AI-powered scientific computing represents a paradigm shift in the realm of scientific research and computation that integrates artificial intelligence into traditional scientific computing, offering unprecedented opportunities for enhanced efficiency and accelerated scientific discovery. As the AI models are data hungry, one challenge comes from the scarcity of data in many important scenarios. In this talk, we will introduce several designs of AI approaches that incorporate the physical knowledge into the AI models to enhance the efficiency and accuracy in evolving the dynamic systems governed the partial differential equations.