A Tsinghua University analysis of 41.3 million research papers published in Nature finds AI adoption increases individual scientists' paper output 3.02x and citations 4.84x, while shrinking the collective breadth of scientific topics studied by 4.63% and cross-researcher collaboration by 22%. AI-augmented work concentrates in data-rich areas, producing a Matthew effect at the field level: individual scientists accelerate while the overall diversity of science narrows.