Chinese tech giant Huawei plans to release two new artificial intelligence (AI) semiconductor products next year. As U.S. export controls make it difficult for the company to secure advanced AI chips from NVIDIA, Huawei is accelerating the development of its own chip and system technologies.
According to reports from the China Securities Journal, Huawei's rotating chairman, Wang Tao, announced at the Huawei Connect conference in Shanghai on September 17 that the Ascend 960 series will include the 960DT chip, set for release in the first quarter of 2027, and the 960PR chip, expected in the third quarter.
Wang stated that the Ascend chips will evolve annually, with plans to launch the Ascend 970 and 980 chips in 2028 and 2029, respectively. He also emphasized ongoing improvements in processing power, memory capacity, and data transfer speeds between chips.
According to the South China Morning Post, Huawei is also speeding up the development of its interconnect technology, UnifiedBus, which connects multiple AI chips into a single system. This technology allows for rapid data exchange among AI chips, enabling them to function as a massive computing system, comparable to NVIDIA's NVLink technology.
Wang revealed that 11 semiconductor products utilizing UnifiedBus technology have been developed, allowing Huawei's so-called 'supercluster' AI systems to connect up to one million AI processors.
Huawei has already shipped over 1,000 'supernode' systems, which combine multiple AI semiconductors for processing tasks, to more than 370 customers. Supernodes are smaller AI computing systems compared to superclusters.
This announcement comes amid heightened competition between the U.S. and China over semiconductor, data center, and software technologies, which are crucial for AI. As U.S. export controls hinder Chinese companies from adequately securing NVIDIA's latest AI chips, the importance of technologies that connect domestic chips into a single system is increasing.
Guo Ping, chairman of Huawei's supervisory board, also stated at a recent meeting that the company's goal in the ICT and computing sectors is to become like NVIDIA. He acknowledged that while individual chips may lag behind NVIDIA, the ability to integrate chips into a cohesive system will help overcome this disadvantage.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.