AINEWS
Ensono's 2026 survey of US and UK IT leaders found 78% now see their old systems as more valuable than two years ago, and 52% are extending them rather than replacing them.
The sequencing that built enterprise IT for decades was mainframe first, AI second: retire the system, lift the data, rewrite the application, then talk to the AI team. That sequence is now running into its own cost curve. Ensono's second annual State of IT Modernization Report, released September 15, found 71% of modernization initiatives ran over budget, with the UK at 77% and the US at 66%. 97% of organizations blamed talent shortages for the cost overruns. 61% had paused, delayed, scaled back, or abandoned a modernization initiative in the past 24 months. The companies pouring money into AI capex are discovering that the substrate for that capex, the transaction systems, the data pipelines, the batch jobs, sits on mainframes and COBOL services that nobody on the AI team can read.
Half the organizations in the Ensono survey, 52%, are now optimizing and extending their existing systems rather than replacing them outright, with the UK at 57% ahead of the US at 48%. 78% say legacy systems play a more important role than they did two years ago, a finding The Register read as evidence that enterprises are "sweating" legacy assets harder as AI capex rises.
Brian Klingbeil, Ensono's chief strategy officer, frames mainframes as "intensely powerful, reliable and efficient sources of computing that can now be augmented and made more agile thanks to AI." His argument is that the modernizers who return AI ROI fastest will be the ones who learn to distinguish what to replace from what to augment. That distinction used to be a budget line. It is now a sequencing decision.
Kyndryl's November 2025 launch of an agentic AI framework for the mainframe sits inside the same arc. The company cited its 2025 State of Mainframe Modernization Survey, in which 88% of respondents said they had implemented or planned to implement AI, including agentic and generative systems, on the mainframe. The expected benefits tracked the Ensono priorities: business agility at 37%, faster repeatable operations at 32%, cost savings at 31%. The blocker was the same: 70% cited multi-skilled talent gaps. Two vendors, two surveys a year apart, point at the same constraint. The human capacity to operate legacy stacks is the binding input, not the silicon.
Asked what was holding back AI ROI, Ensono's respondents ranked workflow integration at 33%, infrastructure limitations at 28%, and governance concerns at 25%. Workflow integration, the top-ranked blocker, is the operational constraint: a model that works in the lab but cannot be wired into the surrounding business process in the time the budget allows. Infrastructure limitations are the legacy constraint restated. Governance is the regulatory one: a control framework written for the previous generation of systems, asked to audit model decisions it was not designed to capture.
The leaders in the Ensono data look measurably different from the laggards. 77% of self-identified leaders have end-to-end visibility into their modernization, against 39% of laggards. 74% measure success in a structured way, against 16%. 64% say their IT roadmap is aligned with the business, against 36%. The AI payoff tracks the same axis: organizations proactively modernizing for AI are nearly twice as likely as the rest to say AI is unlocking new revenue, 58%.
Kyndryl's 88% mainframe-AI figure is the test. By the time Ensono's third annual survey lands in September 2027, that number will show whether the augment-not-replace strategy has converted from pilots to revenue. The 2027 AI capex plans most likely to clear board review are the ones that mapped the legacy substrate before they specified the model.