COBOL is unbelievably ingrained into the fiber of banking, and the developers surrounding it are seriously seasoned veterans in the realm of batch code, deployment and support.
Not only did I go to a university in the early 2000's that actually taught COBOL, JCL, VSAM generation and use, emulation mainframe 'green screen' interfaces (the actual language we used escapes me now 16 years later - ACS? ASC?), AS/400 exposure, even fucking assembler (and not for computer science C/C++ brats dumping out compiler interpretation) for one reason only: Citibank, Federated Insurance, Wells Fargo, FIS, Bank of America all were extremely heavy players in that degree path at that university. Hell, our professors who taught those banks of classes were soon-to-be-retired Citibank senior developer vets teaching us their standards, techniques, and tons of this-is-the-difference-between-academic-code-and-real-world-code lessons. So as much as everyone makes this baseless argument about how it's dying --- even back in early 2000, after the Y2K scare, it seems like the big banking brains were setting themselves up for long-term rollover of fresh meat to take on the mainframes.
I actually went on to work at Citibank for a few years, but I worked on the front-end and middle-ware vs. the back-end mainframe, even with all my newly fresh COBOL skills. All I can tell you is: that shit isn't going anywhere. I know plenty of people who still hack COBOL for a living. And as long as banks still push the agenda at universities and kids who are intimidated by computer science take these courses, not only will it still be taught to a fresh crop of students, it means that bankers know money and also know how much fucking money they'd lose migrating away from it in any sort of planned manner.
Has IBM stopped making AS/400 iterations past it like the System I and such? Hell no. All the answers are there. COBOL is here to stay.