The novel concept of near-field velocity sensing is proposed. In contrast to far-field velocity sensing, near-field velocity sensing enables the simultaneous estimation of both radial and transverse velocities of a moving target. A maximum-likelihood-based method is proposed for jointly estimating the radial and transverse velocities from the echo signals. Assisted by near-field velocity sensing, a predictive beamforming framework is proposed for a moving communication user, which requires no channel estimation but achieves seamless data transmission. Finally, numerical examples validate the proposed approaches.