An honest answer from someone who has been in the industry since 1990 and still builds mainframe software today.
This question comes up constantly, in forums, in university career offices, in LinkedIn comments. A computer science student gets an offer involving COBOL or z/OS and immediately wonders: am I about to bet my early career years on a technology that's about to disappear?
I've been working in mainframe since 1990 and running a mainframe software company since 2004. Here is my honest answer across the four questions I hear most often.
Don't choose, do both, but in the right order. During your internship or first role, give mainframe everything. Learn it deeply: COBOL, JCL, CICS, DB2, VSAM. Don't split your focus while you're trying to build real competence.
In your own time, keep your Java skills warm. The most valuable mainframe professionals in 2026 are the ones who can bridge both worlds: someone who understands COBOL and CICS but can also write a REST API or work with modern CI/CD pipelines is extremely rare and extremely employable. You are not choosing one or the other for life.
The niche concern is real, and it works in your favour. There are far fewer mainframe developers than there are mainframe systems that need maintaining. The supply-demand gap is only getting worse as experienced professionals retire. Companies pay a premium for people who can actually do the work.
If you are in India specifically: Tata, Wipro, Infosys, HCL, and Cognizant all run large mainframe practices servicing US and European banks. Your skills will be relevant across a large employer base, not just one niche company. This is not a path that locks you into a single employer.
Mainframe careers compress the experience curve. In Java, you are competing with millions of developers globally. In mainframe, by year three or four you are already in the top 20% of available talent simply because so few people have done the work long enough to develop genuine expertise.
Senior mainframe architects at banks and financial institutions with eight to ten years of experience earn well above average for the industry in most markets.
The ceiling is real too. You will not become a startup founder on COBOL skills alone, and consumer tech companies have no use for this experience. But if your goal is stable, well-compensated employment with genuine job security, this path is hard to beat.
Take the offer. Work hard, learn everything: COBOL, JCL, CICS, DB2, VSAM, the whole stack. Don't treat it as a fallback. Treat it as a specialism that very few of your peers will ever have.
The insecurity about friends having full-time offers is normal but misleading. A converting internship at a US MNC in a field with genuine scarcity is a better starting position than most of those offers, you just can't see it yet because the path is less familiar.
Mainframe isn't exciting to talk about at tech meetups. But the people who know it well are paid well and have job security that most developers would envy.
The rise of AI code generation tools is actually increasing demand for mainframe professionals, not decreasing it. When developers use AI assistants to generate COBOL for the first time, someone still needs to submit the job, read the output, diagnose the abend, and iterate. That person needs mainframe skills.
AI writes the code. Mainframe professionals are the ones who make it run.
Also worth reading: Starting Mainframe Work at a Bank: First 90 Days · The Hidden Risk in Every COBOL Migration Project · Is Mainframe Work at Risk of Being Outsourced? · The Quest for the Lost Holy Mainframe Treasure (text adventure)