Ant Group Cuts AI Training Costs by 20% Using Chinese Chips
Ant Group has successfully developed a new AI model training technology that utilizes Chinese-made semiconductors, resulting in a twenty percent reduction in costs. According to reports from Bloomberg, the fintech giant integrated chips from domestic tech giants Alibaba and Huawei with a machine learning architecture known as Mixture of Experts (MoE). This method enhances efficiency by dividing tasks into specialized subsets, a technique also employed by Google and China’s DeepSeek. Notably, Ant Group’s training outcomes using these domestic alternatives reportedly match the performance of Nvidia’s high-end H800 chips. While the company continues to utilize hardware from Nvidia and AMD, this development marks a strategic move to decrease reliance on expensive, high-end GPUs. This approach contrasts with Nvidia CEO Jensen Huang’s perspective, which emphasizes the necessity of more powerful, larger GPUs for AI advancement rather than focusing primarily on cost reduction. The achievement highlights the growing capability of Chinese semiconductor solutions in the competitive global AI landscape, offering a viable alternative amidst surging corporate investments in artificial intelligence infrastructure.
Wire timeline
Ant Group Cuts AI Training Costs by 20% Using Chinese Chips
Ant Group has successfully developed a new AI model training technology that utilizes Chinese-made semiconductors, resulting in a twenty percent reduction in costs. According to reports from Bloomberg, the fintech giant integrated chips from domestic tech giants Alibaba and Huawei with a machine learning architecture known as Mixture of Experts (MoE). This method enhances efficiency by dividing tasks into specialized subsets, a technique also employed by Google and China’s DeepSeek. Notably, Ant Group’s training outcomes using these domestic alternatives reportedly match the performance of Nvidia’s high-end H800 chips. While the company continues to utilize hardware from Nvidia and AMD, this development marks a strategic move to decrease reliance on expensive, high-end GPUs. This approach contrasts with Nvidia CEO Jensen Huang’s perspective, which emphasizes the necessity of more powerful, larger GPUs for AI advancement rather than focusing primarily on cost reduction. The achievement highlights the growing capability of Chinese semiconductor solutions in the competitive global AI landscape, offering a viable alternative amidst surging corporate investments in artificial intelligence infrastructure.
TechNode