Infrastructure and agents take center stage at Huawei Connect 2026

David Wang, Huawei rotating chairman and former ICT Strategy & Marketing President speaks at a Huawei database and storage product launch during a press conference in Beijing, China. — (Image credit: Getty Images)

Huawei unveiled a flurry of enterprise hardware announcements at its annual Connect conference in Shanghai, China, underpinned by a steady current of agentic AI.

Speaking on stage at the Shanghai Exhibition Center during the opening keynote, Huawei rotating chairman David Wang unveiled the Atlas 960E SuperPoD and OceanStor M900 ranges.

According to Huawei, Atlas 960E SuperPoD is the industry's first near-packaged objects (NPO)-based SuperPoD, capable of massively accelerating training and inference for 10-trillion-parameter foundation models.

The star of the show, however, was UnifiedBus – a new interconnect architecture that the company claims will help reduce data bottlenecks in cluster computing.

Huawei targets agentic compute gains

Agents, which can have hundreds of interactions per hour with AI models, add to this problem. Interactions between CPUs and NPUs, Huawei said, generate KV cache that’s also greater than memory capacity, which in turn drives up demand for data storage across compute systems.

In his opening keynote Wang said: “The generation and consumption of intelligence will continue … growing by orders of magnitude, reaching infinite degrees of depth and breadth. These changes will place far higher demands on the scale, performance, and reliability of AI infrastructure. Only by building a powerful AI infrastructure can we lay a rock-solid foundation for this intelligent world.”

Wang explained the problem as it stands. “In a conventional server architecture, inter-server latency is an order of magnitude more than that of intra-server – namely 100s of nanoseconds vs 10s of microseconds,” he said. “Therefore, optimizing server architecture for communication overhead is a critical challenge for AI infrastructure.”

“Huawei’s SuperPoD design is equally connecting all components within a SuperPoD via a unified protocol, supporting unified memory access across physical servers and adopting a copper near optical four interconnect architecture,” he added.

“This enables near linear bandwidth and latency across tens-of-thousands of chips, allowing them to operate as one single computer and effectively boost the MFU. To build this complex system, the first thing you need is the interconnect protocol that equally connects all components within the cluster. This protocol is UnifiedBus.”

UnifiedBus is a new system architecture that, according to Huawei, solves these bottleneck issues with four key features:

A unified protocol and memory semantics
Heterogenous compute collaboration
Tiered storage with global pooling
Optoelectronic interconnect and flexible networking

A unified protocol

According to Yang Chaobin, CEO of Huawei’s ICT business unit, as UnifiedBus brings together ten protocols in one umbrella, round trip time (RTT) latency is reduced from seven milliseconds to two milliseconds, while interconnect bandwidth is increased from 100 Gbits/s to 1 Tbit/s.

Heterogenous compute collaboration

CPUs, NPUs, memory and storage are all directly interconnected under UnifiedBus, which Huawei said enables decentralized, peer-to-peer access between them. It also enables flexible NPU-CPU mixing.

Tiered hardware acceleration, meanwhile, enables Attention-FFN disaggregation.

Tiered storage with global pooling

UnifiedBus uses hybrid-media resource pooling, which enables double data rate memory (DDRM) to be used as an alternative memory source for NPUs.

According to Huawei, this reduces latency for services such as search and recommendation, and doubles the performance of vector retrieval for large, complex data loads.

It also, the company said, reduces per-NPU high bandwidth memory (HBM) capacity requirements for the training of 10 trillion-parameter models and improves cluster model FLOPs utilisation (MFU).

Optoelectronic interconnect and flexible networking

Finally, thanks to its networking features, UnifiedBus acts as a “global ultra-high bandwidth, ultra-low latency ‘data highway’” Huawei said, which the company claimed enables more flexible compute scaling.

“Huawei is committed to innovating at the system level, building a computing-foundation product portfolio, and going open source and open system,” said Yang, concluding his keynote.

“We will continue working with customers, partners, and developers to help this ecosystem thrive and offer the world a new option for compute."

Follow ITPro on Google News and add us as a preferred source to keep tabs on all our latest news, analysis, views, and reviews.

You can also follow ITPro on LinkedIn, X, Facebook, and BlueSky.

TOPICS

Jane McCallion is Managing Editor of ITPro and ChannelPro, specializing in data centers, enterprise IT infrastructure, and cybersecurity. Before becoming Managing Editor, she held the role of Deputy Editor and, prior to that, Features Editor, managing a pool of freelance and internal writers, while continuing to specialize in enterprise IT infrastructure, and business strategy.

Prior to joining ITPro, Jane was a freelance business journalist writing as both Jane McCallion and Jane Bordenave for titles such as European CEO, World Finance, and Business Excellence Magazine.