Chinese AI startup DeepSeek has launched its latest V4 model with full support for Huawei’s Ascend AI chips, marking a significant shift toward domestic hardware as China accelerates efforts to build an independent artificial intelligence ecosystem. The move reflects a broader strategy to reduce reliance on foreign technologies amid ongoing geopolitical and supply chain constraints.
The DeepSeek V4 model has been specifically adapted to run on Huawei’s Ascend chipsets, including next-generation clusters based on Ascend 950 systems. This makes it the company’s first major model designed to operate on non-Nvidia hardware at scale, highlighting a transition driven partly by US export restrictions on advanced AI chips.
Technically, the V4 model introduces improvements in efficiency and scalability, including a large context window of up to one million tokens and architectural optimizations that reduce computing and memory costs. The model is available in multiple variants, balancing performance and cost efficiency for different enterprise and developer use cases.
Huawei has confirmed that its Ascend infrastructure will fully support the new model, with its AI supernode systems contributing to both training and deployment. However, the rollout is expected to scale more effectively later in the year as newer Ascend chips become widely available, addressing current throughput limitations.
The collaboration underscores China’s push toward technological self-sufficiency in AI, where domestic chips and models are being tightly integrated to create a localized stack. Industry analysts see DeepSeek and Huawei forming a core part of this emerging ecosystem, especially as access to Western hardware becomes increasingly restricted.
At the same time, the development intensifies global competition in AI infrastructure, particularly for companies like Nvidia that have traditionally dominated the market. By aligning advanced AI models with homegrown semiconductor solutions, China is positioning itself to compete more independently in the rapidly evolving AI landscape.
