With the rapid growth of cloud computing, firms face a dizzying array of choices and pricing structures for performing their computing tasks on the cloud. Unlike captive computing resources, cloud computing occurs as a pay-as-you-go contract, similar to the provision of electricity. We develop a method to reduce the rental cost of completing a given computing task with a certain deadline. The current practice is to use a single computing resource that can get the task done in the cheapest possible manner. Instead, costs can be significantly reduced if the task is switched between multiple resources, some more powerful and others less powerful. We apply our method to a real computing task at Cidewalk and show that costs can be significantly reduced.