I have seen very large jobs that needed to be split up across multiple regions but doing so required replicating large amounts of data. That comes at a cost too

I have seen very large jobs that needed to be split up across multiple regions but doing so required replicating large amounts of data. That comes at a cost too so it requires planning to work out which regions you want to use. Once that is done it is fairly easy to set up spot clusters in each region to process the jobs in the queue.