He added that the significance of geographical fine-tuning in AI was underestimated. “It issues greater than folks assume. It shapes how fashions interpret nuance, bias, and context. Baidu’s edge will seemingly be its integration with China’s digital ecosystem and search infrastructure, not simply uncooked mannequin efficiency.”
On the convention, Baidu EVP Dou Shen additionally introduced two new processors geared toward powering the corporate’s advances in AI. The Kunlunxin M100, optimized for big scale AI inference, might be launched at first of 2026, with the M300, optimized for the coaching and inference of ultra-large-scale multimodal massive fashions, following in early 2027.
Along with these two new processors, Baidu additionally unveiled the Tianchi 256 and Tianchi 512 supernodes on the convention. They’re anticipated to formally launch in 2026, with a single Tianchi 512 supernode, consisting of 512 Kunlunxin P800 chips, able to coaching fashions with as much as one trillion parameters, the corporate stated.
