Finding better ways to handle software complexity (both inherent and accidental) is the holy grail for a significant part of the software engineering community, and especially for the Model Driven Engineering (MDE) one. To that purpose, plenty of techniques have been proposed, leading to a succession of trends in model based software developments paradigms in the last decades. While these trends seem to pop out from nowhere, we claim in this article that most of them actually stem from trying to get a better grasp on the variability of software. We revisit the history of MDE trying to identify the main aspect of variability they wanted to address when they were introduced. We conclude on what are the variability challenges of our time, including variability of data leading to machine learning of models.