DeepSeek has announced a new partnership with Huawei to develop programming tools optimized for Huawei’s Ascend AI chips, marking another major step in China’s effort to reduce reliance on Nvidia’s ecosystem. The Chinese artificial intelligence firm said on September 30 that it is open-sourcing programming infrastructure for the Ascend platform, including compute and communication libraries. The move follows Huawei’s recent unveiling of next-generation AI processors and supernode computing systems.
Targeting Nvidia’s Software Ecosystem
The partnership directly addresses one of the main barriers for rival AI processors, namely Nvidia’s CUDA software platform, which has become deeply embedded in AI development. DeepSeek and Huawei have developed a toolkit based on the open-source GPU programming language TileLang, designed to support rapid AI development on Huawei hardware. The companies are releasing computing and communication libraries that help software interact more effectively with Ascend processors.
What TileLang Brings to Developers
TileLang is a high-level programming language intended to simplify the process of writing optimized software for AI chips. According to DeepSeek, it offers a simpler programming model than Nvidia’s CUDA while still allowing developers to access the full performance potential of compatible hardware. The language lets programmers describe computational tasks at a higher level, reducing the need to manage detailed low-level hardware instructions associated with each processor.
Supernode System Built on Ascend 950
DeepSeek said Huawei provided extensive support in developing the new programming infrastructure, and the two companies jointly advanced a supernode solution based on 128 Ascend 950 chips. This configuration is intended to improve both computation and communication, reducing data exchange bottlenecks for demanding AI workloads. The announcement came just two weeks after Huawei showcased its next-generation AI processors and said it expects its AI systems to be widely used for model training in 2027.
Huawei’s Hardware Roadmap
Huawei is also building more powerful silicon to support domestic AI training. On September 17, the company announced upcoming flagship processors, the Ascend 960DT and Ascend 960PR, expected in the first and third quarters of 2027 respectively. These chips aim to more than double training and inference performance, and Huawei plans to launch a new data format and computing architecture that could allow up to one million Ascend chips to work together for multi-trillion parameter models.
Export Controls and Domestic AI Independence
The partnership reflects a broader push among Chinese technology companies to develop alternatives to American semiconductor technology, especially after US export restrictions limited access to advanced Nvidia processors. Huawei has become the government-backed hardware option for Chinese AI, benefiting from those restrictions, while DeepSeek has publicly sought to reduce its dependence on Nvidia. DeepSeek had already adapted its V4 AI models for Huawei’s Ascend processors earlier in 2026, and the new partnership expands that relationship into foundational software infrastructure.
DeepSeek and Huawei’s open-source initiative could give AI developers, researchers, and businesses another option beyond Nvidia’s software stack, especially for workloads on Chinese hardware. Widespread adoption will depend on real-world performance, compatibility, documentation, and long-term support, areas where CUDA still maintains an advantage. The partnership signals a meaningful step toward a more independent Chinese AI computing ecosystem, though hardware availability and manufacturing constraints remain important limiting factors.