DeepSeek and Huawei Challenge Nvidia's CUDA Dominance in China

DeepSeek, a Chinese artificial intelligence company, has joined forces with Huawei to challenge Nvidia's dominance in the AI software ecosystem. The collaboration aims to reduce dependence on Nvidia through software self-reliance amid growing difficulty in securing hardware due to U.S. government semiconductor export controls.

The two companies have developed a programming language called 'TileLang,' compatible with Huawei's AI chip 'Ascend.' TileLang is designed to offer a simpler programming approach than CUDA while maximizing hardware performance.

DeepSeek emphasized that a general-purpose, easy-to-program high-level language is essential for building an independent and autonomous GPU software ecosystem. The company currently uses TileLang as a core tool in its artificial general intelligence (AGI) research and has completed performance testing on older Nvidia chips.

Huawei provided extensive technical support to DeepSeek throughout the development process, and the two companies have also jointly pushed the computing and communication performance of a supernode solution comprising 128 Huawei 'Ascend 950' chips to its limits.