Huawei Launches Revolutionary SuperPoD Architecture at Connect 2025
At the highly anticipated Huawei Connect 2025 event in Shanghai, Huawei unveiled groundbreaking innovations that are set to redefine AI and enterprise computing. Yang Chaobin, Huawei’s Director of the Board and CEO of the ICT Business Group, took center stage to officially launch the company’s open-access SuperPoD architecture. Supported by a cutting-edge line of SuperPoD-based products, the architecture is poised to deliver a powerful backbone for artificial intelligence (AI) infrastructure and enterprise operations, allowing Huawei and its partners to push the boundaries of computing potential.
An Efficient and Scalable Paradigm for AI Growth
Pexels
The SuperPoD architecture represents a revolutionary shift in large-scale computing paradigms, addressing the escalating demand for efficiency and low-latency solutions in AI-driven workloads. Traditionally, the expansion of computing clusters came with limitations; as additional servers were stacked, compute utilization dropped, and model training became increasingly unreliable. In contrast, Huawei’s SuperPoD resolves these issues through its UnifiedBus interconnect protocol, facilitating deep interconnection between physical servers. This enables them to behave like a singular logical server for tasks such as learning, reasoning, and scaling data processing, enhancing both reliability and scalability.
The SuperPoD offers strong resource pooling and seamless long-term stability, allowing compute and storage units to interconnect with high bandwidth and low latency. Yang Chaobin highlighted that this innovation is critical for supporting industries with rapidly growing AI needs, eliminating bottlenecks while improving data center performance. Already, Huawei’s Atlas 900 A3 SuperPoD units have proven capable, shipped to over 20 customers across sectors like finance, telecommunications, and manufacturing.
A Suite of Innovatively Engineered Products
Pexels
Huawei’s Connect 2025 event also served as the launchpad for an impressive lineup of SuperPoD-based products. These include the liquid-cooled Atlas 950 SuperPoD, ideal for ultra-large-scale AI computing tasks, featuring an orthogonal architecture to achieve cableless interconnection. With its ability to scale up to 8,192 NPUs without convergence, this product ensures uninterrupted high-performance computing for heavy workloads.
Another product, the Atlas 850, is suited for enterprises and meets multiscenario inference needs with its flexible multi-cabinet deployment. Further complementing the lineup is the Atlas 350 accelerator card and the TaiShan 950 SuperPoD, the industry’s first general-purpose SuperPoD that excels in handling database operations, virtual machine migrations, and big data processing. These innovations affirm Huawei’s role in shaping future-ready technologies for industries ranging from energy to telecommunications.
Open Access for Collaborative Innovation
Pexels
As part of its strategy to foster ecosystem growth, Huawei is taking a unique approach by opening its SuperPoD hardware and software to partners and developers. Customers and collaborators will gain access to vital components of the UnifiedBus protocol and SuperPoD reference architecture, as well as AI cards, NPU modules, and other hardware tools. This approach empowers enterprises to customize and develop their UnifiedBus-based solutions, further expanding the capabilities of the SuperPoD system.
On the software front, Huawei has made its UB OS Component entirely open-source, allowing seamless integration into upstream open-source communities like openEuler. This initiative promotes flexible and independent software development for partners, fostering innovation and ensuring that SuperPoD solutions remain compatible with emerging technological advances. As Yang Chaobin emphasized, this commitment to openness will enable the rapid growth of a thriving ecosystem fueled by collaboration and mutual innovation across industries.
Huawei’s SuperPoD marks a significant leap forward in AI and general-purpose computing technology. Through innovative hardware and open-source collaboration, it not only meets today’s industry challenges but also anticipates the needs of burgeoning AI and enterprise data centers. The implications of this launch for industries worldwide are profound, offering scalable, efficient, and future-proof solutions to power the next era of digital transformation.
