No, ML works fine with large output spaces too.

The problem is that the solution space grows super exponentially, and if you need to find an exact one, then the number of samples too.