In this paper, we present a resource allocation mechanism for network slices. We consider a dynamic resource allocation model with multiple independent resource providers. We call our allocation mechanism, the soft-max allocation mechanism, where the slices bid for resources and the resource providers allocate resources such that their revenue is close to the maximum revenue possible. We show that this mechanism translates into a game among the slices with a unique Nash equilibrium. We also show that the network utility obtained at this Nash equilibrium is close to the optimal social utility. We then present a stochastic dual sub-gradient algorithm that provably converges to the unique Nash equilibrium.