For years, application modernisation sat in the category of things companies knew they should do but kept deprioritising. The legacy systems still worked. The cost of staying was manageable. There were always more urgent projects in the queue.
That calculation has shifted.
In 2026, the urgency driver for modernisation is no longer technical debt. It's AI-readiness. And the difference matters, because it changes who owns the problem, how it gets funded, and how quickly it needs to move.
Most leadership teams have made their AI commitments. There's a strategy, there's budget, there's pressure from the board. The question is no longer whether to invest in AI, it's why the investments aren't producing the results anyone expected.
When you look at where the friction actually is, the pattern is consistent. The models are capable. The tooling exists.
What's broken is the infrastructure underneath. According to a 2026 Cognizant study of 1,000 Global 2000 executives, organisations currently allocate 61% of their technology budget to keeping existing systems running.
Gartner found that poor data quality alone can increase AI implementation costs by up to 40%. And only 7% of enterprises say their data is fully ready for AI.
That's not a technology problem. That's a foundation problem.
The companies seeing genuine returns on AI investment are not necessarily the ones who moved fastest or spent the most. They're the ones who dealt with their infrastructure first.
Here's what makes this moment different from previous modernisation cycles: AI has specific, non-negotiable requirements that most legacy architectures simply cannot meet.
Traditional enterprise software was built around one assumption: that humans would be the primary consumers of information. A person runs a report. A person reads the dashboard. A person decides what to do next. The system's job was to store and surface data on request.
Autonomous AI systems need something fundamentally different. They need clean, structured, accessible data, not once a day, but in real time. They need APIs that respond in milliseconds, not batch processes that run overnight.
They need event-driven architectures that emit signals as things happen, not systems designed around scheduled jobs and manual triggers.
When you try to build AI capability on top of infrastructure that wasn't designed for this, you're not adding a new feature. You're asking the system to do something it was never architected to support. The result is what most teams are experiencing right now: slower-than-expected progress, data pipeline work eating all the engineering time, and use cases that looked straightforward in a demo that turn out to be structurally complex in production.
Previous modernisation arguments were defensive. Reduce technical debt. Avoid security risk. Cut maintenance costs. These were valid, but they competed with a hundred other priorities and usually lost.
The AI argument is offensive. It's not about protecting what you have, it's about whether you can compete at all over the next three years.
The gap between companies with modern, AI-readable infrastructure and companies without it is compounding quickly. Every quarter that passes without addressing the foundation is a quarter your competitors with clean infrastructure are using to deploy, learn, and improve. The debt isn't just technical anymore. It's strategic.
"The companies that will win with AI are those that stop thinking in terms of features and start thinking in terms of infrastructure."
A McKinsey analysis found that using AI to assist modernisation efforts can reduce timelines by 40 to 50% and cut costs derived from technical debt by around 40%. That's the financial case for moving now rather than later. But the more important number is the Cognizant projection: companies currently spending 61% of their IT budget on legacy maintenance plan to bring that down to 27% by 2030. The ones who get there first will have a significant structural advantage in everything that follows.
The word gets used loosely, so to be precise: this is not about replacing everything.
Big-bang rewrites are expensive, disruptive, and historically unreliable, close to 70% fail to deliver on time, on
budget, or with the original business logic intact. That's not the approach.
What works is a layered strategy. Leave the stable core systems alone. Systematically expose their data and functionality through modern interfaces. Build API layers on top of legacy systems that are working fine but locked.
Introduce event streaming where there were previously static databases. Create data pipelines that clean, unify, and normalise records that were never designed to talk to each other.
The goal is specific: make your existing systems AI-readable. That's a narrower, more achievable target than a full rewrite. And it's where the value actually comes from.
The single most consistent obstacle to AI progress in established companies isn't the models, the compute, or the team's technical ability. It's data quality.
Not data volume. Most companies have plenty of data. It's the shape and reliability of it that creates the problem.
AI systems are not forgiving of inconsistency the way human analysts are. A human analyst notices that two systems
use different naming conventions and adjusts. An automated system won't, unless you've explicitly handled that.
Duplicate records, missing fields, inconsistent formats, siloed data that was never intended to be joined, all of it creates friction that either slows AI workflows down or breaks them entirely.
This is where the infrastructure work earns its value. Fixing it isn't glamorous. It's data modelling, governance frameworks, pipeline engineering. But it's foundational. The companies that get this right first move dramatically faster on every AI initiative that follows, because they're not spending half their engineering effort cleaning up before they can do anything useful.
The companies making real progress on this tend to follow the same pattern. They start with an honest modernisation audit: which systems are stable but outdated, which data is valuable but inaccessible, where the biggest structural gaps are. Then they prioritise ruthlessly, identifying two or three AI use cases that would deliver meaningful business value in six to twelve months, and working backwards from those to understand what infrastructure changes are actually required.
That direction matters. Working backwards from the use case, not forward from the technology. It prevents over-engineering and keeps the work connected to outcomes the business actually cares about.
The ROI when this is done well is significant. Kyndryl's 2025 State of Mainframe Modernisation Survey of 500 business and IT leaders found returns ranging from 288% to 362% depending on the approach taken. Those are not marginal improvements. They're the kind of numbers that shift a board conversation.
The companies that have already built AI-readable infrastructure are not standing still. They're deploying, learning, and iterating at a pace that organisations still carrying legacy architecture cannot match.
The urgency in 2026 is real, but it's not the urgency of crisis. It's the urgency of compounding advantage. The gap between where you are and where your competitors are building towards closes slowly at first, then doesn't.
Modernisation was always the right thing to do. Now it's the prerequisite for the thing everyone's trying to do.
If your team is working through where your current infrastructure stands in relation to your AI roadmap, or trying to build a realistic modernisation sequence, that's a conversation worth having early.
