DeepSeek and Huawei are developing software for Ascend processors, while DeepSeek has released programming tools intended to broaden support beyond Nvidia’s ecosystem.
DeepSeek has open-sourced a set of programming tools for Huawei’s Ascend AI processors, deepening a partnership aimed at building software that can run on Chinese hardware. The September 30 announcement includes six software modules, according to the South China Morning Post, and covers compute and communication libraries as well as an Ascend-compatible version of TileLang, a programming language for developing high-performance kernels.
DeepSeek said Huawei provided technical support and that the companies advanced a supernode design combining 128 Ascend 950 chips. The developers describe TileLang as a simpler programming model than Nvidia’s CUDA. DeepSeek’s official DeepGEMM-Ascend repository also records an initial September 30 release and lists support for Ascend 950 devices, giving developers code to inspect and test.
The release addresses a key barrier to using AI accelerators: chips need a usable software stack, not just hardware. A broader set of open-source tools could make it easier for developers to port workloads and experiment with Ascend processors. It also gives China’s AI ecosystem another route to reduce reliance on Nvidia’s CUDA-based development environment, though the announcement alone does not establish that the software or chips match Nvidia’s performance or maturity.
For now, the clearest verifiable step is the availability of code and development libraries, alongside the reported 128-chip system collaboration. Developers will still need compatible Ascend hardware and software to test the tools in practice. The project’s adoption by model builders and infrastructure operators will help show whether the new releases can support a durable alternative ecosystem, rather than just a set of initial ports.