On September 2, the 2026 Open Data Center Conference and inaugural Computing Power Expo, or ODX 2026, opened in Beijing. H3C Group showcased products and technical solutions spanning supernodes, high-speed interconnects, 800V high-voltage direct-current power, liquid cooling and AI-native storage.

Liu Xinmin, senior vice president and vice chairman of H3C Group’s Technical Committee, told the conference’s main forum that standardized interfaces and open collaboration could reduce the complexity of software and hardware integration and large-scale deployment for AI infrastructure.

Advancing Supernode Interface Standardization

As AI computing clusters grow, different types of computing chips vary in power consumption, cooling requirements, dimensions and interconnect methods. Equipment from different vendors also faces inconsistent interfaces and high integration costs.

H3C said it supports UXM standardized module technology and is participating in the research and development of related standards. By standardizing power delivery, interfaces and high-speed interconnects, the solution can accommodate PCIe, BFX and OAM modules, enabling different XPU modules to be decoupled and upgraded in stages.

The standards and technologies remain under development and require further industry collaboration. Their effectiveness in enabling cross-vendor compatibility will need to be validated through subsequent product integration and real-world deployments.

G-Link Targets High-Speed Scale-Up Interconnects

For high-speed interconnects, H3C is participating in the development, application and standardization of the G-Link general-purpose high-speed interconnect architecture.

The architecture targets AI high-performance computing and Scale-Up interconnect scenarios, using open interconnect interfaces to link different computing chips and systems.

At the physical connection layer, H3C is advancing NPO interface standardization while exploring orthogonal architectures, Cable Tray designs and other interconnect approaches. The company said its layered, decoupled design is compatible with 112G and 224G interfaces and leaves room for an eventual transition to 448G SerDes.

H3C has also proposed layered, decoupled standards for short-reach, high-speed NPO and CPO Scale-Up interconnects, allowing compute and connectivity components to be upgraded independently. However, 448G SerDes and the related layered interconnect standards remain part of a future technology roadmap and have not yet been deployed at scale.

H3C Unveils Full-Stack 800V HVDC Power Solution

To meet the power requirements of AI servers and high-density racks, H3C showcased a full-stack 800V high-voltage direct-current power solution with modular on-board power components.

The solution includes an 800V Powershelf, common redundant power supplies, or CRPS PSUs, and node power distribution boards, or PDBs. The Powershelf is available in both air-cooled and liquid-cooled designs.

H3C said the solution combines vertical power delivery, or VPD, with an 800V HVDC architecture. By increasing the supply voltage and reducing the number of power-conversion stages, it can cut line losses by 9% to 12% and deliver end-to-end power efficiency of more than 92%.

The efficiency and loss figures come from technical materials disclosed by H3C. Actual performance may vary depending on server configuration, rack power, load levels and testing conditions.

The company said the solution targets high-density computing scenarios where individual chips consume more than 2,000 watts and rack power rises toward 500 kilowatts or even megawatt levels. These figures are design targets for some next-generation high-performance chips and high-density racks, not typical power levels across today’s AI servers.

Liquid Cooling Expands Across Server Components

For thermal management, H3C showcased liquid-cooling solutions covering compute nodes, busbars and Powershelves.

The company said the solution can cover server components including CPUs, memory and power boards. It is also working to bring native liquid-cooling designs to optical modules, solid-state drives and network interface cards.

H3C also said it is helping develop standards for liquid-cooling interfaces and components, including standardized quick-connect fittings and cold-plate designs, to improve compatibility across equipment and components.

The actual coverage and deployment scale of these liquid-cooling products will need to be confirmed based on specific server models, customer projects and subsequent product documentation.

H3C and Xiaohongshu Build DDC-Based AI Computing Network Cluster

During the conference, H3C also detailed an AI data center networking cluster project jointly developed with Xiaohongshu using a DDC architecture.

According to H3C, the project uses a DDC architecture to build a lossless AI computing network cluster for large-scale AI training, with a focus on improving network throughput, latency, reliability and bandwidth utilization.

The company said the project can support access to multiple types of heterogeneous computing resources while reducing the network-tuning work required during cluster deployment. The project’s scale, performance metrics and deployment results still need to be verified against future disclosures from both companies.

Contributing to AI Factory Technical Specification V1

The ODCC AI Factory task force was established during the conference and publicly released AI Factory Technical Specification V1 for the first time.

Zheng Guoliang, director of the Server Systems Department within H3C’s Cloud, Computing and Storage Product Line, attended the related event. According to H3C, the company will participate in the task force’s subsequent standards development and technical validation work.

The specification has been made public but remains under development. Its eventual scope and effectiveness will depend on the standard-setting process, industry participation and validation through real-world projects.

H3C Showcases UniPoD S80000 Supernode

H3C also showcased its UniPoD S80000 supernode at the conference, along with AI computing networks, AI servers, AI-native storage and other products.

Through coordinated design across computing, networking, storage, power and cooling systems, the company aims to reduce the complexity of integrating and deploying high-density AI infrastructure.

However, H3C’s disclosures were based primarily on its conference materials. Shipment volumes, customer deployments and commercialization progress for some products have not been disclosed and will require further confirmation through product documentation, customer case studies and company announcements.