IBM shares fell 13% in their worst single-day drop in more than 25 years after Anthropic claimed its Claude Code tool can sharply accelerate COBOL modernisation. The selloff reflected investor concern that AI could erode a long-standing consulting revenue stream tied to maintaining and upgrading legacy systems.
Anthropic said its AI can automate the mapping, documentation, and risk analysis of massive COBOL codebases, compressing projects that once took years into quarters. The announcement struck at the heart of a niche that firms like IBM, Accenture, and Cognizant have monetized for decades. COBOL still powers critical infrastructure, handling an estimated 95% of ATM transactions in the U.S., and hundreds of billions of lines of code remain active across finance and government systems.
Can AI Upend The COBOL Consulting Model?
The labor bottleneck has long defined COBOL modernisation. As experienced developers retire, institutions have relied on consulting armies to untangle brittle legacy workflows. That scarcity supported premium pricing and multi-year engagements. Anthropic argues AI now shifts that balance by rapidly identifying dependencies and surfacing actionable insights. If modernization becomes faster and cheaper, does that compress consulting margins or expand total demand?
Dan Robinson
IBM maintains that the market reaction overlooks key distinctions. The company has already introduced its own AI-driven tooling, including watsonx Code Assistant for Z, aimed at helping enterprises understand and refactor COBOL systems. CEO Arvind Krishna said recently that adoption has been strong, with many clients using the tool to evaluate modernization pathways.
Senior Vice President Rob Thomas pushed back on the broader narrative: “Translating code is one thing. Modernising a platform is something else entirely. The two are not the same, and the gap between them is where most enterprises run into trouble.”
That distinction matters because much of IBM’s value proposition lies in its vertically integrated mainframe stack, including z/OS and transaction processing architecture. Notably, around 40% of COBOL runs on distributed systems rather than mainframes, complicating the assumption that AI-driven code tools directly threaten IBM Z.
Recent client examples show tangible efficiency gains, including a reported 94% reduction in COBOL code analysis time for one public-sector user. Investors will now watch whether AI-driven productivity gains translate into shrinking service revenues or unlock a new wave of modernization spending in the quarters ahead.
