In academic inter-cloud environments, virtual machines communicate with other virtual machines in different virtual L2 network segments through a virtual router. As virtual routers are deployed on certain physical machines, communication between virtual machines via virtual routers could not be an optimal in physical sense. We are proposing an academic inter-cloud architecture (AIC)  which consists of AIC compute and AIC storage. AIC is a hub of clouds and it provides compute physical resource on-demand for each cloud. It means that AIC provides extension to the clouds. The clouds meet in AIC. In that sense, the virtual machines in each cloud are physically close. However, the communication between virtual machines which are located in different clouds has another level optimization problem. The communication between the virtual machines is through the gateways of each cloud. The latency cloud be much smaller if we can optimize the route. For cloud scalability we may provide bare metal servers at geographically distributed manner but the network latency should be optimized by application requirements. We propose an architecture to solve these issues which has L3 function delegation mechanism to open flow switches to reduce network latency according to users' requirements for short cuts.