Ask a room of leaders what "mainframe modernization" means, and most will describe some version of the same project: a multi-year effort to move off mainframes and onto something newer. It's the default script.
For the majority of institutions still running mainframes, that's not how it plays out. Mainframes aren’t so easy to escape, and the more likely path is modernizing in place. That path isn't the consolation prize it's often treated as. It's where most of the market is heading.
The exit narrative is more bark than bite
While the "rip them out" statement can make headlines, it isn’t quite the reality. Gartner expects only 10% of mainframe users to want out by 2030, and 75% of the vendors selling "mainframe exit" services to pivot their business models or disappear by then as the market resets its expectations. Most organizations running mainframes already operate a hybrid estate, with mainframes and the cloud side by side, each doing the work it does best.
The exit projects that do get started tell the rest of the story. As I've written before, Gartner predicts more than 70% of mainframe exit projects initiated in 2026 will fail to produce their intended benefits, largely because teams overestimate what generative AI can do with complex legacy code. When the destination is "no more mainframe," the odds are against you.
The company that builds mainframes is investing in its future
Instead of ‘winding things down’, IBM is doing the opposite. Its newest generation, the IBM z17, shipped in 2025 on the Telum II processor, which has an AI accelerator built directly onto the chip. Later that year IBM added the Spyre Accelerator, a card designed to run generative and agentic AI on mainframes, right where the data lives. Mainframes are being engineered for the AI era, not retired ahead of it.
This matters for a CTO or CIO’s strategy. If your mainframes stay, the question becomes how well the data on them can participate in everything happening around them.
What mainframe modernization actually means in 2026
For most, mainframe modernization is the work of making a mainframe's data and business logic usable by modern systems (cloud platforms, analytics, and AI) whether or not mainframes go away. A mainframe migration strategy may be less about subtraction and more about modern integration.
That's a different project, with a different risk profile, than a full exit:
| Exit (rip and replace) | Modernizing in place |
|---|---|
| Goal: remove mainframes | Goal: connect mainframes to modern systems |
| Business logic is rewritten on a new platform, or lost along the way | Business logic is extracted and documented where it already runs |
| One high-stakes, multi-year cutover | Incremental: value lands as each data domain opens up |
| Done when mainframes are switched off | Done when mainframe data is trusted, explainable, and usable everywhere |
The hard part is the logic
Most mainframe-to-cloud migration plans get scoped as a move: copy the data out, stand it up somewhere modern, done. But the data is only half of what lives on mainframes. The other half is the logic: the rules, calculations, and exceptions encoded across decades of COBOL, PL/1, and RPG that determine what the data actually means.
That logic rarely sits on one platform, either. "Mainframe" often gets used as a catch-all for both z/OS and IBMi, the descendant of the AS/400 that runs its own enormous installed base on RPG. They're different architectures with different languages, and both are still actively invested in. A modernization plan has to account for the logic on each.
Exit patterns also differ by platform. IBMi, still widely called AS/400, often runs core business applications on a midrange server, and organizations replace it with packaged software more frequently than they replace z/OS mainframes. Whether a team keeps the platform or leaves it, the RPG logic inside still has to be understood before the data can be trusted elsewhere.
Recovering that logic by hand is slow, whichever direction you're heading. In one Fortune 100 institution's mainframe modernization program, six people spent nine months decoding just nine COBOL modules, out of an estate of nearly 100,000. But if you simply move the records without the logic, then you’ve only created a copy of data nobody can fully explain. That’s why the most valuable lineage goes beneath where data moved, to the logic that produced it.
Access is a trust problem, not a plumbing problem
Explainability is a prerequisite for AI-ready data in a regulated environment.
The gap between data and its logic stops being an inconvenience the moment AI becomes the consumer. A model run on mainframe data it can't account for doesn't fail loudly – rather, it produces “answers” that are confidently wrong: fluent, plausible, and untraceable. In a regulated environment, that's the most dangerous output there is.
So making mainframe data accessible comes down to two questions. Can the data be trusted once it lands somewhere new? And can anyone explain how it got its meaning? Zengines answers them with two products and one outcome. The Turnkey Data Migration Platform validates and reconciles data as it moves, and its Logic Discovery feature flags the business logic and exceptions along the way, so the data lands trusted. Contextual Data Lineage reads the calculations, rules, and exceptions inside COBOL, PL/1, and RPG and explains them in plain English, so the data is explainable. Trusted plus explainable is what AI-ready means for a mainframe estate.
Hybrid is the destination
We've described Zengines before as the bridge between managing the mainframe today and modernizing tomorrow. The evidence now points one step further: for most institutions, the bridge isn't temporary.
That changes how you plan. If hybrid is a phase, trusted data is a one-time migration deliverable. If it's the destination, trusted and explainable mainframe data becomes a standing capability that every new cloud platform, data product, and AI agent can draw on. The competitive advantage goes to organizations whose mainframe data is already trusted, explainable, and feeding every new cloud platform, data product, and AI agent, while the system of record stays where it performs best.
Modernization doesn't mean exit. It means access.
Make your mainframe data AI-ready
If your mainframes are staying - either for a little while or a long while – then the priority is how quickly and safely the data inside them can reach everything else. See how the Turnkey Data Migration Platform and Contextual Data Lineage make mainframe data trusted, explainable, and AI-ready. Book a demo.
Frequently asked questions
What is mainframe modernization?
Mainframe modernization is the work of making a mainframe's data and business logic usable by modern systems, such as cloud platforms, analytics, and AI, whether or not mainframes are retired. For most organizations today, it means connecting mainframes to everything around them rather than fully exiting them.
Does mainframe modernization mean migrating off mainframes?
Usually not. Gartner expects only 10% of mainframe users to want out by 2030, and more than 70% of 2026 mainframe-exit projects to fall short of their goals. Most institutions modernize in place, keeping mainframes as part of their future.
Are mainframes becoming obsolete?
No. IBM's z17 (2025) has AI acceleration built into its processor, and the Spyre Accelerator runs generative and agentic AI on the platform itself. IBM is building mainframes for AI workloads, not phasing them out.