DeepSeek V4 Will Run on Huawei Chips, Ditching NVIDIA

DeepSeek's next flagship model won't run on NVIDIA hardware. According to a Reuters report citing The Information, the Chinese AI lab's upcoming V4 model is being built to run on the latest chips designed by Huawei Technologies — a significant shift for a company whose previous models were developed primarily on NVIDIA GPUs.

The move isn't happening in isolation. Chinese tech giants including Alibaba, ByteDance, and Tencent have placed bulk orders for Huawei's upcoming chip totaling hundreds of thousands of units, according to five people with direct knowledge of the purchases. The orders are being placed in preparation for V4's launch, which is expected within the next few weeks.

DeepSeek has been working directly with Huawei and another Chinese chip designer, Cambricon Technologies, to rewrite pieces of V4's underlying code and optimize performance for domestic silicon. The lab is also developing two additional V4 variants, each optimized for different capabilities and built to run on Chinese chips.

This aligns with earlier Reuters reporting that DeepSeek withheld its latest model from U.S. chipmakers including NVIDIA, breaking from standard industry practice. Instead, the lab granted early access to domestic suppliers.

The geopolitical context is hard to miss. U.S. export restrictions have pushed Chinese AI labs to find alternatives to NVIDIA's H100 and H200 chips. DeepSeek's V3 and R1 models sent shockwaves through global tech markets when they demonstrated competitive performance at dramatically lower cost. If V4 achieves similar results on fully domestic hardware, it would validate China's push for semiconductor self-sufficiency — and raise serious questions about the effectiveness of U.S. chip export controls.

Neither Huawei nor DeepSeek responded to Reuters' requests for comment.