Chinese artificial intelligence company DeepSeek has partnered with Huawei Technologies to develop programming infrastructure optimized for Huawei’s Ascend AI processors, strengthening efforts by China’s technology industry to build an alternative to Nvidia’s dominant AI computing ecosystem.
DeepSeek said Wednesday in a post on its official WeChat account that it is open-sourcing programming infrastructure for Huawei’s Ascend platform, including libraries designed to improve computing and communication performance.
The collaboration comes as Chinese technology companies deepen cooperation around domestic AI hardware and software at a time when access to Nvidia’s advanced processors remains constrained by U.S. export controls.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Rather than treating the chip itself as the entire computing platform, the partnership targets one of the more difficult parts of competing with Nvidia: the software layer that allows developers to efficiently program and deploy AI workloads on alternative processors.
DeepSeek said Huawei provided full support for development of the programming infrastructure and that the two companies jointly advanced a “supernode” solution based on 128 Ascend 950 chips. The work covers both computation and communication, suggesting an effort to address the performance of large AI systems as a complete cluster rather than focusing solely on individual processors.
The announcement follows Huawei’s unveiling two weeks ago of its next generation of AI processors and supernode computing systems. Huawei said at the time that it expected its AI systems to see broad use for model training next year.
The significance of the collaboration extends beyond another Chinese AI model company adopting domestic chips.
Nvidia’s advantage in AI computing has been built not only around its GPUs but also around CUDA, the company’s mature software platform that gives developers tools and libraries for programming AI workloads. That ecosystem has become deeply embedded in AI research and commercial model development.
DeepSeek’s comments point directly at that software challenge.
“To build a new generation of independent, self-controlled GPU software ecosystems, the first priority is establishing a high-level language that is universal, easy to program, and still capable of reaching the hardware’s full performance potential,” the company said.
“TileLang was created precisely to meet this need,” DeepSeek added, describing the system as offering “a simpler programming model” than Nvidia’s CUDA.
TileLang is a high-level, open-source programming language designed for AI chips. DeepSeek said it can improve development efficiency while simplifying code logic, potentially making it easier for developers to adapt AI applications to hardware other than Nvidia’s GPUs.
Replacing Nvidia hardware is considerably harder if developers must also rewrite or heavily optimize their software for every alternative chip architecture.
A competitive domestic ecosystem therefore requires compatibility, developer tools, libraries and programming abstractions that allow researchers and companies to move workloads without incurring prohibitive engineering costs.
Huawei’s Ascend Push Gains An Important AI Partner
Huawei has been developing its Ascend processor family as a domestic alternative to Nvidia’s AI accelerators, while Chinese technology companies have increasingly sought ways to reduce their dependence on foreign semiconductor technology.
DeepSeek’s involvement could give Huawei’s hardware push greater relevance because DeepSeek has become one of China’s most prominent AI model developers.
The collaboration also illustrates a broader shift in China’s AI industry from individual companies developing isolated alternatives toward greater integration between model developers, chipmakers and software engineers.
The 128-chip Ascend 950 supernode sends a powerful message in that context. Large language models and other advanced AI systems depend on clusters containing large numbers of accelerators, meaning the ability to efficiently connect and coordinate processors can become as important as the performance of an individual chip.
DeepSeek said its joint work with Huawei optimizes both computation and communication. That suggests the companies are addressing the full stack required to train and run large models, from programming abstractions to chip-level execution and communication across a cluster.
The open-source element could also broaden the impact beyond the two companies. If programming infrastructure, libraries, and high-level tools become available to other developers, the benefits could extend across China’s AI industry, reducing the duplication of engineering work required to support domestic accelerators.
This means the deeper objective is not simply to make DeepSeek’s models run on Huawei chips. It is to make Huawei’s chips easier for a wider pool of developers to use.
That is the more difficult part of China’s effort to build a self-reliant AI computing ecosystem. Producing a capable accelerator is one challenge; persuading developers to build around it, while matching the software efficiency and ease of use of Nvidia’s established platform, is another.
DeepSeek’s partnership with Huawei puts that software battle closer to the center of China’s AI hardware strategy. It is believed that if TileLang and the broader Ascend programming infrastructure can make it substantially easier to move AI workloads onto domestic processors, Chinese companies could gradually reduce the software and hardware dependence that has helped make Nvidia’s ecosystem so difficult to displace.