China's top AI continues using Nvidia amid high costs of local chip transition

China's leading AI developers continue to rely on Nvidia chips for training large language models due to high transition costs to domestic alternatives, according to industry sources. The steep engineering bottleneck stems from Nvidia's CUDA software platform, which requires extensive code rewriting to run on Huawei Technologies' Ascend chips, potentially adding at least 50% in time and costs, as estimated by AI researcher James Wang.

China’s top AI is still trained on Nvidia chips. What is delaying a switch to local tech? High transition costs are keeping AI developers in China reliant on Nvidia chips, industry sources say Beijing’s push for self-sufficiency https://www.scmp.com/tech/big-tech/article/3348168/chinas-tech-self-sufficiency-drive-reaches-new-milestone-powerful-risc-v-chips?module=inline&pgtype=article . While domestic hardware continues to advance, changing chip architecture presents a steep engineering bottleneck. “Training LLMs on Nvidia chips for now remains the norm among Chinese AI developers,” said a person familiar with the industry. One of the core hurdles lies in the software ecosystem. Nvidia’s Compute Unified Device Architecture CUDA platform has long been the industry standard for AI development. By contrast, Huawei Technologies’ alternative – Compute Architecture for Neural Networks CANN – requires developers to rewrite and optimise large amounts of code, according to an AI researcher involved in model development. “Our existing training pipelines are reliant on CUDA,” said James Wang, who develops AI models at a research institute affiliated with a Shanghai-based university. “CUDA code cannot run directly on Ascend and requires extensive rewriting.” Wang estimated that migrating existing workflows to Huawei’s Ascend chips could add at least 50 per cent in time and costs for his team.