Imagine if you could scale up and down the amount of computing capacity you were paying for on a second by second (or even ms by ms) basis! That would be pretty cool :)