Gartner is throwing cold water on the idea that AI can help companies escape their mainframe addiction. More than 70% of mainframe exit projects started this year will fail to deliver expected benefits, the analyst firm predicts, because organizations are wildly overestimating what generative AI can actually do with legacy code conversion.
The report, authored by Gartner analysts Dennis Smith, Alessandro Galimberti, and Tobi Bet, pulls few punches. By 2030, they expect 75% of vendors selling AI-powered mainframe migration tools to either shift strategy or disappear entirely. The core problem is that mainframes house mission-critical applications and decades of interconnected data. AI is genuinely useful for finding and describing technical debt, the analysts note, citing how JUXT used Claude AI and Allium to find a 57-year-old bug in Apollo 11's Guidance Computer code. But automated code conversion hits a wall fast.
Decades of patches, workarounds, and unwritten business logic live in that COBOL code. The people who understood why certain decisions were made retired years ago. AI can translate syntax. It can't recover lost context or reconstruct institutional knowledge that was never documented. Converting lines of code is the easy part. Understanding what the code actually does for the business is where things fall apart.
This market got a jolt when Anthropic promoted Claude Code's COBOL-conversion capabilities, which spooked investors and sent IBM's stock sliding. Gartner's team argues that aggressive investor pressure is pushing vendors to sell AI solutions where they don't fit, such as building deterministic web integrations for AI agents. "The stakes of a miscalculation are immense," they wrote. "Poor decision making regarding migration is not merely a budgetary overage; it is a threat to business and operational continuity." Their advice is blunt: most companies should stop dreaming about a clean exit and focus on improving the mainframe systems they already have. The report ranks the mainframe as still the leading platform for certain mission-critical applications, even with the ongoing push toward cloud-native architectures.