GPU servers, as a new infrastructure for scientific research in universities and laboratories, are reshaping the paradigm of modern scientific research with their powerful parallel computing capabilities. It not only injects momentum into traditional computationally intensive disciplines - achieving seamless simulations from quantum scale to cosmic scale in physics, completing molecular dynamics explorations that took decades to exhaust in materials science, and pushing high-precision fluid simulation from theory to engineering practice in the engineering field - but more importantly, it has spurred and firmly supported the rise of the revolutionary interdisciplinary field of "AI for Science".