On March 24, Ant Group announced a significant breakthrough: it has successfully deployed domestic chips from Alibaba, Huawei, and other providers in conjunction with the Mixture of Experts (MoE) machine learning approach to train AI models, achieving a cost reduction of approximately 20%. Testing has confirmed that the performance of this innovative technology is on par with that of NVIDIA's H800 chips. While Ant Group continues to utilize NVIDIA chips for AI development, its latest models have predominantly relied on alternative solutions, including AMD and domestic chips. A research paper released this month reveals that some of Ant Group's models have outperformed Meta in benchmark tests, although these results await independent verification. This strategic shift underscores Ant Group's commitment to accelerating the localization of AI technology, with the dual goals of cost reduction and reduced dependence on high-end US chips.
