The generation of motor patterns has been the focus of several recent experimental studies. Recordings have shown that populations of neurons in motor cortex transition into a preparatory state while a movement is being planned, and engage in transient collective dynamics of large amplitude during its execution. We investigate this phenomenon in rate models of cortical dynamics. Weakly coupled networks cannot produce the substantial transient departure from background activity observed in the experiments. Strongly coupled random networks with their inherent chaotic dynamics, on the other hand, do not capture the transient nature of movement-related activity. Here we introduce a new class of models with strong and complex excitatory recurrence, and inhibitory feedback of matching complexity to stabilize the dynamics. We show that such inhibition-stabilized networks transiently amplify certain network states. The network activity can be forced to arrive at one of those states by the end of the preparatory period through the delivery of an appropriate external input. Upon a go-signal, the input is withdrawn and the network is released to elicit transient single-neuron and collective dynamics that match the data well. In particular, we reproduce the recently uncovered phenomenon of rotational dynamics during the execution of movement [Churchland et al (2012)]. Additionally, muscle activity may be read out from these noisy transients to produce complex movements. Surprisingly, inhibition-stabilized circuits connect several previously disparate aspects of balanced cortical dynamics. The mechanism that underlies the generation of large transients here is a more general form of “balanced amplification” [Murphy and Miller (2009)], which was previously discovered in the context of visual cortical dynamics. Furthermore, during spontaneous activity in inhibition-stabilized networks, a detailed balance of excitatory and inhibitory inputs to single cell exists that is much finer than expected from shared population fluctuations.