In a context of climate change, assessing trends in hazards related to extreme events is urgent. Specifically, current methods to compute European standards for snow load actions on structures do not account for the non-stationarity due to climate change. We present the first analysis of extreme ground snow load trends for the whole French Alps. Our method is based on non-stationary generalized extreme value (GEV) distribution, time derivative of return level and likelihood ratio test. Thanks to Météo France reanalysis and snowpack models, we study mountain massif scale data available every 300m of altitude from 1958 to 2017. We detect an overall decreasing trend for annual maxima of ground snow load between 900m and 2700m, which is significant in the Northwest of the French Alps until 1800m. Despite decreasing return levels, in 2017 half of massifs at altitude 1800m still exceeds standard return levels. We underline the importance of snowpack modelling and limitations of approaches relying on ground snow load computed with snow depth annual maxima and a hypothesis on snow density.