Nearly one in three business leaders struggles to understand the operational costs of AI as organizations deploy AI on a larger scale, according to a global study by KPMG.
According to a KPMG survey of 2,145 senior business leaders across twenty countries, 29 percent struggle with estimating and managing operational AI costs. A third also cite limited insight into the AI economy as a major barrier to the rollout of AI agents.
The challenge is growing as AI vendors increasingly shift from fixed subscriptions to a consumption-based model. Anthropic, OpenAI, and GitHub are increasingly billing for certain services based on actual usage. According to KPMG, companies must therefore learn to better predict, monitor, and manage the costs of their AI applications.
Investments face more critical scrutiny
Rising costs are causing organizations to adjust their AI plans more frequently. Nearly half of the surveyed companies have delayed or redesigned AI projects because the expected return no longer outweighed the costs. According to KPMG, this does not mean companies have less confidence in AI. Instead, organizations are choosing to concentrate their budgets on applications that deliver clear added value. As a result, cheaper AI models with comparable performance are rapidly gaining popularity.
At the same time, hyperscalers continue to invest heavily in AI infrastructure. Amazon is allocating approximately $200 billion for investments this year, primarily to build additional AI capacity in AWS data centers. Microsoft is expected to invest around $190 billion and is also establishing a new organization to help customers develop AI solutions faster. In addition to costs, AI governance remains a challenge. KPMG emphasizes that clear agreements are needed regarding who is responsible for AI decisions, how AI output is monitored, and who intervenes when systems make mistakes.