Why the move away from ORMs? I've certainly banged my head against the wall with them any number of times, but I wouldn't want to give up the compile time checking of queries when I make a db change.
I use the Django ORM most of the time, and it is great for quick stuff.
Say I have to do a bulk update, with a few conditions thrown in. MySQL becomes too much hassle trying to parse the first digit out of a string ,and updating another field with that. A simple script using the Django ORM won't take long to do that and will be far easier than trying to work around MySQL's limited string functions.
On the other hand Django's ORM doesn't handle some pretty common cases (conditional aggregates are the first thing that comes to mind). Or complex joins on more than one field. I looks around, and there are ways of hacking these things into the ORM, but they really don't seem worth the bother, when you can use SQL for the query.
Knowing one makes it easier to learn the other, and there's value to understanding the underlying abstraction layer, but you'll avoid using it if you can.
In general, the move to SQL was initially painful, but is now a breeze.
With SQL, I don't have to learn an abstraction that is limited to a specific language. So whether I'm working in Java, Python or JavaScript/Node.js, the SQL is the same.
Also, there are numerous technologies that now provide an SQL-like language. For example, learning SQL made it easy to learn CQL (Cassandra), HiveQL, etc.
Or use homegrown database wrapper module(s) and keep compile time checking in the vast majority of the code.