Photo credit: www.cnbc.com
Ant Group Leverages Diverse Semiconductors for AI Advancements
BEIJING — Ant Group, the financial affiliate of Alibaba, is strategically incorporating both Chinese and U.S.-manufactured semiconductors to enhance its artificial intelligence model efficiency, according to insiders familiar with the company’s operations.
This approach of utilizing a mix of chips not only accelerates the training process for AI models but also mitigates the risks associated with dependence on a singular supplier like Nvidia. The industry is increasingly adopting a technique known as “mixture of experts,” which enables the training of models with significantly reduced computational resources, making for a more cost-effective solution.
In a recent publication, Ant highlighted its success in employing lower-cost hardware to train its own mixture of experts models, resulting in a remarkable 20% reduction in computing expenses.
As the operator of Alipay, one of China’s predominant mobile payment platforms, Ant Group was founded by Jack Ma, who also co-established Alibaba.
According to Bloomberg, sources indicate that Ant has integrated semiconductors from Alibaba and Huawei into its AI training processes. While Nvidia chips have been part of its repertoire, the company is now shifting focus toward alternatives from Advanced Micro Devices and various Chinese manufacturers.
Despite requests for comments, Ant Group did not respond to queries from CNBC.
On Monday, the company revealed significant upgrades to its AI offerings specifically tailored for the healthcare sector. These advancements are reportedly being utilized in seven major hospitals and healthcare institutions across key cities such as Beijing, Shanghai, Hangzhou, and Ningbo.
The healthcare AI initiative incorporates models like DeepSeek’s R1 and V3, Alibaba’s Qwen, as well as Ant’s proprietary BaiLing. This specialized healthcare model is designed to provide informative responses to medical inquiries and enhance patient services, according to Ant’s official statement.
In the broader context, the U.S. government has made efforts to curtail China’s progress in artificial intelligence by restricting access to advanced semiconductors crucial for training these models. Nevertheless, Nvidia continues to sell its lower-tier chips to Chinese entities.
Source
www.cnbc.com