With the brand new SuperPoDs and SuperClusters, Huawei’s rotating Chairman Eric Xu mentioned the corporate has what it takes to supply plentiful computing energy for speedy developments in AI
In sum – what to know:
Huawei outlines AI roadmap with SuperPoDs and SuperClusters – The corporate plans to launch its Atlas 950 in 2026 and Atlas 960 in 2027, scaling as much as over a million Ascend chips.
UnifiedBus tackles interconnect bottlenecks – Huawei claims its proprietary protocol permits dependable, low-latency connections throughout huge SuperPoD deployments, addressing cable limitations in large-scale AI infrastructure.
SHANGHAI — Chinese language vendor Huawei rotating chair Eric Xu unveiled the corporate’s roadmap for AI computing platforms, with plans to launch what it claims to be the world’s strongest single system within the final quarter of 2026 and double the efficiency with a brand new launch a yr later.
In his opening keynote on the Huawei Join occasion, going down this week in Shanghai, China, Xu unveiled the corporate’s latest SuperPoD merchandise: the Atlas 950 SuperPoD — with 8,192 Ascend chips, and the Atlas 960 SuperPoD with 15,488 Ascend chips. He defined that SuperPoDs are designed to function as single logical machines, combining a number of bodily items for large-scale AI workloads.
The manager mentioned that the Atlas 950 SuperPoD can be out there within the final quarter of 2026, whereas the Atlas 960 SuperPoD will hit the market within the final quarter of 2027.
Xu additionally confused the significance of computing energy for future AI growth. “Computing energy is, and can proceed to be key to AI. That is very true in China,” the manager mentioned, including that Huawei goals to satisfy long-term demand utilizing semiconductor course of nodes out there on the Chinese language market.
In the course of the occasion, Xu additionally revealed Atlas 950 SuperCluster — with over 500,000 Ascend chips and Atlas 960 SuperCluster — with over a million Ascend chips, constructed from a number of SuperPoDs. The Atlas 950 SuperClusters can be out there within the fourth quarter of 2026, whereas the Atlas 960 SuperCluster can be launched within the final quarter of 2027, Xu mentioned.
“SuperPoDs and SuperClusters powered by UnifiedBus are our reply to surging demand for computing, each in the present day and tomorrow. Our purpose is to maintain pushing developments in AI to create better worth,” the manager added.
With the brand new SuperPoDs and SuperClusters, Xu mentioned Huawei has what it takes to supply plentiful computing energy for speedy developments in AI, including that he believes that the 2 programs will set world benchmarks in key efficiency areas and stay probably the most highly effective tremendous nodes worldwide for years to return. He emphasised that Huawei sees their affect extending past AI, with the potential to ship main advantages for general-purpose computing demand in China.
The Chinese language firm additionally famous {that a} main bottleneck for large-scale AI computing infrastructure is interconnect expertise, specifically the bodily limitations of present cable expertise to hyperlink up huge numbers of chips and SuperPoDs over lengthy distances whereas sustaining a dependable, high-speed, and low-latency connection. Huawei mentioned it has overcome these challenges with UnifiedBus, an interconnect protocol for SuperPoDs.
“SuperPoDs and SuperClusters powered by UnifiedBus are our reply to surging demand for computing, each in the present day and tomorrow,” mentioned Xu.
