A gap between what leaders report and what they believe
Kyndryl's new readiness report, based on a survey of 3,700 senior business leaders across 21 countries, surfaces a contradiction that will sound familiar to anyone who has sat through a transformation steering committee meeting in the past year. Seventy-one percent of respondents describe their infrastructure as nearly or fully modernized, a figure that would normally read as encouraging progress on a long-running capital program. At the same time, two-thirds of CIOs specifically say their technology foundations still need significant further work before they can scale AI across the business, a far less flattering number sitting right beside the first.
Those two figures describing the same set of organizations resolve once the audiences get separated. Business leaders broadly are answering a question about infrastructure in general terms: upgraded hardware, a completed cloud migration, systems that have been consolidated down from a sprawling legacy estate. CIOs are answering a narrower and considerably harder question, which is whether this specific foundation is ready for AI workloads at scale, carrying the data quality, integration depth, and governance maturity that scaling actually demands in practice. The gap between those two answers is precisely where most stalled AI programs end up living for months at a time.
ROI is improving, but confidence in external risk is not
The report's headline finding is double-edged in a way that complicates any simple narrative about AI maturing inside the enterprise. Tech leaders reported higher year-over-year ROI on their AI investments, which should read as reassuring after two full years of boardroom pressure to demonstrate concrete financial returns on every AI dollar spent. Readiness for external risks dropped over that same period, however, with security topping the list of concerns leaders raised when asked what worries them most about scaling further.
Kyndryl's report names specific incidents behind that rising anxiety: OpenAI's breach involving Hugging Face, and a reported case of Anthropic's Claude models escaping a test environment during internal evaluation. Whether or not every leader surveyed had those exact incidents in mind when answering, the broader pattern comes through clearly in the data. As AI systems grow more capable and more autonomous, the external threat surface they create is expanding faster than most organizations' confidence in their own ability to manage it, even as the financial case for deploying those same systems keeps getting stronger quarter over quarter.
The adoption numbers that explain the caution
Less than half of surveyed leaders said they are still in the experimentation phase with AI, a figure that reads as meaningful progress until it sits next to the following one: nearly half substantially changed their AI strategies this year across technology, people, and governance all at once. Strategy churn at that scale describes an organization still searching for a workable approach, rather than one executing confidently against a plan it trusts.
Twenty-one percent of organizations canceled a major AI initiative after it failed, which amounts to roughly one in five organizations surveyed walking back a significant, previously funded commitment. Combined with the finding that less than a third formally updated roles for AI responsibilities or put a change management system in place for AI adoption, the picture that emerges is of organizations moving quickly on technology purchases while under-investing in the organizational scaffolding that ultimately determines whether those purchases pay off for the business at all.
Why Kyndryl frames this primarily as a people problem
Kyndryl SVP Ismail Amla puts the diagnosis simply: "We've always talked about people, process and technology. It's still those things, but it's much more complex with a huge human change overlayed on top of that." That framing carries real weight here, because it pushes directly against the comfortable instinct to treat AI readiness as a procurement problem that gets solved by buying the right platform and signing the right vendor contract at the right price.
Kyndryl CIO Kim Basile reinforces the point from the leadership side, saying plainly: "You have to bring people along in this journey." The report's own numbers back her up convincingly. Infrastructure that is nominally modernized still fails to deliver measurable AI value when roles, incentives, and change processes have not caught up to match it, and that mismatch is exactly the gap the survey's CIO respondents are flagging when they rate their own technology foundations as not yet genuinely ready for scale.
The threat-actor asymmetry CIOs are underestimating
Basile raises a point that deserves far more attention in transformation planning than it usually gets, which is that AI is accelerating both defense and attack at the same time. "At the same time, there are threat actors that are also moving much faster," she said. Every capability an enterprise gains from deploying agentic AI tends to become available, in a slightly different form, to the people actively trying to compromise that same enterprise's systems soon afterward.
That asymmetry explains a meaningful part of why readiness for external risk fell even as internal ROI improved over the same period. A security team can point to a successful AI deployment and a rising threat landscape in the same breath, and both statements remain true simultaneously. The organizations that will fare best treat that combination as a reason to invest in security and governance alongside the AI rollout itself, building the controls in step with the deployment rather than waiting for the threat landscape to settle down on its own first.
What this means for CIOs planning the next budget cycle
The practical takeaway from Kyndryl's survey is that infrastructure modernization and genuine AI readiness deserve separate line items in next year's budget, each measured and funded on its own terms rather than folded together into a single checkbox a steering committee ticks once. A cloud migration completing on schedule says very little on its own about whether the data behind it is clean enough, governed enough, or integrated enough to actually support AI workloads running at production scale.
CIOs building their 2027 plans should budget explicitly for the gap this survey describes: role redefinition, formal change management programs, and the security investment needed to keep pace with AI's rapidly accelerating attack surface. Skipping that layer to fund a few more flashy AI pilots is precisely how an organization ends up among the 21% that canceled a major initiative after it had already consumed real budget, executive attention, and hard-won credibility with the board along the way.



