This paper studies the delay, cost and infrastructure tradeoff of epidemic routing in mobile sensor networks. We consider a mobile sensor network with M mobiles and B static base stations. The mobile sensors collect information when moving around and need to report the information to the base stations. Three different epidemic routing schemes - target epidemic routing, uncontrolled epidemic routing and controlled epidemic routing - are analyzed in this paper. For each of the three schemes, we characterize the scaling behaviors of the delay, which is defined to be the average number of time slots required to deliver a message, and the cost, which is defined to be the average number of transmissions required to deliver a message, in terms of the number of mobiles (M) and the number of base stations (B). These scaling results reveal the fundamental tradeoff among delay, cost and infrastructure in mobile sensor networks.