- DeepSeek’s upcoming V4 model will run entirely on Huawei’s Ascend 950PR chips, according to a report by The Information published April 3, 2026.
- DeepSeek spent months working with Huawei and chip designer Cambricon to port the model to Chinese-made hardware, with only Chinese chip companies receiving early access to V4 testing.
- V4 uses a mixture-of-experts architecture with approximately one trillion total parameters, with 37 billion activating per inference pass, and accepts text, images, and code.
- Alibaba, ByteDance, and Tencent have ordered hundreds of thousands of Huawei Ascend 950PR units, pushing chip prices up 20%.
What Happened
DeepSeek’s next-generation AI model, V4, will be able to run entirely on chips designed by Huawei Technologies, according to a report by The Information published on April 3, 2026. DeepSeek spent months collaborating with Huawei and Chinese chip designer Cambricon to port the model to domestically produced hardware. The report states that only Chinese chip companies received early access to V4 testing, with Nvidia explicitly excluded from the process.
V4 is expected to launch in the coming weeks and represents a significant technical milestone: a frontier-scale AI model designed from the ground up to operate on non-Nvidia hardware. As The Decoder reported, the development marks a major step in China’s push to achieve AI independence from Western semiconductor supply chains, an effort that has accelerated since the U.S. imposed export controls on advanced AI chips in October 2022.
Why It Matters
U.S. export controls have restricted Chinese companies’ access to Nvidia’s most advanced AI accelerators since October 2022, with progressively tighter restrictions applied in subsequent rounds through 2023 and 2024. DeepSeek’s ability to train and deploy a frontier model on Huawei’s Ascend chips demonstrates that Chinese AI labs are making measurable progress in working around these restrictions. DeepSeek’s earlier V3 model, released in late 2025, already demonstrated competitive performance with Western frontier models while using fewer compute resources, a result that surprised many industry observers.
The development is a direct win for Huawei, which has positioned its Ascend chip line as the domestic alternative to Nvidia’s data center GPUs. If V4 achieves competitive performance on Huawei hardware, it validates Huawei’s semiconductor strategy and could accelerate adoption of Chinese-made AI accelerators across the country’s technology sector. The strategic implications extend beyond AI into the broader U.S.-China technology competition.
Technical Details
V4 is built on a mixture-of-experts architecture with approximately one trillion total parameters. During inference, roughly 37 billion parameters activate per forward pass, a design that reduces computational cost while maintaining model capacity. The model is multimodal, accepting text, images, and code within the same context window. Huawei’s Ascend 950PR chip delivers approximately 2.8 times the computing power of Nvidia’s H20, the most advanced chip currently available to Chinese buyers under U.S. export controls. However, the Ascend 950PR still underperforms Nvidia’s H200, which is restricted from export to China.
Chinese technology companies have responded to V4 with large hardware orders. According to Reuters, Alibaba, ByteDance, and Tencent have ordered hundreds of thousands of Ascend 950PR units to run DeepSeek V4 through their cloud services and integrate it into their own AI applications. The surge in demand has pushed Ascend 950PR prices up by 20%, reflecting both the model’s anticipated capability and the constrained supply of domestically produced AI chips.
Who’s Affected
Nvidia faces the prospect of losing ground in the Chinese market as domestic alternatives prove capable of running frontier AI models. Chinese cloud providers, including Alibaba Cloud, ByteDance’s Volcano Engine, and Tencent Cloud, benefit from access to a competitive domestic AI model that does not depend on restricted foreign hardware. U.S. policymakers who designed export controls to slow China’s AI progress must assess whether those restrictions are achieving their intended effect or are instead accelerating Chinese investment in domestic semiconductor alternatives.
What’s Next
DeepSeek V4’s launch is expected within weeks. Performance benchmarks comparing V4 on Huawei hardware against leading Western models on Nvidia GPUs will determine whether the model achieves parity with frontier competitors. Huawei faces production bottlenecks due to U.S. export controls on semiconductor manufacturing equipment, which could constrain the scale of Ascend 950PR deployment despite the surge in orders. The outcome will influence both future U.S. export control policy and the pace at which Chinese technology companies shift their AI infrastructure to domestic hardware.
