Key points
- Huawei has announced four new iterations of its Ascend AI chip over the next three years.
- The new chips aim to compete directly with Nvidia’s offerings in the AI market.
- The launch includes the Ascend 950 series and powerful Atlas supernodes.
- Huawei claims to have overcome key technological bottlenecks in high-bandwidth memory.
Huawei, the Chinese tech giant, has boldly entered the high-stakes AI chip market, announcing plans to release four new iterations of its Ascend AI chip series over the next three years. This announcement, made at the Huawei Connect conference in Shanghai, marks a significant departure from the company’s previous secrecy surrounding its chip development and represents a direct challenge to industry leader Nvidia.
Vice Chairman Eric Xu revealed the roadmap, highlighting the upcoming Ascend 950 series (with two variants launching next year), followed by the Ascend 960 in 2027 and the Ascend 970 in 2028.
The strategic move underscores China’s ambition to strengthen its domestic semiconductor industry and reduce its dependence on foreign suppliers, particularly those from the United States. Xu emphasized the crucial role of computing power in AI advancement, particularly for China’s AI development.
A key innovation unveiled is the Ascend 950’s use of Huawei’s proprietary high-bandwidth memory, a technology previously reliant on suppliers from South Korea and the US. This signifies a breakthrough in overcoming a significant technological hurdle for China.
Further solidifying its position in the high-performance computing arena, Huawei also unveiled the Atlas 950 and Atlas 960 supernodes. These supernodes, described as the world’s most powerful, will support 8,192 and 15,488 Ascend chips, respectively, significantly exceeding the capabilities of its predecessor, the Atlas 900 (CloudMatrix 384). Independent research from SemiAnalysis suggests that, on certain metrics, Huawei’s system outperforms Nvidia’s GB200 NVL72.
Huawei’s aggressive expansion into the AI chip market, coupled with its claims of superior performance and technological advancements, indicates a strong commitment to compete globally.
This move is expected to intensify the competition in the burgeoning AI sector and further accelerate China’s efforts to establish self-reliance in critical technological domains. The innovative “supernode” architecture, enabling ultra-high-speed interconnectivity between chips, is a core element of this ambitious strategy.
