For decades, IBM has occupied an enviable position in enterprise computing: the indispensable steward of aging COBOL systems that power the world’s banks, insurers, and government agencies. That position came under sudden and dramatic pressure on February 23, 2025, when shares of IBM cratered as much as 13% in early trading after Anthropic announced that its Claude Code AI tool could now handle COBOL-to-modern-language migration — a task that has long been one of IBM’s most lucrative consulting and software businesses.
The sell-off, first reported by Slashdot, wiped roughly $25 billion from IBM’s market capitalization in a single session and sent shockwaves through the broader IT services sector. Shares of Accenture, Infosys, and Wipro — all of which derive significant revenue from legacy modernization engagements — also traded lower on the news.
The COBOL Problem That Refuses to Die
COBOL, or Common Business-Oriented Language, was developed in 1959 and remains embedded in systems that process an estimated $3 trillion in daily commerce, according to Reuters. The language runs on roughly 95% of ATM transactions and 80% of in-person financial transactions globally. Despite repeated predictions of its demise, COBOL endures because the cost and risk of replacing it have historically been prohibitive. Failed modernization projects at major banks and government agencies — some costing hundreds of millions of dollars — have reinforced a conservative approach: if the old code works, don’t touch it.
IBM has been the primary beneficiary of this inertia. Its mainframe hardware division, its z/OS operating system, and its consulting arm have all profited from the fact that enterprises feel locked into their existing COBOL infrastructure. The company’s 2023 acquisition of mainframe modernization startup Advanced Modernization and its continued investment in its own Watsonx AI platform for code transformation underscored how central this revenue stream is to IBM’s strategy. Analysts at Morgan Stanley estimated in a recent note that COBOL-related services and infrastructure account for between $5 billion and $7 billion of IBM’s annual revenue, though IBM does not break out the figure explicitly.
What Anthropic Actually Announced
Anthropic’s announcement centered on Claude Code, an agentic coding tool that allows developers to interact with Claude’s large language model directly from the command line. According to Anthropic’s technical documentation, Claude Code can now read, interpret, and translate COBOL programs into modern languages such as Java, Python, and Go — while preserving business logic, handling copybook dependencies, and generating unit tests for the translated output.
The company demonstrated the capability on a sample COBOL banking application with approximately 500,000 lines of code, claiming that Claude Code completed the initial translation in under 72 hours with what Anthropic described as “high fidelity” to the original business rules. A human review team then spent an additional two weeks verifying and refining the output. Anthropic was careful to note that the tool is not a fully autonomous replacement for human engineers but rather an accelerant that could reduce modernization timelines from years to months.
Why Wall Street Reacted So Sharply
The magnitude of IBM’s stock decline surprised even bearish analysts. Several factors amplified the market’s reaction. First, IBM has spent the past three years positioning itself as an AI company, with CEO Arvind Krishna repeatedly telling investors that generative AI would drive the next wave of consulting demand. The irony that a rival’s AI product could undermine one of IBM’s most durable franchises was not lost on traders. Second, the announcement arrived during a period of already elevated concern about AI’s impact on IT services spending. A January 2025 survey by Gartner found that 42% of CIOs planned to reduce spending on legacy modernization consulting over the next two years, citing AI-assisted tools as a primary reason.
Third, and perhaps most significantly, Anthropic’s move validated a thesis that competitors like Google, Microsoft, and several startups have been advancing: that large language models are uniquely well-suited to understanding and translating legacy code because they can process the vast corpora of COBOL documentation, patterns, and idioms that exist in training data. Google’s own code migration tools, integrated into its Gemini platform, have been quietly gaining traction with enterprise clients, though none had made as bold a public claim about COBOL specifically.
IBM’s Response and the Watsonx Counter-Narrative
IBM moved quickly to contain the damage. In a statement released later that day, the company said it “welcomes innovation in legacy modernization” but cautioned that “translating code is only a fraction of the modernization challenge.” IBM pointed to its own Watsonx Code Assistant for Z, which the company launched in 2023 specifically to help clients understand and refactor COBOL applications running on its mainframe platform. The company argued that its tool has an advantage because it is deeply integrated with the z/OS environment and understands the specific runtime behaviors, JCL job control language, and CICS transaction processing that characterize real-world mainframe workloads.
“Anyone can translate syntax,” said an IBM spokesperson, according to coverage in Slashdot. “The hard part is understanding the decades of accumulated business logic, the undocumented edge cases, and the integration points with hundreds of other systems. That requires domain expertise that we have built over 60 years.” The statement also noted that IBM’s mainframe customers typically have regulatory and compliance requirements that make wholesale migration to non-IBM platforms a multi-year, board-level decision rather than a technical exercise.
The Skeptics Have a Point
Industry veterans were quick to point out that the history of automated COBOL migration is littered with failures. Micro Focus, now part of OpenText, has offered COBOL migration tools for over two decades with mixed results. The fundamental challenge is not translating the code itself but replicating the exact behavior of systems that have been patched, extended, and customized over 40 or 50 years — often by programmers who are now retired or deceased, and whose modifications were never formally documented.
“I’ve seen at least five waves of ‘COBOL is finally dead’ announcements in my career,” said a principal architect at a major U.S. bank who spoke on condition of anonymity. “Every time, the tools get better at the easy 80%. But the last 20% — the stuff that actually matters, the edge cases that handle billions of dollars — that’s where projects go to die.” This skepticism is well-founded. The Commonwealth Bank of Australia’s five-year, $750 million core banking replacement, completed in 2012, remains one of the few large-scale COBOL migrations widely regarded as successful, and even that project ran significantly over budget and timeline.
The Broader Implications for IT Services
Regardless of whether Anthropic’s specific claims hold up under enterprise-grade scrutiny, the market reaction exposed a deeper anxiety about the future of IT services revenue. The global IT services market generates approximately $1.3 trillion annually, according to IDC, and a significant portion of that spending goes toward maintaining and incrementally modernizing legacy systems. If AI tools can compress modernization timelines by even 50%, the economic impact on firms like IBM, Accenture, Tata Consultancy Services, and Cognizant would be substantial.
Some analysts see an upside scenario for IBM, however. If AI-assisted modernization actually accelerates the migration away from mainframes, IBM could pivot to capturing revenue from cloud infrastructure and AI platform fees rather than consulting hours. The company’s hybrid cloud strategy, built around Red Hat OpenShift, is designed precisely for this transition. But the timing and execution risk are significant. IBM would need to replace high-margin consulting revenue with what are typically lower-margin cloud infrastructure fees, at least in the near term.
What Happens Next in the COBOL AI Arms Race
The competitive dynamics are now set. IBM will almost certainly accelerate the capabilities of Watsonx Code Assistant for Z, likely announcing new features at its annual Think conference. Google and Microsoft are both expected to make enterprise code migration a centerpiece of their AI strategies in 2025. And Anthropic, flush with billions in recent funding from Amazon and other investors, has signaled that enterprise software development — not just consumer chatbots — is a core part of its commercial roadmap.
For the thousands of enterprises still running COBOL, the announcement may ultimately be more significant as a psychological turning point than a technical one. The mere fact that a credible AI company is publicly claiming it can handle COBOL migration will embolden CIOs to revisit modernization business cases that were previously shelved as too risky or too expensive. Whether Claude Code or any other tool can actually deliver on that promise at enterprise scale remains an open question — but it is now a question that every board with a mainframe budget will be asking.
IBM’s stock partially recovered in afternoon trading, closing down 8.4% on the day. But the message from the market was clear: the COBOL moat that has protected IBM’s mainframe franchise for decades is no longer considered impregnable. In an era when AI can read and reason about legacy code at scale, the competitive advantages that once seemed permanent are being repriced in real time.