[tmarman]

From my perspective, I think limiting QPM is better than limiting the number of documents that you can index. In my current case, I have millions of "documents" (in a Lucene sense of documents) but relatively low usage. Obviously, my goal long-term is to increase the usage. So being able to pay to index a lot of documents but limit the resources (i.e., VM size, but not storage size) to search would be the best way to scale.

In theory, the more users I have the more $ I have to scale the search.

The solution I'm currently using (mostly because SQL Azure doesn't support Full-Text Search) is Lucene.NET (which is still on a very old version but supposedly 4.8 is coming) and AzureDirectory (which leverages Blob storage). It's clunky, but it works... at least for the scale I use it at currently. I would love to be able to use Azure Search and scale it up again just like with all my other services.