On the afternoon of March 24, Ant Group addressed reports from international media regarding the training costs associated with its AntBrain large model. The company affirmed that it is continually refining AI applications across various chip platforms and has achieved notable advancements, which it intends to progressively disclose through open-source initiatives. Prior international media coverage had highlighted that Ant Group leverages domestic chips from Alibaba and Huawei, in conjunction with the Mixture of Experts (MoE) machine learning technique, to train AI models, achieving approximately a 20% reduction in costs. Testing has confirmed that the performance of this technology is on par with that of the NVIDIA H800 chip.
