Nearly six in ten executives at large organisations say artificial intelligence expenditure has significantly exceeded expectations, as businesses struggle to forecast the full cost of deploying AI across their operations.
Research from Capgemini found that 59% of surveyed senior executives experienced substantial AI spending overruns during the previous 12 months. Among those affected, expenditure was 24% above budget or expectations on average, while 21% reported overruns exceeding 30%.
The findings are based on a September 2026 survey of 500 senior executives working for organisations with annual revenue above $1bn across ten countries, including the UK.
More than half of respondents, 53%, identified AI costs as a concern, while only 56% were fully confident in their ability to forecast expenditure as adoption increased.
The consequences extended beyond technology budgets. Two-thirds of organisations that overspent reported negative financial effects, including lower free cash flow or reductions in discretionary expenditure elsewhere.
Indirect costs complicate financial planning —
The research identifies limited visibility of the expenditure associated with running AI systems safely and effectively.
Direct costs can include computing infrastructure, access to AI models, and charges linked to the volume of information processed. Organisations also incur expenditure through system integration, employee oversight, correcting unreliable outputs, and governance.
Only around one-third of executives said they could confidently estimate these indirect costs, while half reported that AI expenditure was being approved without adequate visibility of its full financial requirements.
The problem becomes more pronounced as organisations expand AI from individual applications into connected business processes. A workflow involving several automated steps may generate multiple charges and still require employees to check or correct the result.
Volker Darius, chief executive of Capgemini Invent, said: “Cost discipline alone will not create a profitable AI estate, just as value creation without cost control will not be sustainable.”
The findings suggest that financial oversight is extending beyond technology departments, with chief financial officers becoming more closely involved in approving significant AI investments.
Businesses reconsider models and controls —
Among organisations responding to unexpected AI expenditure, 53% introduced tighter controls, reduced the scope of affected initiatives, or restricted spending more broadly.
Only one in five reconsidered the underlying technology or processes, including whether particular tasks should be performed by AI or employees, or whether different models and configurations could reduce costs.
More than three-fifths of executives expected their organisations to adopt a strategy using several AI models over the following year, selecting systems according to their capabilities, operating costs, response times, and risks.
Smaller or more specialised models may be sufficient for relatively straightforward work, allowing more advanced systems to be reserved for tasks requiring greater processing capabilities.
However, cheaper model access does not guarantee lower total costs if poor output quality increases the amount of work employees must repeat or check.
Measuring the value of completed work —
Capgemini also identified weaknesses in how organisations assess returns on their AI investments.
Although two-thirds of executives believed they could verify the value generated by AI, most organisations could not directly connect expenditure with successful business outcomes.
Some monitored usage and technical costs without measuring completed work, while others maintained separate financial and performance records that were difficult to reconcile.
Only one in five organisations calculated what Capgemini calls AI Yield — the verified value of an AI-assisted outcome after accounting for the full cost of delivering it safely.
The measure considers successful task completion alongside factors such as rework, escalation, and employee involvement in checking results.
For organisations expanding AI investment, the research highlights the importance of distinguishing between reducing the cost of individual operations and establishing whether an entire process delivers a worthwhile financial return.
