The mitigation of airport surface congestion is an important step towards increasing the efficiency of the air transportation system, and decreasing flight delays. This paper proposes a strategy to control the release of departing flights from their gates with the specific objective of reducing their taxi times and fuel consumption, while limiting the impact on airport throughput. The proposed strategy also explicitly accounts for the practical constraints that arise due to limited gate resources at the airport. A stochastic network abstraction of the airport surface is used to model aircraft movement, and the optimal release time for each aircraft is calculated using dynamic programming. Simulations of operations at Boston's Logan International Airport in the US are used to illustrate the advantages of the proposed policies.