Capgemini finds AI budget overruns are straining large companies’ finances

AI budget overruns are reaching beyond technology departments: two-thirds of organizations that overspent reported financial consequences, including lower free cash flow or reduced discretionary spending. Capgemini Research Institute’s “The cost of intelligence,” released November 12, finds companies tightening controls while many still lack a complete picture of what their AI spending buys.

The research snapshot draws on a September 2026 survey of 500 senior executives. Participants worked at organizations with annual revenue above $1 billion, spanning 10 countries and 13 industries. That scope matters: the findings describe executives’ accounts of spending and outcomes at large organizations, rather than a financial audit or a measure of every company using AI.

Nearly three in five executives said AI expenditure had been significantly higher than expected or budgeted over the previous 12 months. Among organizations reporting overruns, spending averaged 24% above expectations or budget. CFO Dive reported that 21% of overspenders exceeded their budgets by more than 30%.

Knowing the bill is not knowing the return

Capgemini found that 60% of organizations lacked consistent, company-wide visibility into AI costs. Its spending-coverage findings reveal a second problem: even when organizations can estimate how much expenditure they track, connecting that money to verified business value is a separate task. The survey distinguishes the ability to account for spending from the ability to explain its results.

61%AI spending systematically tracked

Among organizations able to estimate expenditure coverage, 61% of AI spending could be systematically tracked.

45%AI spending linked to verified value

Among organizations able to estimate expenditure coverage, 45% of AI spending could be connected to verified business value.

Those percentages measure shares of spending, not shares of projects that succeeded or failed. Being unable to connect spending to verified value is not the same as proving that spending produced no benefit. The narrower finding is a measurement gap: organizations could systematically track more of their expenditure than they could connect to a verified business outcome.

The accounting challenge also extends beyond technology bills. Capgemini identifies human oversight and governance as indirect AI costs. Only one-quarter of CFOs said they were fully confident in estimating those expenses. Looking ahead, 56% of respondents were fully confident in their organizations’ ability to forecast AI costs as adoption scales.

Tighter controls, few stopped projects

Finance leaders are getting a larger role in approving major AI investments. CFOs participated in those decisions at 52% of organizations, up from 42% a year earlier. Business-unit leaders and chief AI or data officers also became more involved, according to the snapshot. Investment approval is becoming a more cross-functional responsibility.

When spending exceeded plans, organizations were more likely to tighten financial controls than reconsider delivery or stop an initiative. Among organizations reporting overruns, the responses included:

• 53% tightened budgets or financial controls, or reduced the scope of initiatives.
• 20% reconsidered which activities AI should handle rather than employees, or changed their AI tools, models or configurations.
• 5% paused or stopped an affected initiative.

Measure the outcome, including the human work

Capgemini’s recommended response goes beyond setting a lower spending ceiling. It calls for AI FinOps—financial management of AI spending—and outcome-based measures that make costs and value visible together. Clear accountability and spending guardrails are part of that approach. The snapshot also recommends rewarding measurable value rather than the amount of AI activity an organization generates.

Model selection and workflow design are another part of its prescription. Volker Darius, CEO of Capgemini Invent, urged organizations to measure cost, quality and business value together, choose the most appropriate model for each task, and account for the human oversight needed to deliver outcomes safely.