Huawei has unveiled its latest AI hardware innovations in Shanghai, showcasing rapid progression in its chip technology and reinforcing China’s push for technological independence amidst US restrictions.
Huawei has unveiled a new generation of artificial intelligence hardware in Shanghai, stepping up its challenge to Nvidia and other Western chipmakers as China pushes harder for technological self-reliance. The announcement came at Huawei Connect 2026 and follows years of U.S.-led export restrictions that have limited Chinese access to some of the most advanced AI chips and manufacturing tools.
The company said its new Atlas 960E SuperPoD is the industry’s first near-packaged optics-based SuperPoD, a design intended to speed up both training and inference for AI systems with as many as 10 trillion parameters. Huawei also introduced UnifiedBus, an interconnect architecture designed to reduce data bottlenecks in large AI clusters. The firm said the system combines ten protocols into one, lifts bandwidth to as much as 1 Tbit/s and cuts round-trip latency, while also supporting direct communication between CPUs, NPUs and other components.
Huawei’s latest claims build on the Atlas 950 SuperPoD and related supercluster products it unveiled last year. According to Huawei’s own 2025 launch material, the Atlas 950 SuperPoD used 8,192 Ascend NPUs, while the Atlas 960 SuperPoD was described as using 15,488 Ascend NPUs. The company also said at the time that its Atlas 950 SuperCluster would exceed 500,000 Ascend NPUs, with the Atlas 960 SuperCluster crossing one million, underlining the scale of its effort to create a domestic AI stack.
The new announcements also point to how quickly Huawei is iterating. Reuters reported that the Atlas 960 SuperPoD arrived only months after the Atlas 950 SuperPoD, prompting fresh concern among analysts over the pace of China’s AI build-out. George Chen, a partner and chair of digital practice at The Asia Group, told Reuters that the timing underscored Beijing’s confidence in its technology ambitions. He also said the speed of Huawei’s progress would intensify U.S. unease about the AI race.
Huawei said separately that it plans to launch the Ascend 970 and 980 chip series in 2028 and 2029. The company framed those products as part of a longer-term push to improve computing power while building around components and manufacturing processes that remain available to the Chinese mainland. That strategy reflects a broader Chinese effort to reduce dependence on U.S. technology, even as advanced model training in China still often relies on Nvidia chips, according to industry analysts cited by Reuters.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
