AI Unit Costs Are Falling. AI Spending Is Rising.
The price of each AI transaction has dropped sharply, but usage has grown faster, and most finance teams see the cost after the bill arrives.
The cost of running AI does not behave like a software license. The price per unit of AI computing has fallen sharply over the past two years. Total AI spending has risen faster, because AI agents and reasoning models consume far more computing per task, and because use spreads across the company once it works.
Most of that spending is now in business-unit budgets rather than in the IT budget. Many companies review it from vendor invoices after the month closes. By the time finance sees an overrun, the usage that caused it has run for weeks.
The P&L is affected in three places: operating expense variances in the business units, margin on products and services that use AI, and vendor contracts with consumption pricing and minimum commitments.
Companies managing this well treat AI as a variable operating cost with unit economics: the cost per transaction, per customer served or per document processed, compared with the value of the work. They budget by application, monitor usage close to real time and route routine tasks to lower-cost models.
Three actions for the CEO:
1. Put AI cost in the P&L by use case. Make each business leader accountable for the AI cost and the result in their area.
2. Set unit-cost targets and real-time alerts. Replace monthly invoice review with usage monitoring and limits.
3. Review AI contracts prior to renewal. Test commitments, pricing tiers and data terms for flexibility.
What does one AI-completed transaction cost your company today, and what was it worth?
Subscribe to the GCCM Insights newsletter: https://www.linkedin.com/newsletters/gccm-insights-7232110651324739585/
