Huawei Challenges Nvidia with SuperPoDs & SuperClusters
-
Published Time: 2025-09-23T12:30:05Z
Article
AI Strategy
Huawei: Challenging Nvidia with SuperPoDs & SuperClusters
By Kitty Wheeler
September 23, 2025
3 mins
Share
Eric Xu, Huawei's Rotating Chairman, gives the keynote at HUAWEI CONNECT 2025 (Credit: Huawei)
Huawei launches its Atlas SuperPoDs containing up to 15,488 neural processing units and SuperClusters with over one million NPUs at Huawei Connect 2025
Company portals
-
HUAWEI
Tags
Hyperscale, AI, HPC, Compute, Data Centres, Data Centers, China, Huawei
Huawei Technologies, the Chinese telecommunications and technology equipment manufacturer, is challenging Nvidia.
The company is unveiling its Atlas SuperPoD and SuperCluster portfolio at Huawei Connect 2025 in Shanghai as a direct challenge to Nvidia’s dominance in AI computing infrastructure.
Eric Xu, Huawei’s Deputy Chairman of the Board and Rotating Chairman, announced the systems during his keynote speech – positioning the launch as Huawei’s response to surging global demand for AI computing power, particularly within China’s expanding AI sector.
“Computing power is – and will continue to be – key to AI. This is especially true in China,” he tells the audience.
The announcement is happening as technology companies navigate ongoing semiconductor export restrictions that have limited access to advanced processing units.
Huawei’s strategy centres on utilising manufacturing process nodes available to Chinese operations, ensuring sustainable supply chains for its data centre customers.
“Chips are the building blocks of computing power. And at Huawei, Ascend chips are the foundation of our AI computing strategy,” Eric explains.
How Atlas systems target hyperscale data centres
Huawei introduced two SuperPoD configurations designed as single logical machines comprising multiple physical systems.
The Atlas 950 SuperPoD contains 8,192 Ascend Neural Processing Units, while the Atlas 960 SuperPoD incorporates 15,488 Ascend NPUs.
These specialised processors accelerate machine learning (ML) and deep learning computations through optimised architectures for parallel processing.
The systems deliver performance across four metrics: NPU count, total computing power, memory capacity and interconnect bandwidth.
Eric states that based on publicly available roadmaps from competing companies, the Atlas SuperPoDs represent the most powerful systems in their category globally.
Beyond individual SuperPoDs, Huawei is announcing SuperCluster systems that combine multiple units for exponential scaling.
The Atlas 950 SuperCluster contains over 500,000 Ascend NPUs, while the Atlas 960 SuperCluster incorporates over one million NPUs.
Eric Xu, Huawei's Deputy Chairman of the Board and Rotating Chairman (Credit: Huawei)
These configurations target hyperscale data centres operated by cloud service providers requiring massive computing capacity.
The company is also introducing the TaiShan 950 SuperPoD for general-purpose computing workloads beyond AI training.
This system operates alongside Huawei’s GaussDB database platform, providing an alternative to traditional mainframes and Oracle’s Exadata systems for enterprises requiring unified AI and database infrastructure.
UnifiedBus’ role in addressing infrastructure challenges
Large-scale AI infrastructure faces significant interconnect challenges.
Traditional optical and copper cables encounter physical limitations when connecting thousands of processing units across extended distances while maintaining low-latency connections.
Huawei developed UnifiedBus, a proprietary interconnect protocol leveraging the company’s three decades of connectivity expertise.
The protocol underpins SuperPoD performance and enables reliable communication across distributed systems.
Huawei's Shenzhen HQ campus (Credit: Huawei)
At the Shanghai event, Eric revealed UnifiedBus 2.0 specifications and encouraged industry adoption.
The company aims to establish an open ecosystem enabling additional vendors to develop compatible components, mirroring industry efforts to create standardised high-performance computing interconnects.
“With the world’s most powerful SuperPoDs and SuperClusters,” Xu states, “Huawei has what it takes to provide abundant computing power for ongoing, rapid advancements in AI, both now and in the future.”
“SuperPoDs and SuperClusters powered by UnifiedBus are our answer to surging demand for computing, both today and tomorrow,” Eric concludes.
“Our goal is to keep pushing advancements in AI to create greater value.”
Company Portals
-
HUAWEI
Related Content
- How Deloitte Deploys AI for Industrial Carbon Capture Machine Learning
- How AI is Transforming Operations for CSPs Technology
- Bain & Co: AI Infrastructure Needs to Face $800bn Shortfall Data & Analytics
- How AI & Digital Twins are Shaping Sustainable Business Technology
-