appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Howard Liu

Huawei Technologies on Thursday unveiled a new computing architecture that it says could make as many as 1 million processors work as a single computer, putting its UnifiedBus interconnect technology at the centre of an ambitious effort to rethink how large data centres are built.
The Peerium Computing Architecture combines nested parallelism, unified memory addressing and peer-to-peer interconnect, with UnifiedBus serving as the high-speed fabric connecting processors, memory, storage and networking equipment.
Huawei said an Atlas 950 SuperCluster containing 256,000 computing cards was already being deployed, while its next-generation Atlas 960 system using near-packaged optics, or NPO, was under testing.
Known in Chinese as Lingqu, UnifiedBus has become central to Huawei’s efforts to extract more computing power at the system level, a strategy that has taken on greater importance as US restrictions constrain the company’s access to advanced semiconductor manufacturing technologies.
Here is what you need to know about UnifiedBus, how it works and why it matters for China’s artificial intelligence ambitions.
What is UnifiedBus?
UnifiedBus is Huawei’s high-speed interconnect protocol designed for what the company terms as “SuperPoDs” – systems that tightly link hundreds or thousands of processors into a single logical computing unit.
Huawei now positions the technology as the key interconnect underpinning Peerium, its newly named computing architecture. UnifiedBus uses a single open protocol to link CPUs, NPUs, memory, solid-state drives, network interface cards and switches, with the longer-term goal of extending the concept of a single computer to systems containing as many as 1 million processors.
In a traditional AI cluster, servers communicate with one another over conventional network connections like Ethernet. While easy to expand by adding more machines, larger clusters also face bandwidth bottlenecks and latency spikes when moving data between processors, leaving expensive AI chips waiting idle for data.
Scale-up interconnects like UnifiedBus aim to address this structural limit directly. By providing higher bandwidth and lower latency between processors, memory and storage modules, UnifiedBus allows different hardware components to share data through unified memory addressing and communicate on a more peer-to-peer basis, according to Huawei.
The protocol’s first commercial implementation, UnifiedBus 1.0, was deployed in Huawei’s Atlas 900 A3 SuperPoD, linking up to 384 Ascend AI processors. It began shipping in March 2025.
Huawei subsequently expanded the scale to 8,192 processors with its Atlas 950 SuperPoD. The 256,000-card system disclosed on Thursday goes further still, although it is a SuperCluster made up of multiple SuperPoDs rather than a single 256,000-card SuperPoD.
Why did Huawei develop it?
US restrictions have made system-level innovation more important to Huawei, but rotating chairman Eric Xu Zhijun rejected the idea that UnifiedBus was merely a workaround for China’s limited access to advanced chipmaking technology.
Asked whether the technology represented a China-specific path or one the wider industry would eventually follow, Xu called UnifiedBus “a path for the future” in the AI era.
“In the years to come, other players would certainly follow suit,” he told the South China Morning Post on Thursday, adding that Huawei made the UnifiedBus protocol open to all because it believed “the whole industry will move toward the same direction”.

Xu acknowledged that Huawei remained constrained by advanced manufacturing. “There is no problem from [a] design perspective for Huawei,” he said, but the company was limited by “the availability of sophisticated process nodes in China”. By making multiple processors work as one computer, Huawei believed it could compete at the system level, an area where Xu said the company had global leadership.
He also stressed that the work predated US sanctions. “Those activities started not after the US sanction, but before,” Xu said. Huawei unveiled its first AI chip and AI strategy in 2018 and began investing more heavily in new computing architectures and interconnect technologies after launching SuperPoD and SuperCluster products in 2019.
Xu said Huawei had invested more than 1.8 trillion yuan (US$268.2 billion) in research and development since 2007, with about 10 per cent going towards basic research.
How about Nvidia’s NVLink and other Western alternatives?
Western tech giants are also pouring resources into high-speed interconnects, which have become a key battleground for next generation AI infrastructure.
Nvidia’s dominance in AI hardware relies heavily on its proprietary NVLink and NVSwitch technologies, which bind multiple graphics processing units (GPUs) inside systems, such as its rack-scale platforms, into a unified domain.
Meanwhile, an industry coalition – including Advanced Micro Devices, Intel, Google, Microsoft, Meta Platforms, Amazon Web Services and Alibaba Group Holding – is backing Ultra Accelerator Link, or UALink, as an open standard to counter Nvidia’s ecosystem lock-in. Alibaba owns the South China Morning Post.
By publishing the technical specifications for UnifiedBus 2.0 late last year, Huawei is seeking to build a domestic ecosystem that encourages Chinese chipmakers, server vendors and software developers to standardise on its architecture.


How Huawei plans to bypass US chip curbs with its new UnifiedBus technology

Once reliant on Samsung and SK Hynix, Chinese smartphone makers are turning to CXMT

Huawei quickens AI chip pace, promises next entrant 3 quarters early

Huawei unveils latest tech to boost AI power in push to break China’s Nvidia reliance

Huawei Technologies

US restrictions have made system-level innovation more important to Huawei. Photo: Shutterstock

The Huawei logo is seen during the KubeCon + CloudNativeCon Europe event in Paris, France, March 20, 2024. Photo: Reuters

The Huawei logo is seen at the World Artificial Intelligence Conference in Shanghai in July 6. Photo: AFP

Huawei showcases its Atlas 950 SuperPoD computing infrastructure at the Huawei Connect 2026 conference in Shanghai on Thursday. Photo: Handout

Chinese optical transceiver makers dodge US ban for now as FCC updates rules

Wiring is seen inside a Microsoft data centre in Mount Pleasant, Wisconsin, September 18, 2025. Photo: Reuters

Optical transceivers facilitate the transmission of data in data centres. Photo: Shutterstock Images

The logo of Zhongji Innolight is seen at the launch of the company’s initial public offering in Hong Kong on July 30. Photo: AFP

AI bottleneck-breaker: China’s Huawei touts high-speed optical module

A man walks by a Huawei store in Beijing on July 8, 2026. Photo: Reuters

NPO technology aims to bring optical components closer to compute chips, reducing signal loss and energy costs. Photo: Shutterstock Images

China targets fourfold boost in AI computing capacity by 2030 in major tech push

China’s expansion in intelligent computing clusters is being driven by a sharp increase in AI workloads. Photo: Getty Images

This photo taken on March 20, 2023 shows servers at the Tianjin Artificial Intelligence Computing Centre in north China’s Tianjin. Photo: Xinhua

Huawei’s new Kirin chip puts LogicFolding to the test, with bigger ambitions in AI

Unveiled on Monday, the Kirin 9050 Pro delivers a 42 per cent performance improvement over its predecessor with better power efficiency, according to Huawei. Photo: Weibo/定焦数码

Huawei’s consumer business group chairman, Richard Yu Chengdong, unveils the Mate XT 2 trifold phone in Guangzhou on Monday. Photo: Handout

Tech

Big Tech

Tech Trends

Tech War

The Chinese tech giant is leveraging its UnifiedBus technology to link hundreds or thousands of chips into single computing unit


The Chinese tech giant is leveraging its UnifiedBus technology to link hundreds or thousands of chips into single computing unit.


How Huawei plans to bypass US chip curbs with its new UnifiedBus technology

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}The Chinese tech giant is leveraging its UnifiedBus technology to link hundreds or thousands of chips into single computing unit

The Chinese tech giant is leveraging its UnifiedBus technology to link hundreds or thousands of chips into single computing unit