DeepSeek, Huawei Join Forces to Challenge Nvidia’s CUDA Dominance
Chinese artificial intelligence company DeepSeek has joined forces with Huawei to release its own programming tools designed to rival Nvidia’s software ecosystem, CUDA. The strategy is aimed at reducing dependence on Nvidia through software self-reliance, amid growing difficulty in securing hardware
Chinese AI firm DeepSeek has teamed up with hardware manufacturer Huawei to create programming tools that challenge Nvidia's CUDA platform. The move aims to lessen reliance on Nvidia's products amid U.S. export controls making hardware acquisition difficult. DeepSeek unveiled a free open-source programming language called "TileLang" compatible with Huawei's AI chip "Ascend" via its WeChat platform.
This new software suite offers computation and communication libraries, designed to be simpler than CUDA while maintaining high performance. DeepSeek sees TileLang as crucial for building an independent AI software ecosystem. The company utilizes TileLang in its general AI research and has tested its performance on older Nvidia chips.
Beyond tools, DeepSeek and Huawei are optimizing hardware together, pushing a 128-chip "Ascend 950" configuration to its limits. Huawei displayed its resolve for self-reliance at a Shanghai event, where CEO Eric Xu vowed China would not be dictated to by external forces in tech. DeepSeek aims to deploy 160,000 "Ascend 950DT" chips in its new Inner Mongolia data center, while still using Nvidia chips for model training and Huawei for inference.
Industry analysts believe the software release will lower the hurdle for AI development in China, as Huawei faces domestic demand limitations and encourages a closed yet robust ecosystem of home-grown chips and software.
Written by urgent.news from BusinessKorea's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
