Landslide time-of-failure prediction is crucial in natural hazards, often requiring precise measurements from in situ instruments. This instrumentation is not always possible, and remote-sensing techniques have been questioned for detecting precursors and predicting landslides. Here, based on high frequency acquisitions of the PlanetScope satellite constellation, we study the kinematics of a large landslide located in Peru that failed in June 2020. We show that the landslide underwent a progressive acceleration in the 3 months before its failure, reaching at most 8 m of total displacement. The high frequency of satellite revisit allows us to apply the popular Fukuzono method for landslide time-of-failure prediction, with sufficient confidence for faster moving areas of the landslide. These results open new opportunities for landslide precursors detection from space, but also show the probable seldom applicability of the optical satellites for landslide time-of-failure prediction.