COBOL code that has been working since the 60s is a lot more safe than new code would be. If it is commission critical and complex, keep it in COBOL where it works and pay lots of cash for additional features.

Rather then migrate it over in hopes you don’t have to pay for training, but then deal with tons of risk of systems breaking that haven’t broken in 50+ years.

Keep in mind the sunk cost fallacy and that every million dollars invested in 1960s money is about ten million dollars in today's money. Some of these projects are so foundationally sunk cost to these companies that they might as well be load bearing sunk costs. The company doesn't think that it exists without those sunk costs, and the company doesn't think it could survive a second, similar spending program today, at least not without their shareholders pitching a fit in potentially lots of short-term quarters given current shareholder obsessively short-term thinking.