Huawei has confirmed plans to launch two new artificial intelligence chips in 2027 as the Chinese technology company continues expanding its AI computing business. Rotating chairman David Wang announced the roadmap at Huawei Connect 2026 in Shanghai on 17 September, outlining plans for the 960DT and Ascend 960PR.
The 960DT is scheduled to arrive in the first quarter of 2027, while the Ascend 960PR is expected in the third quarter. Huawei’s latest roadmap brings the 960DT forward significantly from its earlier schedule. It is part of a broader effort to build large-scale AI computing systems around the company’s Ascend chip family.
Huawei brings forward its next-generation AI chips
The new chips are intended to strengthen Huawei’s position in AI infrastructure at a time when access to advanced processors has become an important issue for Chinese technology companies. US export controls have restricted China’s access to certain advanced computing chips and semiconductor manufacturing equipment, making locally developed alternatives more important.
Huawei said development of the Ascend 960 has progressed faster than expected. According to Wang, the Ascend 960DT will be available in the first quarter of 2027, three quarters earlier than the company’s previous roadmap, while the 960PR will arrive in the third quarter, one quarter earlier than previously planned. Huawei describes the 960DT as part of its AI training hardware and the 960PR as an inference-focused chip.
The company also intends to maintain an annual update cycle for its Ascend processors. Huawei said the Ascend 970 and Ascend 980 are planned for 2028 and 2029 respectively. Wang said the company expects improvements not only in computing performance but also in memory bandwidth, memory capacity and interconnect bandwidth as the series develops.
The accelerated timetable highlights how Huawei is developing its AI hardware roadmap while also expanding the infrastructure surrounding its processors. Rather than relying solely on improvements to individual chips, the company is placing considerable emphasis on connecting large numbers of processors into unified computing systems.
UnifiedBus is central to Huawei’s large AI systems
Huawei’s approach relies heavily on UnifiedBus, a technology designed to allow large numbers of AI processors to communicate and operate together. This matters because advanced AI models can require far more computing resources than a single processor can provide, making communication between chips a key part of large AI systems.
Wang said Huawei has developed 11 semiconductors around UnifiedBus for use in its larger computing systems. The company says its largest systems, known as superclusters, can connect as many as one million AI processors. Huawei has also shipped more than 1,000 smaller systems, known as supernodes, to more than 370 customers. However, it has not disclosed those customers’ identities or the number of chips in each system.
Huawei is also developing optical interconnect technology to help scale these systems. The company said its Hi-ONE optical interconnect engine can provide 7.2 Tbit/s of transmission capacity per engine and is designed to support the expansion of large SuperPoD systems. Huawei has used the technology with its Ascend 960 processors to develop the Atlas 960E SuperPoD.
According to Huawei, an Atlas 960E SuperPoD can scale to 4,096 NPUs and deliver up to 8 EFLOPS of FP8 computing performance, along with up to 1 petabyte of HBM capacity. The company says its use of near-packaged optics can reduce power consumption and improve system availability compared with conventional optical connections. These figures are Huawei’s own specifications and have not been independently verified.
Huawei faces a wider Nvidia challenge
Huawei’s chip plans come as the company seeks to expand the role of its hardware in China’s AI ecosystem. Nvidia remains a major provider of AI processors and software globally, with its CUDA platform forming a significant part of the software environment AI developers use. Huawei therefore faces challenges beyond producing processors with competitive computing performance.
Building a strong developer ecosystem is especially important because AI software must be adapted and optimised for specific hardware platforms. Huawei has been expanding its own software and development environment. At the same time, the company says Ascend is now supported as a PyTorch accelerator backend and has expanded into more than 90 leading third-party open-source projects.
Restrictions on advanced chip exports have also contributed to a stronger focus on domestic AI infrastructure in China. Huawei has become one of the companies developing alternatives as Chinese organisations seek ways to expand AI computing capacity without relying as heavily on restricted foreign technology. Reuters reported that demand for Huawei’s AI computing equipment has been strong, although production capacity remains a constraint on the company’s ability to expand internationally.
Huawei’s 2027 roadmap therefore extends beyond two new processors. The company is developing chips, interconnect technology, SuperPoDs and larger superclusters as parts of a broader AI computing platform. Its ability to turn those technologies into widely adopted systems will depend on manufacturing capacity, software support, developer adoption and access to the components required to build increasingly large AI infrastructure.
