The term Voxel is very broad. Due to the popularity of Minecraft, many people think a voxel game can only consist of blocks (like Minecraft, Cube World and Seven Days to Die). However, this isn't true.
Voxel worlds usually use a three-dimensional data set to represent the terrain. In cases like Minecraft, the data set contains what block there is at certain places. In some other games, however, it tells you how much material there is at that specific point. The latter version usually uses some sort of iso-surface algorithm (like marching cubes or dual contouring) to display the terrain. Many games like Space Engineers (which is a mix of the two), the recent Astroneer or Mediavel Engineers are in this category.
It's not necessary to use perfect squares if you implement Voxel terrain the "Minecraft way", you can use irregular shapes, slopes or block sized models for it too. It has nothing to do with the complexity of the graphics.
Implementing a game like this can be achieved in multiple ways. One method of optimization is to separate the terrain into chunks, then only render the visible ones and the ones that are close to the player.