Ant Group cuts AI costs using Chinese semiconductors

Ant Group uses Chinese semiconductors to cut AI training costs by 20%, competing with US firms like Nvidia.

In Short

Jack Ma-backed Ant Group has developed cost-effective AI training techniques using Chinese semiconductors, cutting costs by 20% and producing results comparable to Nvidia. As the company pivots towards local alternatives in response to US bans, its models may significantly enhance Chinese AI development and reduce costs for services.

Jack Ma-backed Ant Group Co. has developed cost-effective techniques for training AI models using Chinese-made semiconductors, reportedly reducing costs by 20%.

The company utilised domestically produced chips from affiliates like Alibaba and Huawei, employing the Mixture of Experts machine learning method, which produced results comparable to Nvidia’s H800 chips.

While Ant continues to use Nvidia for some AI development, it is increasingly leveraging alternatives such as Advanced Micro Devices and Chinese chips for its latest models.

This development positions Ant in competition with Chinese and US firms, especially following DeepSeek’s demonstration of cost-effective model training compared to major investments by OpenAI and Google.

The move highlights the shift of Chinese companies towards local alternatives in response to the US ban on advanced Nvidia semiconductors, including the powerful H800 model.

Ant recently published a research paper claiming that its models sometimes outperform those of Meta in specific benchmarks, a claim that Bloomberg has not independently verified. If confirmed, these models could significantly advance Chinese AI development by reducing inference costs for AI services.

As AI investment grows, Mixture of Experts models are becoming widely adopted due to their efficiency, dividing tasks into smaller data sets.