Mainframe Redux: How AI Is Breathing New Life into Big Iron

The death of the mainframe has been talked about on and off for at least 30 years. But although such doom-mongering had subsided somewhat over recent times, IBM, the only surviving mainframe manufacturer, has lately found it being ignited once again. Second quarter revenues, which were announced in July, plummeted 42% year-on-year at its mainframe division. This contributed to the company's turnover coming in at $17.2 billion, missing Wall Street targets by around $700 million. Chief Executive Arvind Krishna attributed the fall to "low tens" of customers delaying their mainframe purchasing decisions due to anxiety over soaring demand for AI infrastructure, particularly among the hyperscalers. This led to them raiding their mainframe budgets to stockpile servers, storage and memory instead. To make matters worse, only months before the markets had shaved more than $30 billion off IBM's value over analyst concerns that its mainframe business would be disrupted by AI in other ways. Despite making no new technology announcements, Anthropic had claimed in a blog post that products, such as its Claude Code tool, could speed up the re-writing of Cobol into non-mainframe software languages from years to quarters. This is despite the fact, as Michael Stricklen, Managing Director of Professional Services firm EY's Software Strategy Group, told The Stack: > Translating code is not the same as migrating off a mainframe. Anyone who has actually done this work knows that the code is the straightforward part. The data layer, the middleware, the operational dependencies, the disaster recovery frameworks...that is where the real complexity lies. Indeed, IBM managed services spin-off Kyndryl's 2025 State of the Mainframe Modernization Survey Report, based on a global survey of 500 senior leaders, indicated that few of them expect to throw out their big iron any time soon. The strategic importance of such machines may have declined by 11% year-on-year as "hybrid environments" become more common. The re-vitalization of the mainframe But a significant 56% of customers have actually increased and diversified their mainframe usage. This is because they are now positioning it as an "enterprise transaction engine" rather than an "enterprise platform", says Andy Hughes. He is Senior Vice President for Go-to-Market at Kyndryl UK and Ireland. As a result, the number of workloads moving off such machines has dropped by eight percent as organizations discover "new, high-value roles' for them. Hughes explains: > Five years ago, everyone was claiming the mainframe was dead and everything would move to hyperscalers and the cloud. But over the last couple of years, we've seen a marked change in views towards the mainframe and its relationship with new AI platforms. IBM's z17 mainframe refresh [which supports AI] was one of its strongest launches ever. We still hear some people say the mainframe is unaffordable, difficult to develop on, and that they'll move off going forward due to Cobol. But many companies are saying they'll continue to keep their mainframes, so we believe they'll be here for the foreseeable future. An important point here for organizations requiring high volume, secure, and reliable transaction processing (more than 70% of Fortune 500 companies, according to Forbes) is that mainframes are "probably the most effective data processor on the market", Hughes says. In fact, he claims that, in this context, there's "no better or cheaper place in terms of cost per transaction". But the downside of such machines is that their software is often ageing, complex 'spaghetti code' that lacks adequate documentation and is tricky to alter or migrate due to often unclear dependencies. This has frequently led to an 'if it's not broken, don't touch it' approach. As Hughes points out: > CIOs are facing two issues. Everyone wants them to spend their budget on AI and the mainframe is a large cost item in the portfolio. So, they've been struggling, saying 'we've got this critical platform, but we can't access the data effectively from an AI point of view and so we can't afford it'. Which drives a discussion on its role and how to address the challenge of ensuring it's an effective platform. The move to mainframe modernization The upshot is that much of the conversation has now refocused onto mainframe modernization. This includes working out which workloads should stay on it and what can safely be moved elsewhere. It also involves understanding how mainframe data can be exposed to AI to underpin digital transformation. Hughes explains: > A few years ago, it was about moving off the mainframe and rewriting applications to run on the cloud. This involved skilled professionals knowing what the applications were doing, and there were some successes. But many applications couldn't be moved as even the Baby Boomers don't fully understand the complexity of the code due to a lack of documentation and stuff added on...But since the accessibility of AI models increased and IBM enabled them to run on the mainframe about two years ago, everything's started to accelerate, and it's now become not modernize off. It's modernize around. Brian Klingbeil, Chief Strategy Officer at technology advisory and managed services provider Ensono, agrees: > All the easy stuff has now been moved off the mainframe and it's the hard stuff that remains. Some things can't be done outside of that environment or at least, the mainframe's better suited to it. So, there's been an overall slowdown in migration. As for where the mainframe's strength lies, Klingbeil believes: > They're incredible systems of record rather than engagement. So, for example, you'd use them to run the flight scheduling system for an airline, but not the app that passengers would engage with to change their flights etc. Engagement lives in the cloud, but you can now tap into and analyze loads of mainframe data that it wasn't possible to access before. And as Microsoft's Chief Commercial Officer Judson Althoff once said, without good quality data, all you get from AI is the 'ability to make mistakes with greater confidence than ever before'. In other words, much of the corporate focus is now on building cloud-based data analysis systems that "feed off" mainframe data via application programming interfaces, explains Klingbeil. Another option if real-time processing is required is to take advantage of AI inferencing inline in the mainframe itself to guard against banking or other financial fraud, for instance. Unsurprisingly then, the Kyndryl report says: > Technologies, such as agentic AI, Large Language Models, and DevSecOps, are driving renewed interest in mainframe capabilities, positioning the platform as a key component within increasingly dynamic and integrated hybrid environments...In a hybrid world, the evolving capabilities of the mainframe allow it to remain central to IT strategies. The role of AI in the mainframe's renaissance As Klingbeil points out: > We started seeing the shift about 18 months ago. We'd believed that AI could help get people off the mainframe, but it actually gives them an incentive to stay. Some stuff should still be moved but it's a minority. Most things can stay and be augmented dramatically. As a result, he believes: > The mainframe is being re-vitalized because of hybridization. It's making it a more equal player, enabling it to act like the rest of the estate. That's where I see the revolution happening. So, this will extend its life indefinitely as it makes so much more sense in this context. If you're a CIO, your choice is to spend five years and $100 million trying to move everything off a box that may not be movable, you don't know if it'll work, or if there'll be dead bodies as a result. Or you can keep it where it is. So, why risk it when for $5 million, you could get 85% of the benefit by staying put? The choice feels easy. Meredith Stowell, Vice President of Ecosystem at IBM, is, unsurprisingly, equally