DeepSeek V4 Will Run Entirely on Huawei Chips — A Major Win for China's AI Independence

DeepSeek's upcoming V4 model will reportedly be trained and served entirely on Huawei Ascend chips, according to The Decoder — a development that, if accurate, marks a pivotal moment in China's push to build a domestically self-sufficient AI stack independent of US export-controlled hardware.

Why This Matters More Than a Hardware Story

The US government's export controls on advanced semiconductors — specifically NVIDIA's H100, H200, and A100 GPUs — were designed to impede China's ability to train large-scale AI models. DeepSeek's reported use of Huawei Ascend chips for V4 is a direct test of whether those controls achieved their intended effect.

If DeepSeek V4 achieves competitive performance benchmarks while running entirely on Huawei hardware, the geopolitical calculus around semiconductor export controls shifts significantly. It would demonstrate that domestic Chinese chip capabilities have reached a threshold where US-imposed hardware restrictions no longer function as a meaningful bottleneck on frontier AI development.

The Huawei Ascend Ecosystem

Huawei's Ascend 910B — the current generation — has been positioned as a domestic alternative to NVIDIA's data center GPUs. Chinese AI labs, including Baidu and several state-affiliated research institutions, have been building toolchains and frameworks optimized for the Ascend architecture. DeepSeek running V4 entirely on this hardware would represent a significant validation of that ecosystem's production readiness for frontier-scale workloads.

DeepSeek's Track Record Makes This Credible

DeepSeek is not a lab that makes claims it cannot back up. The company's prior releases — including DeepSeek-R1 — demonstrated an ability to train highly competitive reasoning models at dramatically lower cost than US counterparts. Their documented efficiency innovations in training methodology make it more credible that they could coax frontier-level results from non-NVIDIA hardware than almost any other lab attempting the same.

DeepSeek has not officially confirmed the hardware configuration for V4. The Decoder's reporting cites internal sources. A V4 announcement is expected in the coming weeks.