BEIJING -- DeepSeek's next flagship model, V4, will run entirely on Huawei's Ascend 950PR chips — and Alibaba, ByteDance, and Tencent have already placed bulk orders for the hardware in anticipation of the launch, which is expected within weeks. [1]

The significance is structural. Every previous frontier Chinese AI model either trained on Nvidia hardware stockpiled before US export controls tightened or depended on workarounds. V4, if it performs as reported, will be the first model from a leading Chinese AI lab that operates entirely on domestic silicon from training through inference. [2]

The Ascend 950PR is Huawei's newest chip in the Ascend lineup. Mass production is scheduled to begin this month. Demand from DeepSeek's V4 deployment has already pushed prices up approximately 20 percent in recent weeks, according to analysts cited by Tech in Asia. Huawei is projected to take roughly 50 percent of China's AI chip market share by end of 2026. [3]

DeepSeek deliberately withheld early V4 access from Nvidia — granting Huawei exclusive first access instead — a decision reported in February by The China Academy. The sequencing is a message: China's most capable AI lab is now running its stack on Chinese hardware, by choice, not compulsion. [4]

The US export control regime that restricted Nvidia's H100 and A100 chips from China was designed to slow China's AI development. DeepSeek's R1, released in January 2025, disrupted that assumption by demonstrating that efficient training methods could close the capability gap. V4 running on Ascend 950PR closes the hardware dependency entirely. [5]