Ant Group has developed AI model training technology using Chinese-made chips, reducing costs by 20%, according to Bloomberg. The fintech giant used semiconductors from Alibaba and Huawei, along with a machine learning method called mixture of experts, or MoE, which splits tasks into specialized subsets for efficiency. MoE models, also used by Google and China’s DeepSeek, have gained traction amid surging corporate AI investments. Ant’s training results reportedly match those of Nvidia’s H800 chips, though it continues to use Nvidia hardware alongside alternatives like AMD and domestic chips. The company aims to eliminate reliance on high-end GPUs, a contrast to Nvidia CEO Jensen Huang’s stance that advancing AI requires more powerful chips rather than cost-cutting. Huang advocates larger GPUs with more processing cores to drive revenue growth. [Bloomberg]
0 Commentaires