i could have expressed myself better, you are right. of course you need the aerial images from different perspectives to generate 3D data from 2D data. the top-down view from satellite imagery only gives you ground textures and 2D positional informations.
i was trying to stress that these (google) 3D models were being generated by algorithms and not humans. granted, they look silly if you get really close, because the mesh algorithms produce holes and false angles, depending on the source imagery.