The foundational assumption in every FinOps maturity model is that cost control improves as practices mature. Better tooling, clearer accountability, tighter governance: these are supposed to prevent budget overruns. For AI budgets, the data says the opposite.
DoIT’s 2026 AI Spending Survey, conducted by Sapio Research across 500 finance leaders at organizations with 1,000+ employees, found that 79% of enterprises experienced AI cost overruns in the past twelve months. The surprising finding is buried in the segmentation: organizations with mature FinOps practices overran at an 89% rate, with a mean overspend of 30.9%. Early-stage FinOps organizations overran at 69%, with a mean overspend of 16.1%.
Maturity, in other words, correlated with larger overruns, not smaller ones.
The Maturity Paradox Isn’t About Bad Practices
The explanation is straightforward once you look at what mature FinOps teams are doing. They are the organizations spending aggressively on AI: running inference at scale, deploying agents across business functions, embedding AI into production workflows. Early-stage FinOps organizations are still in pilot mode with limited blast radius.
An organization spending $50,000 on AI pilots can absorb a 16% overrun without anyone noticing. An organization spending $4 million on production AI workloads will feel a 31% overrun in the quarterly earnings call.
This tracks with Gartner’s finding that 58% of enterprises exceeded AI infrastructure budget estimates by 40% or more, primarily because they underestimated compute requirements. The organizations that underestimate the most are the ones running real workloads, not the ones still evaluating vendors.
Three Cost Categories That FinOps Dashboards Miss
Traditional FinOps was built for infrastructure: compute, storage, networking. AI spending operates across cost categories that most FinOps tooling doesn’t capture.
Consumption-based SaaS with AI surcharges. Microsoft Copilot runs $66 to $87 per user per month when combined with M365 licensing. Salesforce Agentforce uses conversation-based pricing that creates unpredictable spikes. Token-based APIs range from $0.002 to $0.40 per interaction depending on the model. None of these show up in a cloud bill. They land in SaaS procurement, departmental credit cards, or employee expense reports.
Shadow AI. According to WitnessAI, 68% of employees access generative AI through personal accounts rather than company platforms, and 57% have entered confidential data into publicly available AI tools. The Beri Institute estimates that shadow AI tools purchased through employee expenses can reach $300,000 annually in a 2,000-person organization. These costs are invisible to FinOps teams focused on AWS, Azure, and GCP consoles.
Overlapping tool redundancy. Organizations commonly run three to five overlapping AI tools for identical use cases. One enterprise documented $492,000 in annual costs for three tools performing the same function. Traditional cloud waste detection (idle instances, unattached volumes) doesn’t flag SaaS redundancy.
The Accountability Vacuum
The DoIT survey reveals a structural problem that no dashboard can fix: nobody owns AI spend. Technology leadership claims 55% accountability for AI costs. Finance claims 53%. At the operational level, there is no clear single owner.
This matters because AI spending crosses every traditional boundary. A product team spinning up API calls to Claude or GPT-4 isn’t provisioning infrastructure. They’re making procurement decisions with an API key. When 83% of finance leaders expect quantifiable AI returns within 12 months, but only 15% can calculate AI ROI without significant bottlenecks, the accountability gap becomes a forecasting gap.
The C-suite perception makes it worse. DoIT found a 33-point gap between how C-suite executives and operational managers assess their organization’s FinOps maturity: 93% versus 60%. Leadership believes the problem is solved. The people running the numbers know it isn’t.
The Annual Budget Model Can’t Track Monthly Consumption
The deeper issue is that annual IT budgeting, designed for predictable infrastructure contracts, breaks down completely for consumption-based AI.
Enterprise AI spending jumped 108% year over year in 2026, hitting an average of $1.2 million per organization on AI-native applications alone. BCG’s AI Radar Survey found companies plan to spend 1.7% of revenue on AI in 2026, more than doubling from 0.8% in 2025. Technology companies are already at 2.1%.
When spending doubles annually, any forecast based on last year’s run rate is wrong by definition. The Mavvrik/BenchmarkIT study found that 80% to 85% of enterprises miss AI infrastructure forecasts by more than 25%. The typical response is to add a buffer to next year’s budget, which misses the point entirely: the problem isn’t the buffer size; it’s that annual cycles can’t track consumption-based pricing that shifts monthly.
Practitioners who have managed telecom billing know this pattern. Twenty years ago, mobile data pricing was equally unpredictable. Organizations solved it by moving to real-time usage monitoring and monthly reconciliation, not by padding annual budgets. AI cost management needs the same operational shift.
What the 21% Who Stayed on Budget Did Differently
The DoIT survey doesn’t just document the problem. It identifies what separates the 21% of organizations that stayed within AI budget.
The standout variable: unit economics tracking. Only 26% of finance leaders have implemented AI unit economics (cost per inference, cost per agent transaction, cost per AI-assisted workflow). But the organizations that track at the unit level overrun their budgets at significantly lower rates. This aligns with the tokenomics approach that the FinOps Foundation is now promoting.
The C-suite recognizes the gap. Eighty percent of C-suite respondents expect unit economics implementation within six months. Bridging the distance between intent and execution is the operating constraint.
Three operational changes move the needle, based on the survey data and what practitioners report working:
Monthly AI spend reconciliation. Not quarterly reviews. Not annual budgets with a true-up. Monthly line-item reconciliation of cloud AI services, SaaS AI features, API consumption, and shadow AI estimates. The organizations in the DoIT sample that review AI spend monthly had a mean overspend 12 points lower than those on quarterly cycles.
Cross-functional AI cost ownership. Joint CIO/CFO ownership of AI budgets correlates with higher realized ROI, according to Deloitte’s December 2025 enterprise AI survey. The accountability vacuum closes when both functions own the same number.
Approved tool registries with usage caps. The shadow AI problem doesn’t resolve through policy alone. Organizations that maintain approved AI tool registries with per-team usage caps report visibility into 80%+ of their AI spend. Organizations without registries report visibility into less than 25%.
The Spending Won’t Slow Down
Global enterprise AI spending will hit $407 billion in 2026, up 34.8% from $302 billion in 2025, according to IDC. Generative AI alone accounts for $127 billion of that total. And 42% of CFOs plan to increase AI budgets by 30% or more over the next two years.
The organizations that figure out AI cost governance in 2026 will have a structural advantage over those still debating who owns the budget. For FinOps practitioners, the lesson from this data is uncomfortable but clear: maturity in cloud cost management does not transfer automatically to AI cost management. The tools are different, the pricing models are different, and the accountability structures need rebuilding. Treating AI spend as another line item inside the existing cloud FinOps practice guarantees the 31% overrun.