What really gets me is that so many of the NoSQL databases now offer an SQL interface. I have seen folks go to great lengths to avoid using a relational database yet the end result could be easily served by one. One project I was involved with was on AWS with highly structured delimited files in S3 and used Glue and SparkSQL on top of that to enable them to "query" the data. The source data were mostly from other relational databases (SQL Server, Oracle and DB2) which they dumped into S3 with a file per table. They wrote Python scripts to ETL the data into other S3 files that were also highly structured and documented with Glue so SparkSQL could query them. They did everything to avoid using PostgreSQL and RedShift yet wound up creating a very complex and convoluted system that approximated a database. Their arguments for not using an RDBMS was that it needed to be "scalable and fast" yet was really neither.
Why does it surprise you that they "now" offer an SQL interface? - remember NOSQL is short for Not Only SQL - that should make it clear why they not only may but rather must offer an SQL interface.