Abstract: Artificial Intelligence for Science (AI4S) is giving rise to a new form of knowledge power. On the one hand, it is embodied as “computing power hegemony,” whereby “hard” infrastructure, such as large cloud platforms, imposes a technological discipline on scientific practice through their “black box” nature and API lock-in, fostering new technological dependencies. On the other hand, it manifests itself as a “walled garden”, whereby tech giants leverage foundational models to construct “soft” platform ecosystems, enclosing open knowledge resources as paid services and eroding the public nature of science. These two forms of power reinforce each other, simultaneously undermining scientific autonomy and the public character of knowledge while weakening scientific norms represented by Mertonian norms. A clear understanding of the forms and operational mechanisms of this emerging knowledge power will help reconstruct the public nature of science within this new power landscape.

Key Words: AI for Science; Power discipline; Walled garden; Digital commons; Open science