Huge amounts of data need to be handled in present day society. Despite of increasing (digital) computing power, new approaches to information processing are desired and required. Reservoir computing represents such an alternative approach towards computation. Usually it is implemented using a large network of interconnected nonlinear nodes (or neurons). It has been shown that reservoir computing serves universal computational properties, such that any potential operation could be realized, outperforming other approaches for certain tasks. We have identified that delay-coupled systems are ideally suited for reservoir computing. Time-delayed coupling or feedback originally appeared in control systems where it naturally arises because a finite time is required between sensing the information and reacting to it with a control signal. The presence of delay, however, can also give rise to loss of stability and emergence of complex dynamics. Delay delivers the required complexity needed to implement reservoir computing. In this way, we target to achieve high computational performance with only a small number of nodes, or even one single nonlinear node, thus replacing a network of hundreds of nonlinearly coupled elements by a few nonlinear elements coupled with delay.