A significant portion of daily financial transactions, from paycheck deposits to ATM withdrawals, rely on the COBOL programming language, a system largely invisible to end-users. This week, that decades-old infrastructure became the center of a major market shift impacting IBM.
On Monday, February 23, AI company Anthropic published a blog post detailing how its Claude Code tool can automate the analysis and documentation work historically required to modernize COBOL systems. In response, IBM’s stock price fell 13.2% to $223.35, its worst single-day decline since October 2000. The stock is down 27% for February, on track for its largest monthly drop since at least 1968.
The selloff reflects a fundamental concern: if generative AI can significantly reduce the cost and time needed to understand and rewrite legacy code, it could undermine a core source of customer lock-in within enterprise IT.
COBOL remains foundational to critical infrastructure, supporting an estimated 95% of U.S. ATM transactions and running hundreds of billions of lines of code in finance, aviation, and government. Migrating these systems is notoriously complex, requiring the reverse-engineering of undocumented business logic and adaptation of tightly coupled data structures. IBM has built a high-margin consulting and mainframe services business around managing this complexity, with modernization projects often spanning years and costing tens of millions of dollars.
This ecosystem is reinforced by a severe talent shortage. Original developers have largely retired, institutional knowledge has been lost, and COBOL is rarely taught academically. This scarcity has cemented IBM’s role as the primary steward of these systems.
The market reaction is particularly striking given IBM’s recent strong performance. The company reported Q4 2025 revenue of $19.7 billion, with its mainframe Z Systems line growing 67% year-over-year and a generative AI business portfolio exceeding $12.5 billion. The selloff was not triggered by poor financial results but by uncertainty over their future sustainability.
IBM issued a pointed response, stating, “New AI tools emerge every week, including our own. What they do not change is the fundamental engineering challenge of running mission-critical workloads at scale.” The company highlighted its own watsonx Code Assistant for Z, which translates COBOL to Java, and its internal Project Bob initiative, claiming a 45% improvement in developer productivity.
Industry observers note that code translation is only one phase of a complex migration, which also involves replacing middleware, data formats, and compliance architectures under rigorous regulatory scrutiny.
The decisive question is not technical feasibility but trust. Financial institutions and their regulators must determine whether to trust probabilistic AI models to refactor systems handling trillions of dollars. Core banking modernization is governed by strict protocols requiring every change to be traceable, testable, and explainable. While AI may accelerate analysis, human oversight is likely to remain central to final decisions.
Nevertheless, the potential for AI to lower the analytical and cost barriers to migration could accelerate a long-delayed shift toward cloud-native architectures. This would benefit hyperscale cloud providers and cloud-native banking platforms while challenging vendors whose business models depend on the persistence of mainframe environments.
