The extreme complexity of the future power grid makes research on grid architecture very important. It is the highest level (topmost) description (model) of the entire grid. This paper discusses the concept of “layered and clustered grid structure” from three perspectives: the expansion of the balance area of the interconnected systems, the mathematical foundation of the laminar flow structure, and learning the intelligence of the Internet. Research has shown that a layered and clustered grid architecture (such as GRIP), based on the concepts of “decomposing the system into a layering structure of clusters and implementing global coordination” and “each cluster maintaining its own net power balance and local selfish optimization”, has a mathematical foundation and is as smart as the Internet, such that it is especially suitable for the needs of future grids. These studies can provide a valuable new vision and new methodologies for power grid modernization planning and design.