IBM drops 13% after Anthropic promotes AI coding tool

IBM shares fell as much as 13% on February 23 following comments from AI startup Anthropic about its Claude Code tool, which aims to assist in modernizing COBOL, a legacy programming language primarily used on IBM mainframes.

The decline marks IBM’s largest intraday drop since March 2020, and contributes to a 26% decrease in its stock price this February, the steepest monthly decline since at least 1968.

Anthropic said in a blog post that tools like Claude Code can automate parts of COBOL modernization, which traditionally required extensive manual effort.

Most mainframe computers running COBOL are produced by IBM, and the selloff reflects broader investor concerns about AI’s potential impact on legacy companies’ growth prospects.

The company’s mainframe business remains significant, serving clients in finance and government sectors that rely on COBOL-based applications.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

IBM’s mainframe business looks healthier than the stock move suggests

Investors sold on worries about AI-assisted COBOL modernization, yet IBM’s Infrastructure segment, which includes mainframes, continues to post strong results.
Infrastructure revenue rose 10% in 2025, helped by the Z17 mainframe reaching the highest annual IBM Z revenue in about twenty years 1.
AI tools for COBOL updates are not new. IBM launched a 2023 product that uses AI to work with COBOL and convert code into Java (a widely used programming language) 2.
Banks and government agencies have had paths off mainframes for years, yet many keep them. That choice suggests the platform matters beyond the language running on it 2.

The selloff fits a growing pattern of AI worry in markets

The drop in IBM shares lines up with a recent habit of penalizing established firms right after AI announcements.
Days earlier, cybersecurity names slid after an Anthropic feature release, which fed a sell first, ask questions later mood 3.
Some investors fear generative AI will let teams route around legacy software through “vibe code” (using natural-language prompts to generate working software) 2.
That dynamic leaves many large tech vendors exposed to story-driven selloffs, even before revenue or customer behavior changes.