The 1960 M9.5 Valdivia and 1964 M9.2 Alaska earthquakes caused a decimeters high secondary zone of uplift a few hundred kilometers landward of the trench. We analyze GPS data from the 2010 M8.8 Maule and 2011 M9.0 Tohoku-Oki earthquakes to reveal the persistent existence of a secondary zone of uplift due to great earthquakes at the megathrust interface. This uplift varies in magnitude and location, but consistently occurs at a few hundred kilometers landward from the trench and is likely mainly coseismic in nature. This secondary zone of uplift is systematically predicted by our 2D visco-elasto-plastic seismothermo-mechanical numerical simulations, which model both geodynamic and seismic cycle timescales. Through testing hypotheses in both simple and realistic setups, we propose that a superposition of two physical mechanisms could be responsible for this phenomenon. First, a wavelength is introduced through elastic buckling of a visco-elastically layered fore-arc that is horizontally compressed in the interseismic period. The consequent secondary zone of interseismic subsidence is elastically rebound during the earthquake into a secondary zone of relative uplift. Second, absolute and broader uplift is ensured through a mass conservationdriven return flow following accelerated slab penetration due to the megathrust earthquake. The dip and width of the seismogenic zone and resulting (deep) coseismic slip seem to have the largest affect on location and amplitude of the secondary zone of uplift. These results imply that subduction and mantle flow do not occur at constant rates, but are rather modulated by earthquakes. This suggests a link between deep mantle and shallow surface displacements even at time scales of minutes.