Most companies investing in AI have very little idea what they are actually spending. According to an unreleased KPMG survey reported by the Wall Street Journal, only 26% of companies say they have a comprehensive view of their AI costs. Half have partial visibility, and 22% either have no visibility at all or only find out what they owe once the bill arrives. In short, only one in four companies can meaningfully answer the question: what is AI costing us?
Steve Chase, KPMG’s global head of AI, told the WSJ that the firm is working with companies that have exhausted their entire annual token and cloud computing budgets within a matter of months. One client saw its token usage jump sixfold. “It’s a new resource that needs to be managed that didn’t exist quite that way, and we’re seeing exponential growth,” Chase told WSJ.
AI providers including Anthropic, OpenAI, Microsoft and Salesforce charge enterprise customers, at least in part, by usage. Billing is typically tied to token consumption, the units of text that AI models process when handling a query or generating a response. This model can better align revenue with actual costs for providers and in theory offers more flexibility to customers than flat subscription pricing. In practice, token usage is difficult to predict, especially as more employees and systems tap into AI tools simultaneously and many finance teams are not yet equipped to track it in real time.
The survey findings are a useful reality check on the pace of enterprise AI adoption. Companies are committing serious budget to AI without yet having the visibility to know whether that investment is running efficiently or ballooning out of control. Building that financial infrastructure may turn out to be as important as the AI deployments themselves.