What 396 companies told Mavvrik
Mavvrik's 2026 State of AI Cost Governance Report surveyed 396 enterprise organizations across industries in April and May 2026, and it should be required reading for any CIO who has told the board that AI spend is under control. Nearly half of respondents said AI spending surprises had escalated all the way to the board, meaning the first time senior leadership learned about a material cost overrun was not through a routine budget review but through an unplanned, reactive conversation. That is the kind of escalation that damages a technology organization's credibility regardless of how the underlying AI initiative ultimately performs.
The downstream effects were severe. One in four surveyed businesses delayed or canceled an AI initiative specifically because of unforeseen costs, and two-thirds said unexpected AI expenses had a material impact on at least one business decision. Those are not abstract governance metrics. They represent real initiatives that lost executive sponsorship, real budgets that got reallocated mid-year, and real teams that had to explain to leadership why a project approved at one cost figure ended up requiring a substantially different one.
Why spend visibility breaks down before the board notices
The report's explanation for why these surprises happen so often points away from model pricing and toward how AI spend actually accumulates inside an enterprise. Costs spread across developer tools, infrastructure consumption, third-party platforms, and per-token model usage, each tracked by a different team, often in a different system, with no single owner responsible for the combined total. A CIO who only monitors the model API bill is missing the infrastructure and tooling spend layered on top of it, and that gap is exactly where the surprises Mavvrik documented originate.
Mavvrik CEO Sundeep Goel framed the underlying shift directly: AI is fundamentally changing how infrastructure is consumed and how costs accumulate. That framing matters because it means the traditional IT cost governance playbook, built around predictable, contracted infrastructure spend, does not transfer cleanly to AI workloads, where consumption scales with usage patterns that are genuinely difficult to forecast until an initiative is already in production and generating real traffic. Enterprises that ported over their legacy budgeting cadence unchanged are the ones now discovering the gap the hard way, mid-quarter, in front of the board.
The emergency freeze as a governance failure
One-third of surveyed organizations resorted to emergency spending freezes on AI initiatives, a response that should be read as a governance failure rather than a prudent cost control measure. An emergency freeze means the organization had no earlier mechanism to catch the cost trajectory before it required a blunt, all-stop intervention. Freezes also carry their own cost: paused initiatives lose momentum, teams lose confidence in leadership's commitment to the program, and vendors and partners lose confidence in the enterprise's ability to execute a multi-quarter engagement reliably.
The organizations avoiding emergency freezes in Mavvrik's survey were not necessarily spending less on AI. They had built the visibility to see cost trends early enough to make smaller, planned adjustments instead of large, reactive ones. That distinction, between a governance structure that surfaces problems incrementally and one that only surfaces them at crisis scale, is the practical difference between the CIOs who explain proactively and the ones who explain defensively after the fact.
Fragmented spend is the real cost driver, not model pricing
Benchmarkit CEO Ray Rike captured the report's core diagnostic in one line: you cannot accurately calculate ROI if you do not know your costs. That statement sounds obvious, but Mavvrik's data suggests most enterprises are effectively operating without that knowledge for AI specifically, even as they maintain rigorous cost tracking for every other major technology category. The gap is not a lack of tooling ambition. It is that AI spend crosses more organizational and system boundaries than a typical software license or infrastructure contract, and most cost governance processes were never rebuilt to follow it across those boundaries.
That fragmentation means the fix is not simply negotiating better model pricing or switching to a cheaper vendor, though those levers help at the margin. The fix is building a single, consolidated view of AI spend across developer tools, infrastructure, platforms, and model usage, owned by one accountable function rather than scattered across whichever team happens to provision each component. Enterprises that have not built that consolidated view are, by Mavvrik's numbers, roughly even odds to hit a board-level cost surprise within the next planning cycle.
What Gartner's own data adds to the picture
Gartner chief research officer Rita Sallam's observation reinforces Mavvrik's findings from a different angle: cost per completed task is rising as agentic workflows become more complex. That statement matters here because it confirms the cost pressure Mavvrik's surveyed enterprises are already living through is not a temporary adjustment period that will resolve as AI matures. It is a structural feature of moving from simple AI interactions toward the more complex, multi-step agentic workflows enterprises are actively rolling out, which means the visibility gap Mavvrik documented will keep costing enterprises more, not less, as adoption deepens.
Taken together, the two findings describe a compounding problem: task-level costs are trending upward across the industry, and most enterprises lack the spend visibility to notice that trend before it produces a board-level surprise. A CIO addressing only one half of that problem, either negotiating costs down or improving visibility without accounting for the underlying upward cost trend, will find the fix incomplete within a budget cycle or two. The two problems have to be solved together, on the same timeline, with the same team accountable for both.
Building the cost governance layer before the next board escalation
The practical response is treating AI cost governance as its own discipline with an accountable owner, not a byproduct of existing IT financial management processes built for predictable, contracted spend. That means consolidating visibility across developer tools, infrastructure, and model usage into a single reporting view, setting spend thresholds that trigger review before costs reach board-escalation scale, and building the forecasting discipline to catch a cost trajectory early enough to make a planned adjustment instead of an emergency freeze.
For CIOs who have not yet had an AI spending surprise reach the board, Mavvrik's numbers suggest the honest question is not whether that conversation is coming but how prepared the organization is to have it on its own terms. Building that cost governance layer now, before the next planning cycle locks in AI budgets built on incomplete visibility, is materially cheaper than explaining an emergency freeze or a canceled initiative after the fact.
