Deepseek has released open-source programming tools for Huawei's Ascend chips, aiming to help China's AI industry build an independent software ecosystem. Announced via Deepseek's official WeChat channel with full support from Huawei, the release features TileLang, an open-source programming language originally developed at Peking University that provides a simpler programming model than Nvidia's CUDA. The companies also optimized a supernode cluster consisting of 128 Ascend 950 chips as part of the collaboration.
This partnership addresses a major bottleneck for China's domestic semiconductor sector, where hardware makers have historically struggled to match the software ecosystem built around Nvidia's CUDA platform. While Chinese model developers like Z.ai and Moonshot AI have advanced rapidly, Huawei is working to close the hardware-software gap, having recently unveiled new AI processors and supernode systems intended for widespread model training. Huawei's rotating chairman Eric Xu noted that the company must reduce its reliance on foreign chip sales amid US export controls.
The move comes as industry analysts debate the durability of Nvidia's software advantage, particularly as alternative stacks like Huawei's CANN and open-source languages emerge to support advanced models on non-Nvidia hardware. Although Nvidia maintains a strong lead in complex agent workloads and multi-chip system coordination, initiatives like Deepseek's TileLang integration represent a significant step toward alternative AI chip ecosystems. This collaboration matters because overcoming the software moat is essential for the future independence and scalability of China's artificial intelligence infrastructure.