Huawei Accelerates Ascend 960 AI Chip Release to Q1 2027 with 2x Performance Boost
Huawei Rotating Chairman Wang Tao announced that the launch of the Ascend 960 AI chip has been advanced to Q1 2027 with a claimed 2x performance increase over its predecessor. The company plans to release the 960DT variant in Q1 2027 and the 960PR variant in Q3 2027, followed by yearly upgrades including the Ascend 970 in 2028 and Ascend 980 in 2029. Accelerating its hardware roadmap highlights Huawei's push to narrow the technological gap with leading global semiconductor vendors and strengthen China's domestic AI infrastructure. Maintaining a consistent annual release cadence demonstrates rapid architectural progress despite foreign trade and technology restrictions. The Ascend 960DT model is scheduled to arrive three quarters ahead of the original Q4 2027 baseline, while the 960PR model is accelerated by one quarter. Huawei continues its specialized architecture approach from the Ascend 950 generation, leveraging self-developed HBM (HiBL) technology and splitting chips into PR and DT variants tailored for specific AI inference and training workloads.
## BACKGROUND
The Ascend series is Huawei's lineup of dedicated AI accelerators designed for training and running machine learning models. Large language model (LLM) inference consists of two distinct computational stages: the prefill phase, which processes the initial input prompt in parallel, and the decode phase, which generates output tokens one by one. Huawei optimizes its PR and DT chip variants around these distinct phases to maximize memory bandwidth and throughput.