After a great subduction earthquake, viscoelastic stress relaxation causes prolonged seaward motion of inland areas of the upper plate, as was observed around the turn of the century in the area of the 1960 Mw 9.5 Chile earthquake with Global Navigation Satellite System (GNSS) measurements. However, recent GNSS observations during 2010–2019 indicate a systematic decrease in the velocity of the seaward motion over a region covering the latitudinal range of the southern half of the 1960 rupture. Data from the only long-lived continuous site in this region (COYQ since 1997), situated over 200 km away from the trench, suggest that the decrease in the seaward velocity (or increase in the landward velocity) occurred within a few years prior to 2010. This rapid and regional change cannot be explained by viscoelastic relaxation. We thus propose that the change was caused by a relatively sudden downdip widening of the zone of locking along the megathrust. Using three-dimensional finite element modelling, we find that the observed velocity change cannot be otherwise explained, although the amount of the increase in locking cannot be uniquely determined because of trade-offs between, and uncertainties in, the various parameters involved. For example, the degree of the increase in locking is affected by the value of coseismic slip in 1960 in the southernmost part of the rupture zone. A postseismic deformation model with greater coseismic slip in accordance with the most recent coseismic slip model in the literature better fits COYQ data prior to 2005 and requires greater locking increase afterwards. A model with less coseismic slip requires less locking increase but an additional long-term slow slip event prior to 2005. The rapid surface velocity change and the inferred increase in megathrust locking several decades after a great earthquake present new challenges to the understanding of fault mechanics and subduction zone dynamics.