployment of single liquid-cooled racks exceeding 100kW.

Industry Impact on AI Data Centers

The transition to 800V HVDC will profoundly reshape the data center power supply chain and operational economics. Key impacts include:

  • Enhanced Power Density: Enabling ultra-high-density deployments required for advanced AI training clusters and large language models.
  • Improved Energy Efficiency: Reducing end-to-end power conversion losses by up to 15%, aligning with national PUE (Power Usage Effectiveness) reduction targets.
  • Supply Chain Shift: Driving massive demand for high-voltage silicon carbide (SiC) components, advanced busbars, and intelligent thermal management systems.

Future Outlook and Regulatory Support

Looking ahead, the regulatory push signals a definitive shift toward standardized high-voltage architectures in China's digital economy. The government plans to establish pilot zones for 800V HVDC implementation by late 2027. Furthermore, the integration of renewable energy sources will be optimized through the HVDC microgrid framework. Industry analysts predict that this transition will secure China's leadership in AI infrastructure and set a global benchmark for sustainable computing.

Ultimately, the 800V standard will bridge the gap between escalating AI compute demands and stringent environmental regulations, ensuring long-term viability for hyperscale operators.