While often believed to be a passive agent that merely exploits its host´s metabolism, the influenza virus has recently been shown to actively move across glycan-coated surfaces. This form of enzymatically driven surface motility is currently not well understood and has been loosely linked to burnt-bridge Brownian ratchet mechanisms. Starting from known properties of influenza´s spike proteins, we develop a physical model that quantitatively describes the observed motility. It predicts a collectively emerging dynamics of spike proteins and surface-bound ligands that combined with the virus´ geometry give rise to a self-organized rolling propulsion. We show that in contrast to a Brownian ratchet, the rotary spike drive is not fluctuation driven but operates optimally as a macroscopic engine in the deterministic regime. The mechanism also applies to relatives of influenza and to man-made analogs like DNA monowheels and should give guidelines for their optimization.