DeepSeek V4 to Launch in Coming Weeks, Fully Operated on Huawei Chips

On April 4, reports from the U.S. tech media outlet The Information revealed that China’s AI startup DeepSeek is set to release its next-generation flagship model V4, which will be fully operated on Huawei’s self-developed chips. This is seen as an important milestone in China’s push for semiconductor autonomy. V4 is expected to launch in the coming weeks and has been developed in collaboration with Huawei and chip designer Cambricon over several months, during which the underlying code of the model was rewritten to ensure compatibility with domestic chips. Notably, DeepSeek has not granted Nvidia early testing access for V4, instead inviting domestic chip companies to participate in early optimization, breaking industry norms. Chinese tech giants such as Alibaba, ByteDance, and Tencent have already placed substantial orders for Huawei’s latest Ascend 950PR chips, totaling hundreds of thousands of units, intending to deploy V4 through cloud services and integrate it into their own AI applications. The surge in demand has driven the chip’s price up by approximately 20%. V4 employs a mixture of experts (MoE) architecture, with a total parameter count of around 1 trillion, activating about 37 billion parameters per inference, supporting multimodal inputs of text, images, and code while maintaining low latency. Last year, DeepSeek’s release of the low-cost models V3 and R1 led to a significant drop in global tech stocks, raising questions about whether U.S. AI companies need to spend billions on computing power. Consequently, V4 has garnered significant international attention.