Let's consider the execution of an algorithm. In the classical case you start with _one_ of 2^n states and keep evolving that single state. In the case of a quantum computer you can start with a superposition of all of the 2^n states and evolve them simultaneously. This is called quantum parallelism.
This sounds very powerful but comes with a caveat: at the end of the algorithm you will end up with some other superposition of the 2^n states but you can only read one of them. Which one it is going to be will be decided by the square of the complex amplitudes associated with those states. The greater the squared amplitude, the more likely it is that you will end up measuring the state associated with it. A key requirement when a quantum algorithm is designed is that it should yield a useful state at the time of measurement with high probability. This is in general difficult to do.