AI threatens legacy software systems, IBM worst hit

Shares of IBM saw their sharpest single-day fall in more than 25 years after new fears emerged about how artificial intelligence could hurt one of the company’s most stable businesses.

The stock dropped 13.2 per cent to close at $223.35, marking its biggest one-day decline since October 18, 2000.

IBM’s shares are now down around 25 per cent so far this year as investors rethink how fast AI could change the economics of enterprise software and IT services.

The sell-off was triggered by a blog post from AI startup Anthropic, which claimed that its AI tool, Claude Code, can understand and modernise COBOL, a programming language created in the 1950s that still runs many of the world’s most important computer systems.

COBOL remains widely used in banks, airlines, insurance companies and government departments, and it plays a key role in IBM’s mainframe business.

For decades, updating COBOL systems has been slow, expensive and dependent on large teams of consultants.

This work has generated steady revenue for IBM, as many companies struggle to maintain or upgrade old systems that few engineers fully understand anymore.

Anthropic argues that AI changes this balance by making it much easier to analyse and update legacy code.

In its post, Anthropic said there are hundreds of billions of lines of COBOL still running in live systems every day, even though the number of people who know the language continues to decline.

The company added that AI is especially good at handling the complex and time-consuming tasks that once made modernising COBOL systems too costly.

Anthropic estimates that about 95 per cent of ATM transactions in the US still depend on COBOL, underlining how deeply the language is embedded in financial infrastructure.

The company said its AI can scan huge codebases, trace how different parts of the software depend on each other, create clear documentation for systems that are no longer well understood, and highlight potential risks that would normally take months to uncover manually.

“Modernisation has been stuck for years because understanding old code often costs more than rewriting it. AI flips that equation,” Anthropic said.