The Adaptive Join operator was added in SQL Server 2017 as an alternative to the other join operators: Nested Loops (ideal for joining a small data stream with a cheap input), Hash Match (most effective for joining large unsorted sets) and Merge Join (ideal for joining data streams that are sorted by the join key). It is intended to be used when there is no efficient way to fulfill the order requirement of the Merge Join, and the optimizer cannot reliably predict which of the remaining algorithms (Hash Match or Nested Loops) would perform best.
Because it has to be able to join the data using either the Nested Loops or the Hash Match algorithm, Adaptive Join suffers from the combined restrictions of these operators. As such, Adaptive Join supports only four logical join operations: inner join, left outer join (but not the probed version), left semi join, and left anti semi join; it requires at least one equality-based join predicate, it uses lots of memory, and it is semi-blocking.
Read on for a detailed look at this operator.