Overview
- A KPMG survey found only about one in four companies have a full view of their AI costs, leaving many organizations blind to rapid token consumption and surprise invoices.
- Vendors have shifted to token-based, metered pricing so customers pay per inference unit called a token, which makes monthly bills variable and hard to predict.
- Agentic AI systems multiply usage because one human task can trigger many automated calls to models, which can drive token use far faster than per-token prices fall.
- Several large firms exhausted budgets or imposed limits after heavy use: Uber set employee caps, Microsoft cut some third-party licences and Amazon ended an internal token leaderboard.
- Executives are responding by linking AI spend to business outcomes, creating AI FinOps teams, routing simple requests to cheaper models and testing private or edge inference to regain cost control.