Claude Code shock: Big Blue falls, Indian IT stocks follow

FILE PHOTO: Figurines with computers and smartphones are seen in front of the words "Artificial Intelligence AI" in this illustration taken, February 19, 2024. REUTERS/Dado Ruvic/Illustration/File Photo | Photo Credit: Dado Ruvic

Shares of IBM tumbled over 13 per cent on Monday, wiping out more than $30 billion in market capitalisation and dragging Indian IT stocks lower, after a blog post by Anthropic stoked investor concerns that AI-led tools like Claude Code could accelerate COBOL modernisation.

The decline weighed on Indian IT stocks, with the Nifty IT index dropping nearly 5 per cent on Tuesday. LTIMindtree fell 6.43 per cent, Tech Mahindra declined 6.17 per cent, HCLTech slipped 5.83 per cent, while Infosys and TCS were down 3.56 per cent each.

Anthropic effect

Analysts attributed the stock sell-off to a blog post by Anthropic, which highlighted the fragility of legacy COBOL systems. Common Business-Oriented Language (COBOL) is a high-level, English-like, compiled programming language developed for business data processing needs.

Anthropic noted that hundreds of billions of lines of COBOL still power critical infrastructure even as the pool of engineers who understand the language is shrinking. Much of this code was built decades ago and is poorly documented.

Traditionally, COBOL modernisation required large consulting teams, multi-year timelines, and high costs. Anthropic argued that AI changes that equation; tools like Claude Code can compress modernisation cycles from years to quarters.

Responding to a mail sent by businessline, IBM stated, “IBM has been investing in code modernisation for years – both through skilling initiatives and through our own GenAI capabilities. Over two years ago, we launched watsonx Code Assistant for Z (IBM’s mainframe) because we understand the benefit of AI in modernising code. 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. Translating COBOL is the easy part. The real work is data architecture redesign, runtime replacement, transaction processing integrity, and hardware-accelerated performance built over decades of tight software and hardware coupling. That is the problem IBM has spent decades learning to solve, and AI is the most powerful tool we have ever had to do it.”

Pareekh Jain, founder and CEO of EIIR Trend, said the sell-off may be an overreaction, but the underlying risk is real, noting that the prospect of work being completed in as little as two weeks threatens traditional effort-based billing models.

AI-native firms like Anthropic are moving faster because they have no legacy revenue to protect. In IBM’s case, COBOL and mainframes have been core for decades, and its differentiation could erode if tools like Claude Code enable other service providers, including mid-tier firms, to accelerate modernization. While service providers will still be needed, if AI handles 80 per ent of the work, their role shrinks to the remaining 10-20 per cent, making it more commoditised.

“With Claude Code, there are two views — the agent may not work effectively because there isn’t enough publicly available training data, and speculation around whether proprietary IBM data may have been accessed to train such models,” he said.

While agents and service providers will coexist, the broader concern is cannibalisation, Jain added. Legacy spend may decline faster than new AI-led spend emerges. While AI is creating new opportunities, the pace of new investments depends not just on technology but also on macro conditions. “If the speed of decline is faster than the speed of new spending, investors worry,” he said.

Automation

Aravind Putrevu, VP of Growth at Coderabbit, explained that Anthropic is explicitly saying Claude Code automates “exploration and analysis” and can get one to a modernizable plan faster, even in quarters instead of years.” This is believable because most COBOL programs are not hard because of syntax, but because the business logic is fossilised and the dependencies are tribal. What’s less proven is end-to-end modernization “at scale” without humans owning test strategy, data migration, and production-grade validation. While AI makes the unknown faster, it does not magically make regulated production systems safe.

However, IBM’s position in mainframes is more defensible than the market reaction suggests, he said. “IBM’s moat is platform-level: tight hardware-software coupling, transaction throughput, IO subsystem optimization, accelerators, and decades of performance tuning. IBM is already framing this as ‘translation is not modernization,’ arguing that you can’t replicate the system-level properties by just moving code. Also, IBM’s current numbers show the mainframe cycle is not dead weight right now: Infrastructure was up, and IBM Z was sharply up year-over-year in Q4. COBOL is the headline, but reliability, security posture, and operational predictability are the real reasons these systems stay.”

Meanwhile, Naga Santhosh Josyula, Co-founder, Tablesprint, argued that while AI tools are increasingly automating work that used to take large teams of consultants months or years, consulting won’t disappear; rather, the nature of billable work shifts toward governance, strategy, and quality assurance, with fewer hours spent on repetitive technical tasks.

Published on February 24, 2026