We present an optimal power scheduling scheme to maximize the throughput of a set of distributed multiple-input multiple-output (MIMO) wireless links. This scheme exploits both spatial and temporal freedoms of the source covariance matrices of all MIMO links. In particular, the source covariance matrix of each MIMO link is allowed to vary within a block of time (and/or frequency) slots. This scheme, also referred to as space-time power scheduling, optimizes an integration of link scheduling and power control for MIMO links. The computational problem involved in this scheme is non-convex. However, a gradient-projection algorithm developed for this scheme consistently yields a higher capacity than all other existing schemes.