For me there were two main reasons:
1 - the cost of getting a commercial Ada compiler, much more expensive than most alternatives
2 - initially many though that Ada compilers would be extremely costly to implement. Ironically, C++ compilers are way more complex to implement than Ada ones.
As somebody who works on an Ada compiler, I'm not sure about that.
On some levels, Ada has a better design which simplifies compiler implementation.
On some others, the RM spec is extremely complicated - in my opinion because it was specified before it was implemented, and because of the need for safety - and very hard to implement.
See for example, dynamic access check, pool specific controlled types, anything related to elaboration, build in place for limited types, etc ..
The context dependent grammar, lookup rules, ODR, SFINAE, all the semantic interactions of template meta-programming,...
Then again that is just my perception.
This story and hommage by dwheeler https://www.dwheeler.com/essays/make-it-simple-dewar.html was one of the nicest things written on the late Robert Dewar... and on RMS !
The GNAT Front End is a large and complex piece of software and the language is very powerful thanks to all the features packed in the compiler.
On the other hand, for compliance with the standard there is a complete test suite : ACATS... It should help...