Depending on the size of the project, the processing options, the available RAM and the geometry of the project, the point cloud can be generated in multiple clusters of images. When the point cloud is generated in clusters it takes longer than when there is only one cluster to be generated. This is because the number of clusters is representative of the size of the project, the number of points that will be generated (processing options) and the resources available.
Also, when densifying the point cloud, all the automatic tie points ( the result of step 1) are densified and then filtered to remove the noise. This means that if the results of step 1 are noisy, there will be many points generated that will then be discarded as noise. Is your project noisy?
Are the automatic tie points generated during step 1 noisy? Can you please try to mark some manual tie points around the model and check the reprojection on the images in the rayCloud?
If the project is noisy, we would recommend you to find the cause of the bad calibration (bad dataset, incorrect camera model, wrong processing options, etc) and try to improve it before running step 2. Depending on the cause of the bad calibration, you might need to correct the camera model, adjust the processing options or add some manual tie points and reoptimize the project. If you wish, you can send us the quality report of your project and we can further assist you.