Huawei plans to release a new generation of Ascend AI chips every year, doubling their performance each time. This was stated by Chen Lin, the company's vice president and head of its cloud business in China.
According to him, the market is facing a shortage of computing power and rising memory costs, which negatively impacts model training and iterations, increasing expenses. Chen Lin believes that the solution lies in stable supplies of computing resources and a predictable upgrade schedule.
The executive confirmed the plan of "one generation per year, doubling computing power." The Ascend 950DT chip is set to officially launch in Huawei Cloud in August. The company promises a significant increase in vector performance, memory bandwidth, and interconnects, as well as native support for modern formats like FP8.
Huawei claims this will simplify programming and model tuning, including scenarios for intelligent driving.
Separately, Huawei Cloud announced the creation of a global computing and storage network based on its own Ascend chips and a localized supply chain. Plans include 34 regions and 102 availability zones; large computing clusters have already been deployed in Guizhou, Wuhu, and Inner Mongolia.
The company also outlined goals in the automotive and enterprise sectors. Manufacturers of "smart" vehicles will receive a safer and more reliable cloud infrastructure, along with sustainable computing power development. Huawei will continue to increase investments in AI services for digital business transformation.
It is worth noting that in January, Google DeepMind CEO Demis Hassabis suggested that China has nearly caught up with the U.S. and the West in the field of AI.