Huawei has pulled forward the release of its next-generation Ascend 960DT AI processor to the first quarter of 2027, moving up its internal roadmap by about six months. Announced at Huawei Connect by rotating chairman David Wang, the silicon family is designed to double performance year over year as domestic infrastructure providers race to rival Nvidia.
To overcome manufacturing constraints, Huawei is relying on its Peerium Computing Architecture and proprietary UnifiedBus interconnect, designed to link hundreds of thousands—and eventually millions—of dies directly to memory, storage and networking into a single unified computing fabric. Separately, Huawei said Ascend has crossed its ecosystem inflection point: the CANN open-source community now averages over 5,200 monthly active users, non-Huawei developers outnumber Huawei developers, and over 40 large and multimodal models pre-trained on Ascend are available.
The acceleration matters because China's hardware ecosystem is racing to compensate for US export controls by advancing silicon timelines and leaning on supercomputing interconnect fabrics to meet massive domestic compute demand.
Source: AI Breaking Wire / 21st Century Business Herald
