DeepSeek Opens Ascend Tooling Built With Huawei as a CUDA Alternative

DeepSeek said on September 30 that it partnered with Huawei Technologies to develop programming tools for Huawei’s Ascend chips, as Chinese technology companies seek alternatives to Nvidia’s ecosystem. [2][3] In a post on its official WeChat account, DeepSeek said it was open-sourcing infrastructure for Huawei’s Ascend platform, including the high-level TileLang programming language and related compute and communication libraries. [3]

DeepSeek said the priority in building a new generation of independently controlled GPU software ecosystem is a universal, easy-to-program high-level language that can still fully exploit hardware performance. [3] According to the company, TileLang was created for that need and offers software developers a simpler programming model than Nvidia’s CUDA, improving development efficiency and simplifying code logic. [3]

DeepSeek and Huawei are not merely presenting Ascend hardware as an alternative; they are addressing the programming layer that helps developers use it. The explicit comparison is TileLang versus Nvidia’s CUDA: DeepSeek claims a simpler programming model while emphasizing hardware performance, development efficiency, and cleaner code logic. Adoption will therefore depend on whether the open-sourced language, compute libraries, and communication libraries form a workable toolchain. What the sources don't address: how TileLang and Ascend perform against CUDA in independent benchmarks, or how broadly developers have adopted the new tools.