Huawei has moved the planned release of its next-generation Ascend 960DT artificial intelligence chip to the first quarter of 2027, accelerating its previously announced schedule by several months.
Unveiled during the company's Huawei Connect event, the updated roadmap signals Huawei's ambition to expand its presence in high-performance AI computing. The company says the Ascend 960 series will deliver a major performance increase over earlier generations while supporting yearly advances across its chip portfolio.
Building AI at larger scale
The new chip is designed to support Huawei's broader vision for interconnected AI infrastructure. Through its Peerium Computing Architecture, the company aims to combine large numbers of AI accelerators into unified computing environments for model training and inference.
This architecture is supported by UnifiedBus, Huawei's technology for connecting processors, memory, storage and networking components. The goal is to help organizations build AI systems that can scale beyond individual servers and operate as coordinated computing clusters.
Huawei's Atlas 950 SuperPoD and SuperCluster platforms are among the first systems built around this approach. According to the company, an Atlas 950 SuperCluster can link up to 256,000 accelerator cards, illustrating the scale targeted for future AI workloads.
Huawei is positioning the Ascend 960DT as part of an expanding domestic semiconductor ecosystem and a growing global market for AI infrastructure. The earlier launch timeline could give developers, cloud platforms and research organizations faster access to new computing capacity.
The accelerated roadmap reflects how rapidly AI hardware is evolving. As more companies develop scalable chips and interconnected computing systems, the next generation of AI could become more powerful, distributed and accessible across industries.
