According to sources familiar with the matter, Ant Group Co., a company backed by Jack Ma, utilized Chinese-manufactured semiconductors to develop methods for training AI models. This approach was expected to reduce costs by 20%. The company reportedly employed Chinese chips, including those from Alibaba Group Holding Ltd. and Huawei Technologies Co., to train models using the Mixture of Experts machine learning technique. While Ant is still using Nvidia chips like the H800, they are increasingly turning to alternatives such as those from Advanced Micro Devices Inc. and Chinese manufacturers for their latest models.

Ant's utilization of local semiconductor alternatives for AI development reflects the competition between Chinese and US firms in the field. The company's results have raised interest, particularly as they claim to outperform Meta Platforms Inc. in certain benchmarks. These advancements could signify a significant step forward for Chinese artificial intelligence by reducing inferencing costs and supporting AI services.

The adoption of MoE models, a technique that divides tasks into smaller sets of data akin to having specialized teams for each job segment, has gained popularity within the industry. This method has been utilized by notable companies such as Google and DeepSeek. While training MoE models traditionally relies on high-performance chips like Nvidia's GPUs, their cost can be prohibitive and hinder broader usage. Ant's research paper outlines their effort to efficiently train LLMs and scale models without the need for premium GPUs, addressing this limitation.