DeepSeek V5 Expected to Complete Training on Ascend 950 Next Year

Odaily reports: Recently, sources familiar with the industry chain revealed that DeepSeek's next-generation large model, DeepSeek V5, plans to adopt Huawei's Ascend 950 chips as its core training compute, with the project timeline pointing to a launch next year. At present, DeepSeek's core R&D liaison team with Ascend is permanently based in Hangzhou, and the two sides have already completed extensive low-level operator adaptation and custom development during the V4 phase.

According to the cooperation path disclosed by the sources, in DeepSeek's previous generation version, Ascend compute only handled inference tasks and had not yet entered the training stage; V5 will be its first large model to natively rely on Ascend 950 for training. (Caiwen)