DeepSeek and Huawei have announced a collaboration on programming tools and infrastructure for Huawei’s Ascend AI chips. In its September 30, 2026 announcement, DeepSeek said it plans to make the infrastructure open source, including libraries for computing and communications. The work also covers a reported supernode solution based on 128 Ascend 950 chips.
Programming infrastructure for Ascend
The tools are intended for Huawei’s Ascend platform. DeepSeek says the planned open-source infrastructure will include computing and communications libraries—collections of code that help software perform specific tasks and work with a platform.
DeepSeek also said Huawei gave full support to the infrastructure-development effort. The collaboration is part of an effort to reduce reliance on Nvidia’s ecosystem.
TileLang and Nvidia CUDA
DeepSeek highlighted TileLang, an open-source, high-level programming language for AI chips. A high-level language lets developers work with programming concepts without writing every instruction at the hardware level.
DeepSeek says TileLang can improve development efficiency and simplify code logic. The company describes TileLang’s programming model as simpler than Nvidia’s CUDA system.
The 128-chip Ascend 950 solution
The collaboration also includes a reported supernode solution based on 128 Ascend 950 chips. Its stated optimization focus is both computation and communications, bringing the programming-software work together with a specific multi-chip system configuration.
