The FinOps gap nobody budgeted for
A new report from Harness, published July 29 and based on a survey of 700 FinOps and engineering leaders, puts a hard number on a problem most CIOs have felt anecdotally all year: roughly a quarter of every dollar spent on AI is wasted. That figure lands at a moment when AI line items are becoming some of the fastest growing entries on enterprise technology budgets, which makes the waste rate more consequential in absolute dollar terms than it would have been eighteen months ago when AI spend was still a rounding error for most organizations.
The report's more useful finding is not the waste percentage itself but the structural reasons behind it. More than half of surveyed organizations have no dedicated owner for AI cost management, meaning spend accountability sits nowhere specific inside the org chart. A similarly sized share forecast AI spending largely through guesswork rather than any structured modeling approach, and more than 40% are still tracking usage in spreadsheets, the same tooling most finance teams abandoned for traditional cloud cost management years ago.
Why AI cost is harder to control than cloud cost
Enterprises spent the better part of a decade building FinOps discipline around cloud compute: tagging, budgets, anomaly detection, chargeback models. AI spend is breaking that playbook because token-based and usage-based pricing from model providers moves in ways traditional cloud reserved instance and spot pricing never did. A single prompt engineering change, a runaway agent loop, or an unexpectedly popular internal AI feature can spike costs by orders of magnitude within hours, and only about one in five organizations in the Harness survey can actually detect that kind of spike within hours rather than days or weeks.
Harish Doddala, VP of cloud and AI cost management at Harness, said what struck him most was not the size of the spend or its growth rate but how consistent the gaps were. 'What surprised me most wasn't the size of the spend or the speed of the growth, but how consistent the gaps are across every size and geography,' Doddala said. That consistency argument is the part CIOs should take most seriously: this is not a problem that resolves as an organization matures its AI program, because large, sophisticated enterprises in the survey showed the same gaps as smaller ones.
The board conversation this data should trigger
For CIOs and CFOs jointly overseeing AI budgets, a 25% waste rate translates directly into a board-level conversation about accountability, not just tooling. If AI spend at a mid-size enterprise runs into eight figures annually, a quarter of that being wasted amounts to real money that could fund an entire additional initiative elsewhere in the roadmap. The Harness data suggests the fix starts with organizational ownership before it starts with software: assigning a named owner for AI cost governance is a prerequisite for any tooling investment actually working as intended.
This also reframes how CIOs should be evaluating AI vendor contracts going forward. Usage-based pricing models that seemed attractive for their pay-for-value framing are the same models driving the forecasting guesswork the survey documents, since actual spend depends on usage patterns that shift week to week. Enterprises negotiating new AI platform contracts should push for cost predictability features, spend caps, and granular usage reporting as standard contract terms from the outset, well before renewal season, rather than accepting them as an afterthought once an unexpectedly large bill arrives.
What good AI cost governance actually looks like
The organizations in the Harness survey that are managing AI cost well share a small set of practices: a named cost owner who sits at the intersection of engineering and finance, automated anomaly detection tuned specifically for usage-based AI pricing rather than repurposed cloud alerting, and forecasting built on actual usage telemetry rather than budget guesswork carried over from the prior fiscal year. None of this is exotic FinOps theory, it is the same discipline enterprises applied to cloud a decade ago, just not yet applied to AI.
The urgency here is that the gap is compounding. As AI usage scales across more business functions, the absolute dollar impact of a 25% waste rate scales with it. CIOs who treat AI cost governance as a 2027 planning item rather than a 2026 fix are choosing to let a quarter of a growing budget line leak for another full fiscal year, at a moment when finance leadership is already scrutinizing AI ROI closely.
The takeaway for CIOs building next year's budget
Harness's findings arrive as a useful complement to the broader IT spending data showing double digit growth in AI-related infrastructure and software budgets industry wide. Growth in the top line spend number without a corresponding investment in cost governance is exactly how a 25% waste rate becomes structural rather than transitional. CIOs building 2027 budgets should treat AI FinOps as its own line item, with a named owner, dedicated tooling, and reporting cadence measured in hours rather than the monthly or quarterly cycles most cost management still runs on today.
The practical first step for any organization recognizing itself in these numbers is an audit: pull actual AI usage and billing data across every vendor and internal system for the last two quarters, and compare it against what was forecast. The gap between those two numbers is the waste Harness is describing, and it is the fastest way for a CIO to make the business case for the governance investment the data says is overdue.
