• DeepSeek’s V4 AI model may run on Huawei chips.
• The move reflects a shift to domestic AI chips and software in China.

China’s push to reduce reliance on US technology is showing up in AI infrastructure. A recent report by The Information, cited by Reuters, said that DeepSeek’s upcoming V4 model may run on Huawei chips rather than the NVIDIA hardware that still powers most large AI systems today. The shift points to a broader change in AI infrastructure.

According to the report, DeepSeek has been adapting parts of its model to work with Huawei’s Ascend chips. At the same time, large Chinese firms such as Alibaba, ByteDance, and Tencent have placed orders for hundreds of thousands of Huawei chips ahead of the V4 release, according to Reuters. The scale of those orders suggests a wider shift in infrastructure planning.

A move away from NVIDIA dominance

For years, companies building large AI models have depended on GPUs from NVIDIA. Its CUDA software stack and hardware, like the A100 and H100 chips, have set the standard for training models.  

DeepSeek appears to be taking a different route. Reuters reported that DeepSeek withheld early access to V4 from US chipmakers such as AMD and NVIDIA, while giving Chinese suppliers such as Huawei more time to tune their software. Models built for one chip type do not always run well on another. DeepSeek is said to be rewriting parts of V4’s code so the model can run on Huawei’s chips.

How V4 may compare to current frontier models

DeepSeek has already drawn attention with earlier models such as V3 and R1. Reuters has reported that DeepSeek claimed it built earlier models at a lower cost than many US rivals. Details on V4 remain limited, but it is being built on domestic chips.

The key difference is how those results are achieved. US models often rely on large clusters of high-end GPUs, which are expensive and energy-intensive. DeepSeek’s approach focuses more on efficiency per unit of compute.

This does not mean Huawei chips match NVIDIA’s best hardware. Reuters reported in 2025 that Huawei and other Chinese chipmakers had struggled for years to match NVIDIA’s top-end chips for training models. Performance in real-world AI systems depends on hardware, software, and data. By adjusting the model to the hardware, DeepSeek may be narrowing that gap.

Why Huawei chips are central to this shift

Huawei has been developing Ascend AI chips. These chips are already used in some data centres and AI workloads within China. Export controls from the United States have limited access to advanced NVIDIA chips, which has increased demand for local options.

The reported surge in orders from companies like Alibaba, ByteDance, and Tencent shows that demand is not limited to one or two projects. It reflects a move to secure supply chains and reduce reliance on foreign hardware.

Huawei’s role goes beyond hardware. Reuters reported that DeepSeek worked with Huawei and Cambricon as it rewrote parts of V4’s code.

A parallel AI stack is taking shape

These changes suggest that two AI ecosystems may be forming. One is centred on US technology, with NVIDIA hardware and software at its core. The other is forming around Chinese companies, with Huawei chips and local software stacks.

DeepSeek’s V4 model may become a key test of how viable that second system is. If the model performs well, it could encourage more companies to follow the same path.

At the same time, this shift could affect how AI costs are structured. Reuters has linked DeepSeek’s earlier low-cost model claims to investor concerns about the high spending levels of some US AI firms.

What this means for the broader AI race

The race to build better AI models is often framed as a contest of model size or benchmark scores. Control over chips, software, and supply chains can shape what is possible and at what cost.

DeepSeek’s move to Huawei chips does not settle that race, but it adds a new dimension. It shows that high-level AI work is no longer tied to a single hardware path. Different regions may build their own stacks, each with its own trade-offs.

If V4 delivers strong results, it could mark a turning point. It would show that competitive AI systems can be built outside the NVIDIA ecosystem.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.