The tech debt reckoning: why legacy modernisation stopped being optional

Every tech leader knows the feeling. The board wants AI on the agenda, the roadmap is full of shiny things, and underneath it all sits a core system older than half the workforce. For years, living with that tension was a defensible choice. In 2026, it is getting harder to defend.

The numbers tell part of the story. McKinsey research has found that CIOs estimate 10 to 20 percent of the technology budget meant for new products gets diverted into resolving tech debt issues, that debt can represent up to 40 percent of the technology estate in large enterprises, and that 60 percent of surveyed leaders say it has increased materially over the past three years. Reporting in CIO cites analysis putting the waste from legacy inefficiency at more than US$370 million a year for the average global enterprise, and notes that in financial services, legacy platforms, often COBOL-based, consume 70 to 75 percent of annual IT spending.

Why the urgency now

Three forces have converged. The first is AI. McKinsey’s research found companies with fragmented or legacy systems were 30 percent more likely to experience AI implementation delays. The UK offers a cautionary tale at national scale: its Public Accounts Committee warned the government wants to “mainline AI into the veins of the nation” while an estimated 28 percent of central government systems qualified as legacy in 2024, with outdated IT costing public services up to £45 billion a year in missed productivity savings.

The second is security. Closer to home, New Zealand’s health system has seen multiple IT outages in Auckland and Northland over the past year, along with health portal hacks. Budget 2026 responded with $153.6 million over four years for Health NZ to expand national cyber security monitoring and upgrade IT safety systems. Old systems are no longer just slow. They are the soft entry point.

The third is people. BMC’s 20th annual mainframe survey of 1,000 technologists shows veteran engineers retiring, though with a genuinely hopeful twist: millennials now make up just over half of respondents and Gen Z has jumped from 1 percent in 2018 to 15 percent. A separate Kyndryl survey found more than 80 percent of organisations have deployed or plan to deploy large language models in mainframe environments. The skills cliff is real, but the platforms are not being abandoned. They are being reinvented, and the organisations that plan the handover deliberately will fare far better than those that let it happen by attrition.

The New Zealand picture

Health NZ has described its own digital environment as complex, fragmented and fragile, and is putting $300 million of its own budget into the first three years of the Health Digital Investment Plan, launched in November 2025 after long delays. The money targets ageing devices, radiology systems and core IT platforms, with the Shared Digital Health Record now scheduled for national launch by mid-2026. It is a frank public admission of legacy pain, and a funded plan to deal with it.

And New Zealand has proof this can be done well. Inland Revenue’s transformation, a roughly $1.5 to $1.7 billion programme, retired a COBOL mainframe built in the late 1980s and closed in June 2022 with six releases across four stages delivered on time and under budget. By 2023/24 it had been credited with around $2.9 billion in additional revenue and a $1.3 billion reduction in compliance costs. That programme is now the reference point every business case in the country gets measured against, fairly or not.

The quiet lesson across all of it: modernisation has stopped being a maintenance conversation and become a capability conversation. The question boards are starting to ask is not what the old system costs to run, but what it prevents the organisation from doing. For NZ tech leaders, that reframing may be the most useful tool of all, because the leaders who can answer it honestly are the ones who will get funded.