At its Huawei CONNECT 2025 event in Shanghai, Huawei unveiled Xinghe AI Fabric 2.0, a full upgrade of its AI Fabric data center networking solution first introduced in 2018. The announcement was made during the inaugural Data Center Innovation Summit, which brought together over 600 global industry leaders, experts, and scholars under the theme “Leading AI DC Innovation for an Intelligent Future.”
Arthur Wang, President of Huawei’s Data Center Network Domain, noted that rapid AI iteration and evolving cloud architectures are driving data center networks to a critical inflection point. Xinghe AI Fabric 2.0 is designed to empower enterprises to build always-on data center networks with full computing power, accelerating digital and intelligent transformation.
Three-Layer Architecture
The new fabric introduces a three-layer architecture comprising AI Brain, AI Connectivity, and AI Network Elements:
- AI Brain – Combines the NetMaster network agent and StarryWing Digital Map to deliver drag-and-drop service orchestration, automated deployment, integrated application-network O&M automation, and end-to-end automation across heterogeneous networks and security domains.
- AI Connectivity – Uses Network Scale Load Balancing (NSLB) algorithms to boost network throughput to 95%, improving training and inference efficiency by more than 10%. A new iReliable three-level reliability technology, built on Huawei’s Rock-Solid Architecture, is said to deliver 10× higher reliability.
- AI Network Elements – Includes the CloudEngine series of general-purpose computing switches, XH-series intelligent computing switches, and StarryLink optical modules. These provide precise traffic awareness and visualization of packet loss and latency, as well as group-based isolation for enhanced security.
High-Density Switching Portfolio
Huawei also showcased an expanded line of ultra-high-density data center switches and cooling systems:
- 128 × 800GE fixed switch – XH9330, claimed to be the industry’s highest-density 800GE fixed switch.
- 64 × 800GE fixed switch – XH9320, along with a complete portfolio of 800GE StarryLink optical modules.
- 128 × 400GE fixed switch – XH9230-LC, featuring all-port liquid cooling for improved energy efficiency and heat management.
- Xinghe liquid-cooled cabinet, designed for dense AI and HPC workloads.
These systems target AI training clusters and high-performance cloud environments where bandwidth, latency, and cooling efficiency are critical.
Performance and Reliability Targets
- Network throughput: Up to 95% utilization.
- Training and inference efficiency: Gains of 10% or higher through advanced scheduling.
- Reliability: Up to 10× improvement versus previous generation via three-level iReliable technology.
- Visibility: Packet loss and latency monitoring with precise traffic visualization.
