Why attention grows with the square of length

Why does attention's work grow so quickly when you make the input longer?