Quick Facts
- 59% of executives say AI spending ran significantly over budget in the past 12 months, averaging 24% above expectations.
- Two-thirds of organizations that exceeded AI budgets faced negative financial consequences, including reduced free cash flow.
- Just 12% of enterprises can track both the cost and verified business value of AI across their organization.
Nearly six in ten large enterprises spent more on AI than they planned last year, and most cannot explain what they got for it. The finding comes from a September 2026 survey of 500 CXO-level executives at organizations with annual revenue above $1 billion, conducted by the Capgemini Research Institute across 10 countries and 13 industries.
The report, titled The Cost of Intelligence: Governing AI Economics in the Agentic Enterprise, found that among organizations that exceeded their AI budgets, spending ran 24% above expectations on average. About one in five overshot by more than 30%.
The Scale of the Problem
Two-thirds of organizations that blew their AI budgets faced real financial damage. That included lower free cash flow and cuts to discretionary spending. Capgemini cited one unnamed company whose AI bill hit $500 million in a single month after it failed to set usage caps.
Concern among executives is rising. 53% of enterprise CXOs say they are worried by rising AI costs, up from 45% the year prior.
No One Can See the Full Bill
The visibility problem is as serious as the spending problem. Six in ten organizations say they lack a comprehensive view of AI expenses. Half said AI spending was approved without adequate visibility into the full costs.
Only 56% of executives said they were fully confident in their ability to forecast AI costs as adoption scales. Among organizations that could estimate spending coverage, they tracked 61% of AI expenditure systematically but could link only 45% to verified business value.
Just 12% say they can track most or all of AI's costs and its value across the enterprise. That means roughly 90% of large enterprises are operating without a clear picture of what their AI investments are returning.
Agentic AI Is Accelerating the Problem
The report points to agentic AI as a primary driver of cost spikes. These are AI systems that autonomously execute multi-step tasks, consuming far more compute than standard chat interactions. Data from AI gateway company OpenRouter showed agentic token usage grew 14 times in six months, while human usage increased just 2.8 times over the same period.
Agentic coding tasks consume roughly 1,000 times more tokens than a standard code chat. The same agent running the same task can vary by up to 30 times in cost depending on how it is designed.
Bernd Meurer, CTO and Field CTO for Europe at BT Group, put it plainly: "Token costs have fallen drastically. But with more usage, and depending on how you design the agent, costs can rise substantially."
That dynamic explains a wider pattern. Worldwide generative AI spending is on track to reach roughly $644 billion in 2025, up nearly 76% in one year. Token prices fell sharply over the same period, yet total spend kept climbing. Cheaper models expanded usage rather than cutting costs.
CFOs Are Taking the Wheel
Finance leaders are stepping in. CFOs now help approve major AI investments at 52% of organizations, up from 42% a year ago. The shift reflects mounting pressure on finance teams to demonstrate returns on AI spending before committing further capital.
When budgets ran over, most organizations tightened rather than rethought. More than half introduced stricter spending controls or reduced the scale of affected initiatives. Only one in five reconsidered which activities should be AI-led versus human-led.
The Measurement Gap
Capgemini introduces a metric it calls "AI Yield," which captures the verified value of an AI-assisted outcome after accounting for its full delivery cost. Just one in five organizations currently calculates it. And 57% of organizations cite demonstrating sufficient value as their biggest obstacle to further AI investment.
Volker Darius, CEO of Capgemini Invent, said organizations must "stop treating its cost as a technology or procurement issue and start managing AI as a portfolio of economic outcomes." He added that "cost discipline alone will not create a profitable AI estate, just as value creation without cost control will not be sustainable."
Most enterprises are still running AI like an experiment. The bills suggest otherwise.
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.
