DeepSeek says it has partnered with Huawei to develop programming tools for Huawei's Ascend chips, including TileLang, an open-source CUDA alternative (Reuters)

Chinese AI firm DeepSeek announced on September 30 a collaboration with Huawei Technologies to develop programming tools tailored for Huawei's Ascend chips, reflecting a growing trend among Chinese tech companies to reduce dependence on Nvidia's ecosystem. DeepSeek disclosed the news via its WeChat account, revealing plans to open-source the programming infrastructure for Huawei's Ascend platform, which includes compute and communication libraries.

This move came shortly after Huawei unveiled its upcoming AI processors and supernode computing systems, with Huawei forecasting widespread adoption of its AI systems for model training by 2027. DeepSeek emphasized that Huawei provided comprehensive support in the development of the programming infrastructure, together the companies worked on a "supernode" solution utilizing 128 Ascend 950 chips, optimizing both computation and communication processes.

Additionally, DeepSeek highlighted the significance of TileLang, an open-source high-level programming language for AI chips, in enhancing development efficiency and simplifying code logic. The company described TileLang as a universal and programmer-friendly language capable of reaching the hardware's full performance potential, contrasting it favorably with Nvidia's CUDA.

DeepSeek emphasized that creating an independent and self-controlled GPU software ecosystem begins with establishing a universally adopted, easy-to-program high-level language that can fully utilize the hardware's capabilities.