DeepSeek V5 is reportedly set to train fully on Huawei Ascend

Priya reports that DeepSeek is preparing an imminent V5 launch and that founder Liang Wenfeng calls it the company’s biggest bet yet. The model is rumored to have 2 trillion parameters, not 3 trillion.

Priya says V5 would be the first DeepSeek model trained fully on Huawei Ascend chips rather than Nvidia, and that reaching the same training scale would require roughly four times as many Ascend accelerators as the Nvidia equivalent. DeepSeek is reportedly keeping its open-weight strategy. Priya also headlines the leak as beating Astra.