IBM's $7 Billion Legacy Empire Jolted by AI Upstart

On February 23, 2025, a single announcement sent IBM shares tumbling. Anthropic, the AI research company, revealed its Claude Code tool could now translate ancient COBOL programs into modern languages like Java and Python. For IBM, this struck at the heart of a lucrative, decades-old franchise: the costly, complex work of modernizing the legacy systems that still power global finance and government.

The market reaction was swift and severe. IBM’s stock dropped as much as 13% in early trading, erasing about $25 billion in value. Competitors like Accenture and Infosys also slid. The reason is a colossal problem few have solved: an estimated $3 trillion in daily commerce still flows through systems written in COBOL, a language from 1959. Replacing it has been a notorious, high-risk endeavor, creating a captive market for IBM’s mainframes and consulting services, a business worth an estimated $5 to $7 billion annually.

Anthropic’s claim is that AI can dramatically compress that timeline. The company demonstrated Claude Code converting a 500,000-line COBOL banking application in under 72 hours, with human review taking two more weeks. It positions the tool not as a full replacement for engineers, but as a powerful accelerant.

IBM responded by emphasizing that syntax translation is the easy part. “The hard part is understanding the decades of accumulated business logic, the undocumented edge cases,” a company spokesperson said, pointing to IBM’s own Watsonx tool and its deep integration with mainframe environments. Industry veterans echo this skepticism, noting a long history of failed automated migration promises where the final 20% of complex, undocumented logic derails projects.

Regardless of the technical debate, the market’s sharp reaction reveals a new anxiety. If AI tools can reliably shrink modernization from years to months, the economic model for a significant portion of the trillion-dollar IT services industry faces disruption. For the first time, the formidable moat protecting IBM’s legacy empire looks vulnerable, not to a human competitor, but to a machine that can read a half-century of code.