The keyword is "high impact". The link you provided is only ranked by publication count, it doesn't consider the impact factor of journal/conferences. If you look at stats from ICML this year[0], the only Chinese institution TsingHua is ranked #13 in the top #20 list.
Take a look at recent major advances in deep learning: information bottle neck, mask-rcnn, transformer, normalizing flow, all originated from US/Canadian/EU institutions. Chinese academia has very serious systematic issues that discourages innovation and encourages quantity over quality.