Ant Group Cuts AI Training Costs 20% with Chinese Chips

Ant Group has successfully reduced artificial intelligence model training costs by 20% through the use of Chinese-made semiconductors from Alibaba and Huawei, adopting a machine learning method known as Mixture of Experts (MoE). This shift comes as a response to U.S. export restrictions, which have limited access to advanced Nvidia chips, prompting Ant to explore alternatives from AMD and local manufacturers. The performance of Ant's models reportedly matches that of Nvidia's H800, despite still utilizing some Nvidia hardware. This trend reflects a broader movement among Chinese tech firms toward local chip solutions to bolster self-reliance in AI development. Additionally, Ant's advancements in AI are contributing to increased investor confidence in Chinese tech stocks, underscoring the country's ambitions in the global AI landscape. Meanwhile, Nvidia's CEO maintains that future AI advancements will necessitate more powerful chips, contrasting with Ant's cost-cutting approach.