The exit narrative has been winning for thirty years. The mainframe has been running.
That came from a conversation with an AI and Data Governance advisor to Fortune CIOs. He was not being ironic.
He was describing something he has watched happen repeatedly: the exit narrative arrives with a vendor, the project starts with ambition, the complexity appears on schedule, and the mainframe keeps running while the migration quietly fails or quietly gets shelved.
For thirty years, enterprise IT has been told that the mainframe is a liability. Too expensive. Too specialized. Too dependent on skills that are retiring. The future is cloud, microservices, containers, whatever came out last Tuesday.
The vendors selling this narrative have been consistent. The results have been less so.
Gartner published its State of the IBM Mainframe report in 2026. The finding that made headlines: for some organizations running VMware workloads, moving to mainframe is cheaper than Broadcom’s new licensing. The analyst who wrote it said he was surprised.
He should not have been. The math has always been there. The narrative buried it.
The mainframe processes $10 trillion in transactions every day. 95% of the world’s top banks run z/OS. These are not organizations that missed the memo about cloud. They are organizations that ran the numbers and made a deliberate choice.
The exit conversation focuses on code. Can AI decode the COBOL? Can we map the dependencies? Can we understand the business rules?
These are real questions. AI has genuinely improved the answers. Code comprehension is faster, intent surfacing is better, dependency mapping is more complete than it was five years ago.
But the exit narrative misidentifies the problem.
The COBOL is not the mainframe. The COBOL is the business logic that has accumulated over forty years of running real operations for real companies with real consequences. Decoding it is step one. There are nine more steps.
The harder questions are the ones that rarely get asked before the contract is signed:
Can the target platform match mainframe throughput at scale? Can it provide five-nines availability without building HA from scratch? Can it process transactions with the integrity guarantees that financial regulators require? At what cost? With what risk?
These comparisons are rarely done rigorously. The vendor has a product to sell. The CIO has a board asking about modernization. The mainframe team is busy keeping payroll running.
The project gets greenlit on narrative, not analysis.
Here is what the exit conversation misses: the mainframe is exceptional infrastructure.
Throughput at scale. Transaction integrity. Five-nines availability. Cost-per-transaction that cloud architectures struggle to match at volume. Built-in high availability, disaster recovery, and data synchronization that cloud architects spend months designing and years debugging.
These are not legacy limitations. They are engineering achievements that took fifty years to build and are still unmatched for specific workloads.
The organizations that sleep peacefully are not the ones who refused to modernize. They are the ones who understood the difference between modernizing the platform and replacing it.
Modernizing in place: wrapping COBOL in APIs, improving observability, reducing knowledge dependency, adding AI-assisted operations. The business logic stays. The interface to it evolves.
Replacing: migrating workloads to a platform that was not designed for them, rebuilding reliability that was already there, hoping the target environment can match what you had.
The honest conversation with a CIO is not “how do we exit mainframe.” It is “why have we underinvested in it for twenty years while we waited for it to go away.”
Modern tooling for mainframe exists. AI-assisted diagnostics, automated security scanning, developer-friendly interfaces, observability layers that surface what the system has been telling you for thirty years. These are not theoretical. They are available now.
The investment gap is real. The skills gap is real. The institutional knowledge problem is real. These are solvable. They are not solved by migration: they follow you to the new platform, where you have the same knowledge problem plus the complexity of a system that was not designed for your workloads.
An AI and Data Governance advisor to Fortune CIOs put it simply: “People can sleep peacefully with mainframe.” It is a line worth sitting with. He was not offering reassurance. He was reporting an outcome he has watched repeat itself across organizations and decades.
He was not describing complacency. He was describing the outcome of good engineering decisions made decades ago and maintained with discipline since.
The vendors who sold the exit narrative moved on to the next contract. The mainframe team who kept payroll running at 3 AM is still there.
The platform that processes $10 trillion a day does not need a defense. It needs investment, modernization, and the honest conversation that the exit narrative has been preventing for thirty years.
Stop treating your cost solution like a cost problem.
People can sleep peacefully with mainframe. The ones who figured that out are not the ones who stayed because they were afraid to leave. They are the ones who ran the numbers, asked the hard questions, and made the right call.
The platform is sound. The investment gap is the real issue.
The cursor has been blinking since 1964. It will still be blinking when the next modernization wave arrives.
Also worth reading: So You Decided Mainframe Is Cheaper Than Broadcom. Now What? · Why AI Pilots Succeed and Productions Fail · Why Mainframe Is Different