Chinese artificial intelligence developer DeepSeek has entered a partnership with Huawei to build chip programming tools, Reuters reports. The stated goal of the collaboration is to reduce reliance on Nvidia, the US supplier that currently dominates the market for accelerators used in training and running large AI models.

The deal links two companies that sit at the center of China's push to build a domestic AI compute stack. DeepSeek drew global attention for training competitive large language models at a fraction of the cost typically associated with frontier AI development. Huawei, meanwhile, designs its own line of processors, including the Ascend series of AI accelerators, and has positioned itself as the leading domestic alternative to Nvidia's GPUs.

Chip programming tools form a critical, and often overlooked, layer of that stack. Writing efficient code for AI accelerators requires compilers, libraries, and development frameworks tailored to the underlying hardware. Nvidia's strength in this area — its CUDA software ecosystem, built up over nearly two decades — is widely regarded as one of the deepest moats protecting its market position. Developers around the world write and optimize their workloads against CUDA, which makes switching to alternative hardware costly and slow.

That is precisely the gap the DeepSeek–Huawei partnership targets. By developing programming tools jointly, the two companies aim to make Huawei's silicon easier to use for the kind of demanding AI workloads DeepSeek runs. For Huawei, better software support increases the practical attractiveness of its accelerators. For DeepSeek, it opens a path to train and serve models on domestic hardware rather than continuing to depend on Nvidia products, which have been subject to US export restrictions aimed at limiting China's access to top-tier AI chips.

The partnership arrives amid sustained pressure on China's access to advanced semiconductors. Washington has progressively tightened export controls on Nvidia's highest-performance accelerators, forcing Chinese AI developers to work with downgraded chips approved for sale in the country or to seek domestic alternatives. Huawei's Asceng processors have emerged as the most prominent of those alternatives, though they still trail Nvidia's flagship products in raw performance and, critically, in software maturity.

A collaboration between a top Chinese AI lab and the country's leading chip designer addresses both sides of that equation. DeepSeek brings direct, hands-on experience with the practical requirements of training frontier-scale models, along with a demonstrated focus on engineering efficiency. Huawei brings the hardware. Combining the two could accelerate the development of a software ecosystem that makes Chinese-made accelerators viable for the most demanding workloads.

The move also carries signal value for the broader market. If Chinese AI developers can build tooling that narrows the software gap with CUDA, the competitive position of domestic accelerators strengthens — not just for DeepSeek's own workloads, but for other Chinese companies evaluating where to buy compute. Conversely, it marks another step in the bifurcation of the global AI hardware market, with a US-aligned ecosystem centered on Nvidia and a Chinese ecosystem assembling around Huawei and its partners.

Reuters, which first reported the partnership, did not disclose financial terms or a timeline for the resulting tools. Neither company has publicly detailed which hardware lines the tools will support or when developers outside the partnership might gain access to them.