Huawei plans to make its next-generation Ascend 960DT AI chip available in the first quarter of 2027, according to its September 17 Huawei Connect keynote. The company also moved the Ascend 960PR to the third quarter of 2027, advancing its roadmap for AI computing hardware.

TechCrunch separately reported the Q1 target, citing a Huawei spokesperson, while TrendForce reported the same timing from Chinese coverage of the event. The announcements describe planned availability, not chips already shipping.

A broader systems roadmap

Huawei paired the chip update with the Atlas 960E SuperPoD, a system built around Ascend 960 processors and its Hi-ONE optical interconnect technology. The company says a single system can scale to 4,096 neural processing units, with up to one petabyte of high-bandwidth memory.

Those figures are Huawei's announced specifications, rather than independently verified deployment results. Its keynote describes near-packaged optics and UnifiedBus interconnects as central to coordinating processors across larger computing systems.

Rotating chairman David Wang also outlined an annual chip-generation cycle, with Ascend 970 and 980 planned for 2028 and 2029, respectively.

What remains to be demonstrated

The nearer launch target gives AI infrastructure buyers a more specific date to evaluate Huawei's next platform. It does not establish delivered performance, production volume, or superiority over competing accelerators.

The announcement is therefore a roadmap update: Huawei is bringing forward its planned hardware availability while expanding the surrounding interconnect and system architecture. Practical competitiveness will depend on shipping systems and measured workload performance, not the schedule alone.