View of Ali Group building in Chonggin, China on March 23, 2025.
cfoto | Future Publishing | Getty Images
BEIJING – The Alibaba-Affiliate Ant group uses both Chinese and US semiconductors to build more efficient artificial intelligence models, according to sources familiar with the issue.
Chip combinations not only reduce the time and cost of training AI models, but also limit reliance on a single supplier, such as NVIDIA, have focused on industry trends tapping multiple networks known as a mixture of experts, according to sources.
Earlier this month, the company said it could effectively train its own MOE models using low-cost hardware, reducing computing costs by 20%.
ANT runs Alipay, one of China’s two main apps for mobile payments. Jack Ma founded the company and its affiliate Alibaba.
Bloomberg reported Monday that AI used Alibaba and Huawei chips to train AI models, citing sources. According to the Bloomberg Report, ANT also used Nvidia chips, but now relies on advanced microdevices and alternatives from Chinese chips.
Ant declined to request CNBC’s comment.
On Monday, the company announced a “major upgrade” to healthcare AI solutions. It says it is used by seven major hospitals and healthcare institutions, Beijing, Shanghai, Hangzhou and Ningbo.
The healthcare AI model is built on bail from DeepSeek’s R1 and V3 models, Alibaba’s Qwen and Ant’s own bail. ANT’s healthcare-specific model can answer questions about healthcare topics and, according to the company’s statement, will also help improve patient services.
The US is trying to limit Chinese AI development by restricting Chinese companies’ access to the most advanced semiconductors used in training models. Nvidia is able to sell low-end chips to China.
