This paper borrowed the idea of molecular diffusion and proposed a proactive caching scheme based on content diffusion. In this mechanism, the conception of content concentration was introduced to analyze user demands for different contents in different network regions. In order to achieve fast content placement and increase the probability of providing responses to user requests by the nearer caching nodes, the content replicas were proactively pushed forward or migrated according to the content concentration difference between nodes. Furthermore, by synthetically considering the content popularity, a probabilistic content placement method was also implemented. The simulation results have shown that this scheme can effectively decrease the average access cost and improve the cache hit ratio.