The purpose of this paper is to characterize the statistical distribution of worldwide largest earthquakes. We analyze the distribution of worldwide shallow events with M s 7.0+ since 1900, by following the Occam's razor philosophy; we start from the simplest possible model (stationary Poisson process), and we inspect more complicated models only if the data show significant departures from the simplest one. The results show two important characteristics; first, worldwide M s 7.0+ earthquakes tend to cluster in time and space, with features similar to smaller events. Second, for some seismic regions there is evidence in favor of long-term fluctuations of the earthquake rate. These results support the hypothesis of universality, suggesting that an ETAS model with a background varying with time can be considered reliable to describe the seismicity distribution over a wide space-time-magnitude window. From a practical point of view, our findings suggest that the paradigm that seismic zones are stationary systems, implicitly assumed in seismic hazard assessment, should be regarded with caution.