When the headlines began to circulate more forcefully, traders in lower Manhattan were already looking at red screens: In a blog post, Anthropic described how its artificial intelligence (AI) tool, Claude Code, could update COBOL systems more quickly and affordably than with conventional techniques. Within hours, IBM’s stock was plunging, closing at a loss of about 13%, the largest one-day decline in decades. Billions disappeared. And a blog was the catalyst for all of it, at least on the surface.
When artificial intelligence is brought up, it’s difficult to ignore how brittle investor confidence gets.
A large portion of the global financial system is still powered by COBOL, that obstinately robust programming language from the 1960s. withdrawals from ATMs. reservations for airlines. The government gains. IBM Z mainframes, big, humming devices housed in temperature-controlled rooms in banks and government organizations worldwide, power a large portion of that code. Although these systems lack the glitz and glamour of Silicon Valley startups, they are reliable.
| Category | Details |
|---|---|
| Company Name | International Business Machines |
| Founded | 1911 |
| Headquarters | Armonk, New York, USA |
| CEO | Arvind Krishna |
| Market Cap (2026) | ~$225 Billion |
| Core Platform | IBM Z Mainframes |
| AI Platform | watsonx |
| Rival Company | Anthropic |
| Anthropic Product | Claude Code |
| Official IBM Website | https://www.ibm.com |
| Official Anthropic Website | https://www.anthropic.com |
Anthropic’s assertion was simple. Claude Code was able to map dependencies, document complex workflows, analyze large COBOL codebases, and speed up modernization projects that had previously taken years. One gets the impression that the business wasn’t overstating its technical prowess. AI-assisted code analysis has significantly advanced. Workflows are increasingly being described by engineers in simple terms, and they are responding with working code. That feels authentic.
But in record time, Wall Street seemed to go from being a “helpful tool” to a “existential threat.”
The morning after the drop, the winter air outside IBM’s Armonk offices felt still. With coffee cups in hand and badges swiped, staff members talked about product roadmaps instead of stock charts. Teams were probably reviewing ongoing modernization projects inside conference rooms. IBM’s Watsonx Code Assistant has been providing AI-assisted COBOL migration tools for years. Within those walls, the notion that AI could aid in the reworking of legacy systems is not new.
Investors appear to think that abandoning IBM’s mainframes is inevitable if learning COBOL becomes inexpensive. It might be too neat of an assumption.
Language translation is rarely the only step in modernizing enterprise software. Syntax is addressed when translating COBOL into Java or another contemporary language. Data migration, performance assurance, regulatory compliance, and decades of embedded business logic are not all automatically resolved by it. Deeply integrated workflows that have been fine-tuned, optimized, and stress-tested over time are being executed by a mainframe that processes billions of encrypted transactions every day.
Whether AI can handle that full load is still up in the air.
The distinction between code analysis and infrastructure replacement is another deeper level of the IBM COBOL Anthropic controversy. Mainframes provide exceptional throughput, integrated security, and high uptime. IBM has highlighted that its most recent systems are capable of handling massive amounts of AI inference workloads and processing tens of billions of transactions daily. It’s ironic that the AI wave that seems to be threatening IBM may actually make the case for maintaining robust on-premise systems near mission-critical functions.
As we watch this play out, it seems like the market responded more to narrative than to subtleties.
Speed is embodied by Anthropic. This AI startup moves quickly, releasing updates and whitepapers at the speed of Silicon Valley. IBM is a symbol of continuity. It caters to organizations that measure risk differently than venture capitalists, such as banks, governments, and insurers. Usually, blog posts are not enough to influence these customers. Before making changes to infrastructure that handles cross-border transactions or national payrolls, they proceed cautiously, testing and validating.
IBM is not exempt from this. Engagements involving a lot of consulting may decrease if AI technologies shorten the early discovery stage of modernization from months to days. Systems integrators might experience pressure on their prices. There may be minor but significant changes in the economics of legacy transformation. In the long run, it might be more beneficial to lessen the friction around mainframes than to replace them.
This episode, in a sense, offers a more comprehensive insight into the psychology of the market today. AI is expected to overthrow all incumbents, according to investors. Occasionally, that assumption turns out to be accurate. In other cases, it goes beyond reality. That tension seemed to be reflected in the IBM stock chart, which plummeted on a Monday afternoon as engineers silently continued to provide shipping updates.
The unsettling reality is that even though AI-generated code is amazing, mistakes are still made. Research comparing code written by humans and machines frequently reveals that AI produces more errors, especially in complex systems. A bug is annoying for a consumer app. It can be disastrous for a bank that processes billions of transactions. Even if it doesn’t become popular on social media, that distinction is important.
Compared to the market’s initial response, the tone of conversations with analysts and enterprise architects seems more measured. Many contend that IBM’s tools and Claude Code could coexist and even enhance each other. A combined ecosystem, with AI models speeding up discovery and IBM infrastructure guaranteeing dependability, seems more realistic than a winner-take-all situation.
In the end, the IBM COBOL Anthropic conflict might be more about acceleration than displacement. AI is shortening timelines, reducing some expenses, and posing new queries regarding how businesses handle aging infrastructure. But it’s rarely as easy as rewriting code to destroy foundational infrastructure.
The 13% decision on Wall Street appeared to be final. The real world is probably messier, slower, and much more incremental.
