I'm not enough into that industry to actually give a good estimate, here, but the amount of COBOL systems still up and running is certainly not even close to non-zero, and it's going to stay that way for a while. From what I gather for companies moving away from COBOL is more of a "programmers are hard to find" situation, not "these systems absolutely must be replaced" one. It's well-supported and scaled with their business, as in, in places they're running the same 60 year old code on new mainframes because if there's one thing that IBM mainframes are then it's excessively backwards-compatible.

As far as the language is concerned: It's not hard, it's just weird, dating back from an age where people thought randomly calling things "divisions" would make businesspeople capable coders. The reason I'm not in that space isn't because of the language but because of the type of software you write there, it's all bookkeeping and representing business procedures, as said: Bureaucracy.

Also I'm not sure what "modernising" actually meant, there: SEPA instant payment was introduced, meaning that mainframes won't batch up the day's transactions and then talk to each other every night so cross-bank transfers took a day to process, now they're doing it in ten seconds. Most banks already supported instant transfers within their own systems so they should only have had to rewrite the external interface as the rest was already up to the task.