The AI line item on this quarter’s cloud invoice doesn’t match anything in the forecast tab. Six months ago it was a rounding error. Now it’s the fastest-growing number in the budget, and finance, IT, and the vendor each have a different answer for who owns it.
What is token economics, and why is it suddenly a CFO problem?
Token economics, or “Tokenomics,” is the financial discipline of measuring, benchmarking, and governing what an enterprise pays per unit of AI inference (the token), instead of treating AI spend as an extension of software licensing. This year, the Linux Foundation formalized the discipline by launching the Tokenomics Foundation, with founding support from Google Cloud, IBM, Microsoft, SAP, Accenture, and JPMorganChase. A dozen of the world’s largest technology and financial institutions agreeing to standardize how AI cost gets measured is itself the signal: AI spend had gotten too complicated to track the old way.
Why AI costs don’t behave like software licensing
A seat count is predictable. A token count is not. It moves with every prompt, every agent reasoning step, and every tool call an application makes on a user’s behalf, and it can double inside a single product decision an engineering team makes without ever looping in finance. IDC research puts a number on how unevenly that cost lands: output tokens run 3 to 5 times more expensive than input tokens, and a multiagent pipeline that isn’t monitoring its input-to-output ratio can understate its true inference cost by 3x or more. Two enterprises already know what that miscalculation costs in dollars: one exhausted an entire year’s AI budget by April after rolling out a coding assistant at scale, and a hospital system logged more than $6 million in unplanned charges before its finance team traced the source.
Who actually owns the AI budget line?
The deeper problem isn’t the invoice. Only one in five CEOs have created a separate AI budget line in the first place, which means for most enterprises this spend is still buried inside a budget that was never built to isolate it. CFOs now influence more than two-thirds of enterprise technology investment decisions, while CIOs remain essential for the technical build. That’s two roles with converging influence and a mandate that hasn’t caught up. Nearly three in four organizations already name excessive AI spending a major risk to their future technology investment plans.
“Every dollar of wasted AI agent spend is a governance failure.
Eighteen months ago, that framing would have stayed inside IT. Now it’s pushing the ownership question into board-level conversations, next to the AI budget line itself.
What CFOs can do now: Building a full-stack token cost model
The starting point is visibility, not restriction. A full-stack token cost model — compute, storage, networking, power, cooling, and facilities, expressed per million tokens by workload — is the analytical foundation every AI infrastructure and purchasing decision now depends on. IDC research shows that routing tokens to right-sized models, rather than defaulting every workload to the largest frontier model available, cuts cost per token by 30 to 60 percent with no new hardware investment. That’s the shift worth pushing for this budget cycle: a defined per-million-token cost figure attached to every new AI request before it reaches sign-off.
- Ask engineering for input-to-output token ratios by workload before the next planning cycle closes.
- Require any new AI deployment request to show a per-million-token cost estimate, the same way a seat-based software request shows per-seat pricing.
- Put the ownership question on the agenda directly. CFOs demanding embedded governance, modular architecture, and real-time cost visibility from vendors are already ahead of most of the market.
What to expect new
Expect the CFO-CIO ownership question to formalize rather than resolve itself quietly: joint AI-FinOps governance structures, full-stack token cost modeling required before any new infrastructure commitment, and CFOs setting vendor selection criteria that used to belong to IT alone. The enterprises moving on this now are building that model before the FY27 budget cycle locks. Everyone else is having the ownership argument for the first time, mid-budget-season, with finance and IT starting from different numbers.
