DeepSeek and Huawei Technologies are developing open-source software for Huawei's Ascend AI chips, seeking to make it easier for developers to run and optimize AI workloads without Nvidia hardware. The collaboration includes TileLang, a programming language DeepSeek describes as simpler than CUDA, and a 128-chip Ascend 950 supernode with optimized computing and data-transfer software. The initiative targets Nvidia's software advantage: Nvidia says more than 7.5 million developers use CUDA and related tools, while export restrictions and Chinese rules have curtailed its ability to compete broadly in China's data-center computing market. DeepSeek's work adds to its wider shift toward Huawei infrastructure, including plans for an Inner Mongolia facility using more than 160,000 Ascend chips. CUDA retains a substantial lead through its extensive libraries and tools, and shifting AI workloads from it is expected to require years.