Programmers are free to pick up the language, and there are plenty of resources out there to help anyone who wishes to learn COBOL to do so.
I could imagine a programmer boot camp that focused on COBOL, and it might be in the interest of the Big Banks to create just such an institution. However, the mission of a computer science program is quite different from the mission of a coding boot-camp. Computer Science programs teach the core concepts of the field of computer science. There are no important, underlying CS concepts that are best taught in COBOL, and this makes it a poor choice for most of what we do. While nods can be made to industry in language choice, computer science is what we are here to teach.
The reason that there are so many institutions that still rely on COBOL is because they have a tremendous amount invested in that architecture from decades ago when the language really was still in use.
However, there are also important reasons that the language is dying. For instance take a look at this quora post by the user "Jack Menendez":
The assumption Dead is a matter of definition. However, the reasons why new applications are never written in COBOL is:
COBOL has a special type called a Packed Decimal that is a fundamental type for a mainframe. COBOL code and databases used by COBOL are full of Packed Decimals. To a mainframe Packed Decimal works kind of like an integer on intel but it is a kind of decimal. So there are machine level instructions on the mainframe that can deal with the Packed Decimal add/subtract/Div etc. The intel architecture does not have any idea what Packed Decimals are so COBOL compilers that target Intel must create code that simulates them. COBOL runs much slower on the Intel architecture for this reason and can create binaries that are relatively large.
COBOL programs are non-reentrant by their nature. In fact the code and internal data fields on the mainframe normally occupy the same contiguous piece of storage. This makes it hard to use COBOL in environments such as application servers that expect reentrant code. Application servers ported to the mainframe typically cannot use COBOL.
COBOL code has a syntax that is not like other languages, writing "Hello World" in COBOL is kind of an ordeal for a beginner.
The increased salary for COBOL programmers is reasonable since there needs to be a way to reward programmers who take the time to learn languages that have little applicability elsewhere. It is also part of the expense for maintaining dated software on decaying architectures.
Every year, more and more of this COBOL code is being replaced as the maintenance costs go up and the ability to expand its functionality decreases. And this is for the good! While there will always be room for people who truly love the language to keep the flame alive, the large base of industry should be where people can be efficient, and that is no longer in COBOL.