Huawei has revealed the Atlas 960E SuperPoD and moved up the release of its Ascend 960DT AI chip to the first quarter of 2027, according to a report published by Channel Insider. The Chinese tech giant is shifting its approach to competing with Nvidia, focusing on building massive interconnected systems rather than winning on individual processors alone. The infrastructure platform is designed to combine computing, networking, memory, storage, and software into a unified stack, though constrained chip production and Nvidia's dominant CUDA software continue to pose significant obstacles to wider market penetration.
The SuperPoD architecture can expand to 4,096 NPUs within a single system and achieve 8 EFLOPS of FP8 compute performance with as much as 1 petabyte of high-bandwidth memory, the report states. Multiple SuperPods can link together into a SuperCluster capable of supporting as many as 512,000 NPUs, while a multi-rail design could eventually stretch that framework to 1 million NPUs. The Ascend 960DT chip now arrives three quarters ahead of schedule, with the Ascend 960PR expected in the third quarter of this year. Huawei plans to roll out a new Ascend generation annually, targeting the Ascend 970 for 2028 and the Ascend 980 for 2029. The company's UnifiedBus interconnect and Hi-ONE near-packaged optics technology enable the system to operate with just 5,500 optical engines instead of the 48,000 traditional 800G optical modules typically required, cutting power use by more than 550 kilowatts and delivering 99.8% system availability.
According to the report, Huawei's CANN software platform—its Compute Architecture for Neural Networks—has transitioned to sustained, community-led open-source development and now counts more than 5,200 monthly active developers. Ascend hardware works with over 90 third-party open-source projects, including PyTorch, Triton, vLLM, and veRL, and more than 40 models have been natively pretrained on the Ascend and CANN stack. Yet Nvidia maintains a substantial software edge through CUDA despite Huawei's push to expand Ascend adoption, Reuters noted in coverage cited by the report. Rotating chairman Eric Xu acknowledged that the company is already struggling to manufacture enough AI computing equipment to satisfy demand within China and therefore has no plans for large-scale overseas expansion at this time.
Huawei's infrastructure-first strategy matters because AI competition is moving toward sprawling interconnected systems that demand far more than accelerator chips alone, the report explains. Deployments at this scale require high-speed optical networking, storage, power management, and cooling, along with sophisticated integration across all those layers. Customers weighing systems of this magnitude are evaluating entire stacks—networking capacity, software compatibility, energy demands, support, and suppliers' ability to deliver hardware at the required volume—not just processors. Channel Insider previously reported that AI adoption is exposing network readiness gaps and creating demand for partners capable of modernizing infrastructure to handle increasingly data-heavy workloads, while power and cooling are emerging as critical components of the equation as AI clusters grow denser. DeepSeek reportedly intends to deploy 160,000 Huawei AI chips at a data center in Inner Mongolia, a project that could further cement Huawei's position in China's AI infrastructure market, the report states.
The Atlas 960E's significance lies as much in its architecture as its raw specifications, particularly if the AI race continues shifting toward massive networked systems, according to the report. Companies providing networking, optics, storage, cooling, and integration for those accelerators may become increasingly central to how the AI infrastructure market evolves. For partners and infrastructure vendors, the trend toward thousands-of-NPU deployments opens a broader opportunity across the full stack that surrounds the processors themselves. The question is whether scale and system integration can offset the software and production challenges that still constrain Huawei's ambitions beyond China. Huawei's bet on infrastructure over individual chip performance reflects a recognition that the battleground has shifted, but whether customers outside China will commit to an entire ecosystem remains the decisive test.