Deep Seek has released six open-source software modules built for Huawei Technologies' Ascend AI chips, the two Chinese firms disclosed on Wednesday1,2. The tools include TileLang, positioned as an open-source alternative to CUDA, Nvidia's proprietary GPU programming framework.
The modules mirror Deep Seek's existing open-source tooling for Nvidia hardware, transplanting the same developer workflow onto Huawei silicon. Deep Seek described the goal as building a new "independent and controllable" software ecosystem for GPUs.
Deep Seek said it partnered with Huawei Technologies to develop the programming tools specifically for the Ascend chip line. Six software modules were released simultaneously. The South China Morning Post characterized the release as part of China's broader push to end dependence on Nvidia processors under ongoing U.S. export curbs.
Nvidia's dominance in AI training rests not only on chip performance but on CUDA, the software layer that locks developers into Nvidia hardware. TileLang targets that lock-in directly, offering an open-source programming interface for Ascend GPUs. ANALYSIS By replicating its own Nvidia-oriented toolchain on Ascend, Deep Seek lowers the switching cost for Chinese AI labs that already use Deep Seek's open-source stack but face constrained access to Nvidia silicon.
The partnership pairs China's most prominent open-source AI lab with its leading domestic chip designer, concentrating both the hardware and the software layers of an alternative AI compute stack under Chinese control.