❯ DeepSeek V5 plans Ascend pretraining as its Huawei collaboration extends deeper into infrastructure

A pretraining plan from supply-chain sourcesSTAR Market Daily reported on November 3 that multiple supply-chain sources say DeepSeek plans to pretrain V5 on Huawei Ascend chips; The Paper specifically identifies Ascend 950. Ascend previously mainly ran trained DeepSeek models to answer requests. Pretraining develops foundational capabilities from large datasets. DeepSeek has not publicly confirmed V5’s training arrangements or release schedule.

Cooperation reaches the software stackThe collaboration already covers the V4 family. The report describes DeepSeek’s September 30 release of Ascend-oriented infrastructure, including TileLang compilation tools, high-performance computing libraries and distributed communication libraries, alongside Huawei’s supernode networking and low-level programming support. These components map model operations onto chips and coordinate data exchange between them.

Training requires system-wide coordinationLarge pretraining runs keep many chips working together for extended periods, increasing requirements for operation support, communication efficiency and stability. Previously disclosed full-parameter V4-Pro post-training on Ascend further optimized an existing foundation model. The V5 plan concerns foundational pretraining, extending cooperation into the model’s initial development if implemented.

▮ SIGNALAscend adaptation now reaches compilation, computation and communication; training capability depends on sustained coordination across hardware and cluster software.