Huawei quickens AI chip pace, promises next entrant 3 quarters early
Huawei Technologies said on Thursday that it would launch its next-generation artificial intelligence chip in the first quarter of 2027, moving the target launch forward by nine months as the firm aggressively expands its ecosystem amid China’s self-sufficiency push. David Wang Tao, rotating and acting chairman for Huawei, said on Thursday that the Ascend 960DT chip, an AI chip designed for model…
Huawei Technologies has announced that it will introduce its next-generation artificial intelligence chip, the Ascend 960DT, in the first quarter of 2027, six months earlier than initially planned. This move is part of the firm's broader strategy to bolster its ecosystem amid China's self-sufficiency push. The Ascend 960DT chip is designed for model training and promises to double performance compared to its predecessor.
In the third quarter of 2027, Huawei will launch the Ascend 960PR chip, intended for AI model inference, a quarter ahead of schedule. Huawei's chairman, David Wang Tao, stated that the company will continue its annual upgrade cycle, with the Ascend 970 and 980 chips set to launch in 2028 and 2029, respectively. These announcements come ahead of high-level trade talks between China's President Xi Jinping and US President Donald Trump, where AI and semiconductor exports are expected to be a key focus.
The Ascend 950, Huawei's current most powerful processor, has been used for AI workloads as a substitute for Nvidia's advanced graphics processing units (GPUs), which are restricted in the domestic market due to US export controls. In May, Huawei unveiled its Tau Scaling Law and LogicFolding architecture, semiconductor breakthroughs that aim to achieve chip performance comparable to 1.4-nanometre manufacturing process chips by 2031 without relying on advanced lithography tools unavailable in China.
The company also announced the Atlas 960 SuperPoD computing cluster, which will be deployed by the end of 2027. Powered by Ascend 960 chips, the system is expected to enhance performance in training and inference for 10-trillion-parameter models by 2.3 and 2.5 times, respectively, compared to the Atlas 950 SuperPoD, scheduled for deployment by the fourth quarter of 2026.
AI training involves processing massive data sets to teach models basic patterns, while inference refers to using a finished model to respond to user queries and generate output.
Written by urgent.news from SCMP Tech's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
