On September 4, seven Chinese government departments issued a joint implementation plan for 2026-2030, explicitly proposing the exploration of an 800V high-voltage direct current (HVDC) power supply architecture for computing facilities. This policy shift is driven by the increasing power demands of next-generation AI computing, as traditional 54V rack-level power supplies face limitations, making 800VDC essential for reducing intermediate conversion losses and supporting ultra-high-density deployments.
Overview: The Shift to 800V in AI Data Centers
On September 4, 2026, seven Chinese government departments, including the Cyberspace Administration of China, released the Implementation Plan for Promoting the Coordinated Transformation and Development of Digitalization and Greenification (2026-2030). The document explicitly proposes exploring an 800V high-voltage direct current (HVDC) power supply architecture for computing facilities to support ultra-high-density deployments exceeding 100kW per rack. This policy aligns with global industry shifts, notably Nvidia's updated 800VDC white paper in August, which transitioned focus from theoretical necessity to deployment and validation. Nvidia is currently advancing 800VDC standardization with Google, Microsoft, and the Open Compute Project (OCP), with over 80 equipment manufacturers participating in the ecosystem.
Technical Drivers: Overcoming the 54V Bottleneck
The transition to 800V is primarily driven by the escalating power demands of next-generation AI computing. Traditional 54V rack-level power supplies face severe limitations in high-density environments. By adopting 800VDC, data centers can significantly reduce intermediate conversion steps, lower transmission currents, and decrease copper material usage, thereby enhancing overall power path efficiency. Furthermore, 800V necessitates a fundamental restructuring of the data center power system hierarchy. Nvidia's roadmap includes rack-level Power Racks, row-level Power Centers, and future data center-level DC Power Blocks. The Power Rack, compatible with existing AC infrastructure, is slated for actual deployment in the second half of 2026, while the Row Power Center is expected to achieve larger-scale deployment conditions by 2027.
Policy Framework: Integrating Green and Digital Transformation
The significance of the seven-department document extends beyond mere energy conservation. It integrates liquid cooling, low-power chips, 800V HVDC, ultra-high-density racks, computing-power-electricity coordination, direct green power connection, and new energy storage into a unified framework for green and low-carbon development. By explicitly mentioning the exploration of 800V architecture, the policy signals that domestic computing infrastructure design is formally accommodating higher power densities. Additionally, the document advocates for aggregating computing power in clean energy-rich regions and upgrading outdated facilities, indicating that 800V is no longer just a single-platform technical route but a nationally guided application direction.
Industry Impact: The Three-Tier Supply Chain Ecosystem
The 800V ecosystem can be categorized into three critical tiers:
- Power Systems and Power Racks: Companies like Megmeet have disclosed collaborations with Nvidia on 800VDC Sidecar solutions for megawatt-level AI racks. Zhongheng Electric reported a 98.48% year-on-year revenue increase in data center power to 808 million RMB in the first half of 2026, though 800V products remain in the industrialization phase. Overseas, Delta has showcased a 660kW 800V In-Row Power Rack, while LITEON expects to complete validation by late 2026.
- DC Protection and Supercapacitors: Liangxin has developed 600A and 800A computing-specific DC contactors and solid-state breakers, while Jianghai's supercapacitors address rapid AI load fluctuations.
- Solid State Transformers (SST): Jinpan Technology has completed SST prototype designs for 10kV AI infrastructure, and Sifang is promoting SST market applications, representing a longer-term architectural shift.
Implications for AI Data Center Operations
For AI data center operations, the 800V transition dictates a new paradigm of power distribution and thermal management. The push for single racks exceeding 100kW mandates seamless integration with liquid cooling systems, as seen in LITEON's 800V liquid-cooled Power Rack. The shift also emphasizes power buffering capabilities to handle instantaneous power demands of AI workloads, making supercapacitors indispensable. Moreover, the policy's emphasis on computing-power-electricity coordination implies that future data centers will not only consume power more efficiently but also align their operational loads with direct green power availability, fundamentally altering site selection and operational strategies.
Future Outlook: From Concept to Mass Deployment
The market focus has shifted from questioning the viability of 800V to identifying which companies can secure customer validation and mass orders. In the short term, Power Racks, power modules, HVDC, and DC protection are closest to order realization. Medium-term value will accrue to busbars, connectors, and high-voltage power devices. Ultimately, SST and energy storage will integrate into broader data center power systems. While 800V is a highly deterministic industry trend, order realization speeds vary significantly among companies. The critical milestones for the second half of 2026 through 2027 will be customer certifications, batch deliveries, and revenue proportion increases, marking the transition from thematic trading to genuine industrial validation.