This work tackles the problem of user tracking in the multi-user scenario. User tracking is one of the key functional elements in millimeter wave (mmWave) communications that heavily rely on directional beamforming to overcome significant path loss. A naive strategy is to track users one by one but this not only introduces great overhead but also reduces the tracking update frequency when the number of users or antennas is large. In addition, user tracking based the angular information, namely angle of arrival (AoA)/angle of departure (AoD) often requires a good initial estimate and an iterative procedure to ensure the accuracy. Aiming to improve the tracking efficiency for multiple users, we propose to track multiple users simultaneously by partitioning users into groups based on the beamspace multi-input and multi-output (MIMO) channel representation. We formulate the joint user grouping and precoding as a mixed-integer nonlinear programming (MINP) problem and propose a low-complexity grouping algorithm. Simulation results demonstrate the significant improvement of the proposed multi-user tracking scheme over two existing approaches in terms of the spectral efficiency.