Huawei Unveils Ascend 960 AI Chips, Plans Annual Release Cycle
SingTao · 2 SOURCESabout 3 hours ago4 MIN
Summary
Huawei unveiled its next-generation Ascend 960 AI chips at the Huawei Connect 2026 conference in Shanghai on Thursday. The chips, comprising the training-oriented DT model and the inference-optimized PR variant, are both launching ahead of schedule. Company Rotating Chairman Wang Tao (汪濤) announced the accelerated timeline while committing to an annual release cadence extending through the end of the decade.
Key Points
- Ascend 960DT, designed for AI model training, will launch in Q1 2027—three quarters earlier than originally planned
- Ascend 960PR, built for inference workloads, is scheduled for Q3 2027—already one quarter ahead of the initial roadmap
- Both chips surpass the Nvidia H200 SXM in performance benchmarks but fall short of the B300's capabilities
- Huawei projects 2028 and 2029 releases for Ascend 970 and Ascend 980, with computing power expected to double with each generation
- The Hi-ONE near-package optics product—the industry's first mass-produced NPO offering—features 7.2 terabits per second total transmission capacity
Why It Matters
The accelerated timeline underscores Huawei's determination to narrow the gap with Nvidia in the AI chip market despite ongoing US export restrictions on advanced semiconductors . If Chinese open-source AI developers increasingly adopt the Ascend 950 and 960 platforms for native pre-training, this could reduce the sector's reliance on foreign-made processors and reshape the competitive dynamics of global AI infrastructure .
The accelerated timeline underscores Huawei's determination to narrow the gap with Nvidia in the AI chip market despite ongoing US export restrictions on advanced semiconductors . If Chinese open-source AI developers increasingly adopt the Ascend 950 and 960 platforms for native pre-training, this could reduce the sector's reliance on foreign-made processors and reshape the competitive dynamics of global AI infrastructure .