The AI DC Innovation Summit, held during Huawei Connect 2026 under the theme "Leading AI DC Innovation, Winning Together in the Agentic Era," drew more than 600 industry leaders, experts, and academics from around the world for discussions on data center infrastructure and sustainable development in the age of AI agents.

At the event, Huawei launched a newly upgraded Xinghe AI data center network solution designed to improve the utilization of computing power resources, keep workloads running without interruption, and help enterprises build efficient computing power centers for AI and agent applications.

Data Center Networks Become a Critical Link in AI Computing Power Infrastructure

Zhang Bai, President of the Data Center Network Domain at Huawei's Data Communication Product Line, said AI is pushing data centers to evolve into computing power centers. In the Agentic era, agents continuously call on models, tools, and data, making the data center network key infrastructure for unlocking computing power and raising token production efficiency.

The upgraded Xinghe AI data center network solution combines core capabilities including the Panshi high-reliability architecture 2.0, Xingyi Digital Map 2.0, the Xingyu hyper-converged architecture, and network-level packet load balancing (NPLB).

Huawei also upgraded its NetMaster network agent, which can automatically diagnose 95% of faults and pinpoint root causes within minutes, helping operations teams manage networks and handle failures more efficiently.

CloudEngine SF9300 Series Switches Launched

On the hardware side, Huawei debuted the UB (UnifiedBus) network switch CloudEngine SF9300 series, its first global launch.

The series uses a two-layer multi-plane architecture that can cut network construction costs by up to 30%, while UB streamlined forwarding technology reduces end-to-end latency by 40%.

For reliability, the CloudEngine SF9300 series introduces link layer retransmission (LLR). When link bit errors or localized faults occur, the system can retransmit data within microseconds, achieving zero packet loss during link flapping and improving the stability of AI training and inference workloads.

100T/51.2T NPO Switches Introduced

Huawei also released the fully in-house developed 100T/51.2T NPO switch CloudEngine XH9300 series.

The product uses a consolidated light source design and carries a proprietary 3.2T OE optical engine, cutting interconnect power consumption by 40%. Near-package optics remove the oDSP signal processing step, reducing forwarding latency by 26%.

On the maintenance side, the OE optical engine in the CloudEngine XH9300 series uses a pluggable snap-in design that can improve equipment servicing efficiency by up to tenfold, easing operations complexity and downtime risk in high-density AI data centers.

Financial Data Center Network High Availability Standard Released

During the summit, the Beijing Financial Technology Industry Alliance, together with Huawei and 16 other contributing organizations, released a group standard titled Technical Specifications for High Availability of Financial Data Center Networks.

The standard sets out technical requirements and test methods for network high availability in financial data centers, covering both intra-data center networks and inter-data center networks.

It divides financial data center networks into two categories and five network types, and defines high availability systematically across three dimensions: architectural redundancy, recovery from faults and sub-healthy states, and fault prevention.

The standard gives financial institutions a technical basis for safeguarding business continuity and encourages closer alignment between financial network infrastructure and AI technology.

Institute of High Energy Physics Lifts Computing Power Efficiency by More Than 10%

At the summit, Qi Fazhi, Director of the Computing Center at the Institute of High Energy Physics under the Chinese Academy of Sciences, shared details of the institute's work with Huawei.

The institute has deployed Huawei's Xinghe AI high-efficiency data center network solution. Using a network-wide load balancing algorithm, the solution raised computing power efficiency by more than 10% and eased problems such as traffic concentration and uneven loads in AI training jobs.

With the network optimized, massive volumes of observational data can be transmitted more reliably and efficiently, supporting researchers processing deep-space signals and conducting high energy physics research.

Continued Push on AI Data Center Network Upgrades

Huawei said it will maintain its open collaboration approach, working with customers and partners to advance data center network technology and continuing to invest in highly reliable networks, intelligent operations, optical-electrical convergence, and efficient interconnects.

As AI models grow larger and agent applications roll out faster, data centers will shift from platforms that simply host computing power to comprehensive infrastructure built for token production and the execution of intelligent tasks. Network performance, stability, and operational efficiency will become major factors in the overall output of AI computing power centers.