IBM Shares Plunge After Anthropic Claims AI Can “Flip the Equation” on COBOL Modernization

What Happened (Facts)

IBM shares dropped about 13% on Monday, Feb. 23, 2026, in what multiple outlets described as the company’s steepest one-day decline in decades.

The selloff was widely tied to a new Anthropic announcement and blog post claiming its Claude Code tool can dramatically reduce the time and cost required to modernize COBOL—a legacy programming language still deeply embedded in high-volume transaction systems (banking, airlines, government, etc.) that often run on IBM mainframes.

Anthropic’s argument was essentially:

COBOL modernization has historically been expensive because teams must first map dependencies, understand workflows, and identify risks across huge, interdependent codebases—often taking months or years.

AI can automate much of the “exploration and analysis” phase, surfacing dependencies and documentation faster than manual consultants can.

That “flips” the cost equation—making modernization achievable in quarters instead of years, at least in some cases.

Why markets linked this to IBM: IBM has long benefited from mainframe-centric enterprise computing and the services ecosystem around maintaining and modernizing complex legacy systems. Investors interpreted Anthropic’s pitch as a direct threat to a meaningful slice of IBM’s services and modernization work.

The move also arrived in a broader “AI disruption” tape: Anthropic recently launched Claude Code Security (a codebase-scanning security feature), which helped spark a selloff in several cybersecurity stocks as investors worried AI might commoditize parts of the vulnerability discovery workflow.

IBM didn’t stay silent. In a response post, IBM emphasized that the hardest part of “COBOL modernization” isn’t just translating a language—it’s the vertically integrated enterprise stack around it (middleware, databases, security, transaction monitors, storage, and operational requirements) that can’t be waved away by rewriting code.

What Is Analysis (Interpretation)

1) Why a blog post could erase tens of billions in value (for a day)

IBM didn’t “lose” its COBOL business overnight—what the market repriced was future margin power.

Modernizing legacy systems has historically been a slow, high-touch consulting-and-integration problem. If AI meaningfully reduces the “discovery” portion (understanding what the system does, how it fails, what depends on what), it could compress billable hours and weaken the advantage of firms that have built large services practices around that complexity.

That is the fear investors priced in: not that IBM’s mainframes disappear tomorrow, but that the work around them becomes cheaper, faster, and more contestable, inviting new entrants (AI-first toolchains, cloud-native integrators, or even internal teams empowered by agents).

2) The counterpoint: COBOL isn’t the whole iceberg

IBM’s rebuttal is important because it points to what big enterprises actually mean when they say “legacy modernization.”

In many banks, airlines, and governments, the most critical systems aren’t just “COBOL files.” They’re ecosystems embedded in tightly controlled environments with:

transaction monitors

security and identity layers

databases and message queues

operational constraints

audit/compliance requirements

decades of “tribal knowledge” encoded in runbooks and incident practices

IBM’s argument—essentially “translation captures almost none of the real complexity”—is plausible.

Even if Claude Code can help map dependencies faster, large institutions still need governance, testing, migration planning, rollback strategy, and regulatory sign-off. AI can shrink the “understanding” step, but it doesn’t automatically solve the “operating safely in production” step.

So the real competitive question becomes: Does AI merely accelerate IBM’s modernization pipelines—or does it allow competitors to do the same work without IBM?

3) AI might expand the modernization market, not just cannibalize it

There’s another possibility investors often underweight during disruption scares: when a task becomes cheaper, demand can grow.

If modernization has been delayed for years because discovery costs were too high, AI-assisted tooling could unlock projects that were previously “not worth it.” Anthropic’s framing is exactly that: modernization stalled because understanding cost more than rewriting; AI changes that equation.

If that’s true, IBM could face margin pressure per project but still see more total projects—especially if IBM builds or partners into the AI tooling layer (IBM is already positioning offerings around modernization and has its own enterprise AI stack narrative). In that scenario, IBM’s risk is less “collapse” and more “commoditization + higher volume + different economics.”

4) Markets are reacting to “AI as a universal substitute,” not to measured adoption curves

The speed of the selloff looks like a narrative-driven repricing: investors increasingly treat every credible “AI can do X” claim as an immediate business-model threat.

We’ve seen the same pattern across other sectors recently: cybersecurity, customer support tooling, basic marketing content, even some analytics workflows—stocks move on perceived substitution risk before customers have fully validated the tools at scale.

That doesn’t mean the threat is fake. It means public-market reactions can run ahead of enterprise reality, especially when the story is simple (“AI replaces consultants”) and the numbers feel huge (“hundreds of billions of lines of COBOL”).

5) What would actually validate the “IBM casualty” thesis?

If you’re trying to separate signal from hype, the key indicators aren’t the viral blog post—they’re adoption and procurement behaviors:

Do large regulated enterprises allow agentic tools into sensitive codebases at scale?

Are modernization contracts being rebid with AI-first vendors winning?

Do IBM’s services bookings or mainframe-adjacent revenues show measurable erosion tied to AI tooling?

Does IBM respond with its own automation capabilities that keep it in the center of modernization programs?

In other words: the market moved on possibility; the proof will come from enterprise buying cycles.