Ant Group, founded by billionaire Jack Ma, has used domestic Chinese chips to develop artificial intelligence (AI) model training techniques at 20% cheaper costs.
The information was exclusively published on March 24 by Bloomberg news agency citing a number of informed sources. Accordingly, Ant Group uses domestic chips from Chinese corporations such as Alibaba and Huawei to train AI models using the Mixture of Experts (MoE) machine learning method.
The results are similar to when using Nvidia's H800 chip (USA).
Ant Group logo at a conference in Zhejiang province, China.
MoE is a technique that breaks down tasks into smaller datasets, similar to assigning work to specific groups of experts, making the process more efficient. Training MoE models often relies on high-performance, high-priced chips like Nvidia’s. That price is a barrier for many small companies.
Ant Group is still using Nvidia semiconductors to develop AI but is now relying more on alternatives from AMD (US) or domestic chips for its latest models.
Following the success of DeepSeek, China's homegrown AI model that has been causing a stir around the world in recent months as it is said to have capabilities comparable to Western models but at a much lower cost, Chinese AI companies are trying to reduce their dependence on Nvidia's new generation of high-performance chips amid intensifying US-China technology competition.
While not the most advanced, the H800 is considered one of the relatively powerful processors and is currently banned from export to China by the US. Nvidia can still sell low-end chips to China.
According to CNBC, Ant Group's combination of using Chinese and American chips not only reduces the time and cost of training AI models but also limits dependence on a single supplier.
Ant Group owns Alipay, one of the two largest mobile payment apps in China. Billionaire Jack Ma founded Ant Group and Alibaba.
Source: https://thanhnien.vn/ant-group-giam-20-chi-phi-huan-luyen-ai-nho-chip-trung-quoc-185250324175003403.htm
Comment (0)