Council Post: Hidden Token Costs Make AI Prohibitively Expensive, But FinOps Can Trim The Bill

Token costs are making AI prohibitively expensive for many enterprises, but FinOps can help rein in spending, argues Brian Wilson, CEO of Kion, in a Council Post. He says that while inference costs for GPT-3.5-level capabilities fell dramatically—from $20 to 7 cents per million tokens between 2022 and 2024, according to Stanford’s AI Index 2025—overall enterprise spend rose sharply in 2025 and many organizations exceed budgets. Wilson attributes the disconnect to how tokens accumulate as agents run autonomously, retry queries, and expand usage across teams like sales, marketing, finance, and engineering. He argues organizations need granular visibility and governance to justify spend and maximize value.