Alibaba-affiliate Ant uses Chinese, U.S. chips to cut AI costs

BEIJING — Alibaba-affiliate Ant Group is incorporating both Chinese and U.S.-made semiconductors in the development of more efficient artificial intelligence models, as per a source familiar with the situation.

The utilization of a mix of chips not only streamlines the process and cost of training AI models but also diminishes dependence on a single supplier like Nvidia, the source revealed. This approach aligns with the industry trend of utilizing diverse networks, commonly referred to as a mixture of experts, enabling models to be trained with reduced computational resources.

Earlier this month, the company disclosed in a paper that it had successfully employed cost-effective hardware to train its MoE models, resulting in a 20% reduction in computing expenses.

Ant Group, known for operating Alipay, one of China’s leading mobile payment apps, was founded by Jack Ma and is affiliated with Alibaba.

Bloomberg reported on Monday, citing sources, that Ant Group had incorporated chips from Alibaba and Huawei in the training of AI models. While the company previously utilized Nvidia chips, it has now shifted towards alternatives from Advanced Micro Devices and Chinese chips.

Ant Group declined CNBC’s request for comment on the matter.

The company recently announced significant enhancements to its AI solutions for healthcare, which are currently being utilized by prominent hospitals and healthcare institutions in Beijing, Shanghai, Hangzhou, and Ningbo.

The healthcare AI model leverages DeepSeek’s R1 and V3 models, Alibaba’s Qwen, and Ant’s proprietary BaiLing. This specialized model can provide answers to medical queries and enhance patient services, as highlighted in the company’s statement.

The U.S. has implemented measures to restrict China’s AI advancement by limiting Chinese firms’ access to cutting-edge semiconductors essential for training models. Despite this, Nvidia retains the ability to supply lower-end chips to China.