In a significant departure from its typically discrete approach, Huawei has publicly detailed its upcoming plans for chips and computing power, signaling a bold challenge to global rivals like Nvidia. At the Huawei Connect conference in Shanghai, Vice Chairman Eric Xu outlined a comprehensive roadmap aimed at bolstering China’s technological self-reliance in artificial intelligence hardware.
Ascend AI Chips: A New Generation of Huawei Chips
Huawei revealed its strategy to roll out four new versions of its Ascend AI chips over the next few years. Following the earlier Ascend 910C, launched in early 2025, the roadmap includes:
- Two variants of the Ascend 950 in 2026
- The Ascend 960 in 2027.
- The Ascend 970 in 2028.
A critical component of this announcement is Huawei’s new, proprietary high-bandwidth memory (HBM) technology.
Vice Chairman Xu emphasized that this innovation represents a vital step forward, addressing previous dependencies on foreign suppliers for this crucial component. This development aligns with broader national efforts to build domestic alternatives to U.S.-controlled technologies, particularly given escalating trade tensions and export restrictions.
Atlas Systems Redefine Scale
Alongside the chip roadmap, Huawei unveiled plans for advanced computing platforms known as supernodes:
- Atlas 950: Scheduled for Q4 2026, this system is designed to support an impressive 8,192 Ascend chips.
- Atlas 960: Slated for Q4 2027, this more advanced version will support an even larger configuration of 15,488 Ascend chips.
These new Atlas systems build upon the foundation of Huawei’s existing Atlas 900 (also known as CloudMatrix 384) but promise substantial improvements in scale and performance. Huawei claims that its upcoming supernodes will outperform Nvidia’s GB200 NVL72 system in key performance metrics, specifically citing advantages in interconnect speeds, memory capacity, and overall compute power.
New Huawei Chips: Implications and Challenges
Huawei’s announcements mark a strategic shift from secrecy to open competition. For years, China has faced considerable obstacles in developing advanced semiconductor infrastructure, largely due to international trade restrictions and reliance on foreign suppliers.
These new plans indicate Huawei’s determination to overcome those limitations and compete at the forefront of AI hardware. The timing of this revelation is particularly significant, occurring amidst increasing global trade tensions and a concerted push for technological sovereignty. By revealing this ambitious roadmap, Huawei is signaling its readiness to contend with major players in the AI hardware space.
However, significant challenges remain. Scaling manufacturing for advanced AI chips[1] requires flawless yields, mature fabrication processes, and stable supply chains for memory and core silicon components. The volatile international trade environment could still complicate access to critical components. Furthermore, proving the reliability and performance of these new Atlas systems in real-world workloads will be crucial for their adoption and long-term success.
References
- ^ advanced AI chips (www.techjuice.pk)