IBM's COBOL Fortress: Existential Threat or the Best-Timed Overreaction in a Decade?

✦

Anthropic published a blog post on Monday. IBM lost $31 billion in market capitalisation. The stock plunged 13% in its biggest single-day percentage loss since October 2000. With the decline, IBM shares have fallen 27% in February, on track for its biggest one-month slide since at least 1968.

The catalyst was not an earnings miss, not a management departure, not a lost contract. It was a product marketing page from an AI startup describing how its Claude Code tool could help modernise COBOL codebases.

We are launching primary research — 30 expert calls with IBM channel partners and value-added resellers — to find out whether the market's reaction is prescient or panicked.

The irony is difficult to overstate. IBM reported better-than-expected fourth-quarter results in January. Revenue from continuing operations hit $19.7 billion, up 12 percent year-over-year, yielding net income of $5.6 billion. For the full year, revenue was up 8 percent to $67.5 billion, while net income jumped 76 percent to $10.6 billion. IBM Z mainframe revenue surged 67% year-over-year in Q4. CFO James Kavanaugh called it a "record z17 launch, achieving the highest annual revenue for IBM Z in about 20 years." The company guided for more than 5% constant currency revenue growth in 2026 and an additional $1 billion in free cash flow.

Four weeks later, a single blog post erased every dollar of the post-earnings rally and then some.

Key Insights

• IBM just posted its strongest mainframe performance in two decades, and the market erased it in a session. IBM Z revenue grew 67% year-over-year in Q4, with distributed infrastructure up 3% and infrastructure support up 1%. CEO Arvind Krishna noted that each generation of mainframe has performed better than the last, calling it "continuing improvement." He attributed the z17's success in part to increased interest in digital sovereignty and on-premise control. The February sell-off has now overwhelmed an earnings beat, a guidance raise, and the best mainframe cycle in a generation.
• Anthropic's tool targets the discovery phase, not the full migration. Claude Code automates the exploration and analysis phases of COBOL modernisation that traditionally required large consulting teams. It can map dependencies across thousands of lines of code, document workflows, and identify risks that would typically take human analysts months to surface. But discovery is not migration. IBM's own response makes clear that the hardest and most expensive work sits in architecture redesign, runtime replacement, and transaction processing integrity. If AI compresses discovery from months to weeks but leaves the implementation phase intact, the revenue impact on IBM is real but contained.
• IBM already sells its own version of this tool. IBM announced watsonx Code Assistant for Z in August 2023, a generative AI product that enables faster translation of COBOL to Java on IBM Z. Krishna said in July 2025 that IBM's AI coding assistant for mainframes "has got very wide adoption" and that in the majority of cases, customers are using it to understand their COBOL code base and decide what to modernise. The question is not whether AI can accelerate COBOL analysis. IBM has been demonstrating that for three years. The question is whether a competing tool changes the migration destination.
• The consulting segment, not the mainframe itself, is the more immediate pressure point. Full-year consulting revenue was up just 2%, flat at constant currency. IBM's consulting arm earns significant fees from scoping, planning, and managing COBOL modernisation engagements. If AI tools compress the analysis phase, the revenue at risk is the discovery work — not the mainframe hardware or software licences.
• The mainframe modernisation market is growing, not shrinking. The market is valued at $8.18 billion in 2025 and is predicted to reach $25.94 billion by 2035 at a 12.7% CAGR. Who captures the value is the open question.
• The broader software sell-off amplified an already nervous market. Software stocks have been broadly weaker this year on concerns over AI-related disruption. Much of the selling has come on new AI tools released by Anthropic, OpenAI, and Alphabet. The pattern is indiscriminate: Anthropic publishes, incumbents sell. Evercore ISI analyst Amit Daryanani wrote that while mainframe migration could be perceived as a negative for IBM, the company has already provided customers with several modernisation options.
• Enterprise customers are not rushing to leave the mainframe. Eighty percent of banks plan to modernise existing COBOL code through AI-assisted refactoring rather than offload to x86, which lowers integration risk. Krishna stated that more and more clients have recognised that for certain workloads, the mainframe offers the lowest unit cost economics. The installed base is sticky for reasons that go beyond inertia — regulatory compliance, transaction throughput, and decades of optimised performance keep workloads on Z.

Participation Opportunity

Woozle Research is inviting professional investors to sponsor or co-sponsor this primary research. Participation is collaborative. All funds receive full access to research outputs including interview summaries, transcripts, and the final synthesis report.

• Launch date: February 26, 2026
• Delivery date: March 10, 2026
• Participation cap: Limited to 5 funds
• Catalyst: Anthropic Claude Code COBOL tool, IBM mainframe franchise repricing, channel sentiment shift
• Research scope: 30 expert calls with IBM channel partners, value-added resellers, and systems integrators who sell, implement, and support IBM Z infrastructure and COBOL modernisation services across banking, insurance, and government verticals
• Deliverables: Raw data, transcripts, synthesis report, analyst access

This research will proceed with a minimum of one fund and is limited to a maximum of five. Email to confirm your interest.

The Catalyst

The story of IBM and COBOL is older than most of the people trading the stock. COBOL was developed in 1959 through a public-private partnership that included the Pentagon and IBM itself. It handles an estimated 95% of ATM transactions in the US. Hundreds of billions of lines of COBOL run in production every day, powering critical systems in finance, airlines, and government. Despite that, the number of people who understand it shrinks every year.

For decades, the scarcity of COBOL talent and the catastrophic risk of failed migrations created an impenetrable moat around IBM's mainframe business. Customers did not stay because they loved COBOL. They stayed because leaving was too dangerous and too expensive.

Anthropic's argument is that AI changes the economics. Legacy code modernisation stalled for years because understanding legacy code cost more than rewriting it. Claude Code can trace execution paths, map data flows between modules, document business logic buried in uncommented code written decades ago, and identify migration risks. Anthropic claims it can help teams modernise COBOL codebases in quarters instead of years. The company simultaneously released a Code Modernization Playbook providing step-by-step implementation guidance.

The market treated this as a revelation. The technical community treated it as old news.

The perils and opportunities presented by COBOL migrations, and the potential for AI to accelerate refactoring of legacy applications, are not new. Just last week, Infosys chairman Nandan Nilekani said the rise of AI means the cost of rewriting legacy apps has become affordable and made such moves imperative. AWS, Microsoft, IBM spin-out Kyndryl, and NTT have all launched mainframe migration initiatives in recent years. All that activity didn't dent IBM's mainframe business. Last month, Big Blue reported its highest mainframe revenue in 20 years.

That is the tension at the heart of this trade. The threat narrative is real but not new. Every prior attempt to automate mainframe migration has failed to move the installed base. The question is whether this time is different because the AI tools are genuinely better, or whether the market is once again confusing a compelling product demo with enterprise-grade adoption.

IBM's own positioning reveals how it thinks about the risk. Krishna said sovereignty was a key issue for mainframe customers, and that GenAI made mainframes easier to leverage and modernise. IBM's strategy is not to prevent COBOL modernisation. It is to own the modernisation process and keep the workloads on Z. If a bank uses watsonx Code Assistant to convert COBOL to Java but runs that Java on an IBM z17 with the Telum II processor, IBM retains the hardware annuity, the software licence revenue, and adds a new AI tooling subscription.

The bear case requires not just that AI accelerates COBOL analysis, but that it accelerates migration off the platform entirely — onto AWS, Azure, or Google Cloud.

That is where the channel partners become essential. They are the ones fielding customer calls this week. They know whether the conversations are shifting from "help us modernise on Z" to "help us get off Z." They know whether Anthropic's name has entered procurement discussions or whether it remains an investor talking point disconnected from enterprise reality.

The next 60 days, leading into IBM's Q1 earnings in late April, will determine whether this sell-off finds a floor or deepens into a structural repricing of the mainframe franchise.

Key Intelligence Questions

Our primary research will focus on the commercial and operational dynamics that determine whether Anthropic's Claude Code COBOL tool changes the trajectory of IBM's mainframe franchise or represents a continuation of a decades-old migration narrative that has never meaningfully dented the installed base. Each question targets a specific input to the investment case that public data cannot resolve.

Channel Sentiment: Has the Customer Conversation Changed This Week?

The most time-sensitive question is the simplest. IBM channel partners and VARs are the front line of customer engagement. They sell IBM Z hardware, negotiate software licence renewals, and scope modernisation projects. If Anthropic's announcement has shifted enterprise buyer sentiment, channel partners will know before anyone else.

The distinction that matters is between awareness and action. CIOs at banks and federal agencies read the same headlines investors do. Some will ask their IBM account teams about Claude Code. A smaller number will initiate formal evaluations. An even smaller number will change procurement decisions. The gap between headline awareness and genuine pipeline disruption is the variable that determines whether the 13% sell-off was rational or reflexive.

Channel partners can tell us directly: in the 48 hours since Anthropic's announcement, have any existing IBM Z customers raised questions about alternative modernisation pathways? Are any active deal cycles for z17 hardware or software renewals being delayed or reconsidered? Have any customers specifically mentioned Claude Code by name in procurement or strategy discussions?

Migration Destination: Does Faster COBOL Analysis Mean Leaving IBM, or Staying?

This is the central analytical question. Anthropic's tool targets the discovery and analysis phase of COBOL modernisation. It helps organisations understand what their COBOL code does, map dependencies, and plan migrations. But understanding the code is not the same as moving it. The destination of the migrated workload is what determines the revenue impact on IBM.

Eighty percent of banks plan to modernise existing COBOL code through AI-assisted refactoring rather than offload to x86. If that pattern holds, AI-accelerated discovery leads to faster modernisation on Z, not faster migration off it. IBM captures value through watsonx Code Assistant for Z licences, consulting fees for the implementation phase, and continued hardware revenue. If the pattern breaks — if faster discovery creates momentum toward cloud migration — IBM faces a structural headwind on its most profitable segment.

Channel partners who manage modernisation projects across multiple enterprise accounts can clarify which pattern is dominant. Are customers using AI tools to modernise in place on Z, or are they using AI tools to plan exits? The answer varies by vertical, by workload criticality, and by regulatory environment. Banking and government may behave very differently from retail or logistics.

Competitive Positioning: Is Anthropic a Real Competitor or an Investor Narrative?

Anthropic makes large language models. It does not sell mainframe consulting, systems integration, or enterprise IT infrastructure. Claude Code is a developer tool, not an end-to-end modernisation platform. IBM's watsonx Code Assistant for Z supports developers through every phase of the lifecycle — from application discovery and analysis to automated refactoring, code explanation, generation, optimisation, transformation, and testing. It integrates with z/OS runtimes, CICS, IMS, and Db2. It is purpose-built for the IBM Z environment.

The competitive question is whether enterprise customers at regulated institutions will trust a general-purpose AI coding tool with production-grade COBOL modernisation, or whether they will prefer a vendor with 65 years of mainframe expertise, purpose-built tooling, and contractual liability. IBM has access to the largest private data set of COBOL data with which to train its models. That training advantage may matter enormously for accuracy in complex, undocumented legacy codebases.

Channel partners can provide a grounded assessment of whether Anthropic is entering real competitive evaluations or whether the threat is being priced by investors who do not understand how enterprise mainframe procurement works.

Consulting Revenue at Risk: Where Does the Value Sit in a Modernisation Engagement?

IBM's consulting segment generated $5.3 billion in Q4 revenue. A meaningful portion of that — though IBM does not disclose the exact figure — comes from legacy modernisation engagements that include COBOL analysis, architecture planning, and implementation. If AI tools compress the analysis phase, the question is how much of total engagement value that phase represents.

Industry practitioners generally estimate that discovery and analysis account for 20 to 30 percent of a typical COBOL modernisation project's total cost. Implementation, testing, data migration, and production cutover represent the balance. If AI compresses a $10 million engagement's discovery phase from $2.5 million to $500,000, that is a meaningful but not existential reduction. If AI compresses the entire engagement from $10 million to $2 million, the implications are far more severe.

Channel partners who scope and deliver these engagements can provide granular data on phase-by-phase cost structures. They can clarify whether the analysis phase that Anthropic targets is genuinely the cost bottleneck, or whether the integration, testing, and compliance validation phases account for the majority of contract value. This is the single most important data point for sizing the consulting revenue at risk.

Platform Stickiness: What Keeps Workloads on Z Beyond Inertia?

The deepest question in the IBM bear case is whether the mainframe's durability is driven by genuine technical superiority or mere switching costs. If it is switching costs alone, then any tool that reduces the cost of switching accelerates attrition. If it is technical superiority, then modernising COBOL on Z — converting to Java, adding AI inference, running on Telum II — actually strengthens the platform.

IBM's analysis shows that mainframes handle 72% of transactional work yet consume only 8% of IT budgets. That unit economics argument is its most powerful retention tool. The new z17 systems are proving that mainframe computing and AI can coexist. Financial-services, government, and telecom clients are already running inference directly on Telum II. If AI inference runs natively on Z, the platform becomes more valuable to customers, not less.

Channel partners who manage enterprise accounts across banking, insurance, and government can assess the depth of platform dependency. Are customers locked in by inertia and fear, or by genuine performance, cost, and compliance advantages that cloud alternatives cannot replicate? The answer determines whether faster COBOL analysis is a catalyst for migration or a catalyst for modernisation in place.

How to Participate

Woozle Research is inviting professional investors to sponsor or co-sponsor this primary research. Participation is collaborative. All funds receive full access to research outputs including interview summaries, transcripts, and the final synthesis report.

• Launch date: February 26, 2026
• Delivery date: March 10, 2026
• Participation cap: Limited to 5 funds
• Catalyst: Anthropic Claude Code COBOL tool, IBM mainframe franchise repricing, channel sentiment shift
• Research scope: 30 expert calls with IBM channel partners, value-added resellers, and systems integrators who sell, implement, and support IBM Z infrastructure and COBOL modernisation services across banking, insurance, and government verticals
• Deliverables: Raw data, transcripts, synthesis report, analyst access

This research will proceed with a minimum of one fund and is limited to a maximum of five. Email to confirm your interest.

This document is for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Woozle Research conducts primary research on behalf of institutional investors. All research is conducted in compliance with applicable regulations.