The initial corporate excitement surrounding autonomous agents and generative models is giving way to sober calculations across global boardrooms. According to a study titled The Cost of Intelligence published by the Capgemini Research Institute, 53 percent of global executives express growing concern over uncontrollable AI costs. In Germany, this apprehension is even more acute, with 63 percent of business leaders reporting anxiety over runaway expenses. The findings are based on a survey of 500 CXO-level executives across ten countries and 13 major industries. The data highlights that deploying enterprise AI without rigorous economic governance is rapidly straining corporate budgets.
To rein in ballooning expenses for API tokens and intensive computing infrastructure, technology leaders are preparing a notable strategic pivot. Nearly six in ten executives, specifically between 58 and 60 percent of respondents, plan to reallocate their IT budgets toward cheaper open-weight models over the next twelve months. By turning to architectures where model weights are openly accessible, enterprises seek greater flexibility to run and optimize systems on owned or tailored infrastructure. Corporate architects view this transition as a viable route to escape expensive vendor lock-in associated with proprietary frontier models. This planned reallocation signals a decisive turn in how enterprise IT procures intelligence.
However, Capgemini's findings reveal that the financial challenges run far deeper than token pricing alone. Only about one third of surveyed executives, precisely 33 percent, are currently able to quantify the indirect costs generated by their AI initiatives. These hidden expenditures encompass essential human-in-the-loop review cycles, laborious data cleansing routines, and ongoing governance overhead. Because these friction costs are rarely factored into initial pilot estimates, overall operational expenditure often escalates unexpectedly during wider deployment.
The lack of financial clarity is particularly evident when evaluating real return on investment. Only 20 percent of enterprises calculate what Capgemini terms the AI Yield, which measures the actual net business value remaining after deducting all verification, error, and remediation costs. For the remaining 80 percent of companies, the tangible economic return on their deployments remains largely a matter of guesswork. Consequently, merely 56 percent of executives feel confident in producing reliable cost forecasts when attempting to scale AI across organizational boundaries.
Mounting financial pressure is effectively ending an era of unchecked experimentation, forcing corporate leadership to treat AI deployments like any other capital investment. Without a rigorous framework that accounts for ongoing human supervision and data maintenance, projects risk eroding their projected business case. The coming twelve months will serve as a key proving ground to determine whether transitioning to open-weight models delivers the anticipated budget relief. Ultimately, sustained success in enterprise AI will depend less on raw model capabilities and far more on strict financial discipline.

