The First Rule of Tokenomics Is: Don't Count Tokens

Measuring AI expenditures purely by token consumption provides an incomplete picture, as tokens account for only 25% of total costs. The remaining 75% comes from critical underlying infrastructure, including compute, memory, database services, and operational overhead. Organizations must shift toward evaluating total infrastructure expenses alongside business outcomes through platforms like ValueOps AIPM to determine true ROI.

Key Takeaways

Direct token consumption accounts for only 25% of an organization's total AI expenditure.

Unseen infrastructure, storage, and databases make up 75% of actual AI running costs.

Measuring overall outcomes and full infrastructure costs provides the true value of AI investments.

It sounds like a paradox for a discipline literally named after tokens. But the more conversations we have across the industry, the more this idea keeps surfacing: counting tokens is the wrong instinct. Token costs vary too much — by model, by provider, by workload — to be a stable unit of measurement. If you're managing AI spend by watching the token counter, you're watching the wrong number. What you actually need to measure is outcomes.

Here's the thing, though: even "measure outcomes instead of tokens" undersells how big this problem actually is. Tokens themselves are only a fraction of what AI actually costs.

The 25/75 Problem

Companies are increasingly estimating that only about 25% of their total AI cost is actual direct model consumption — the tokens. The other 75% is everything running underneath and around it: GPU and supporting infrastructure, compute, memory, database, and all the services and labor keeping it running. If you're only watching your token bill, you're missing three-quarters of what AI actually costs your organization.

Think about what happens every time someone uses an AI tool. There's the prompt going in and the answer coming out — that's the part everyone thinks of as "the AI." But before that prompt reaches a model, something has to collect it, format it, and decide what context or data it needs. After the model responds, something has to package that response, store it, and ideally reuse it rather than regenerating it from scratch every time. None of that is token cost. All of it is infrastructure cost, and it's been sitting there, mostly untracked, for years — quietly absorbed into general IT overhead because nobody bothered to trace it back to a specific AI workload.

That worked fine when AI was a small part of the bill. It doesn't work anymore. As AI-driven processing becomes a bigger share of what it costs to run modern applications — in a lot of cases, bigger than the cost of the people maintaining them — that "everything else" stops being background noise and starts being the majority of the story.

Why Token Visibility Alone Isn't the Answer

This is exactly why the conversation can't stop at "let's get better visibility into token spend." Token visibility is necessary. It's nowhere close to sufficient. If you can see your token bill perfectly and still have no idea what your GPUs, your storage, your databases, and the people supporting all of it are actually costing you in service of that same AI workload, you still can't answer the only question that matters: what did this actually cost, and was it worth it?

Where ValueOps Advanced Infrastructure Portfolio Management Comes In

This is the part of the story that doesn't get talked about enough, because Tokenomics is the hot topic right now — and for good reason, it's the most visible, most urgent-feeling piece of the puzzle. But it's a capability that sits inside something bigger: ValueOps Advanced Infrastructure Portfolio Management (AIPM).

AIPM is the overarching platform that aggregates data across your hybrid estate — public cloud, private cloud, on-premises, multicloud — and connects it back to the work it's supporting: the applications, the projects, the strategic objectives your organization actually cares about. Tokenomics is the capability inside AIPM built specifically for AI and token consumption. But the same underlying platform is already built to see the other 75%: the compute, the storage, the databases, the infrastructure that AI runs on top of, alongside everything else your hybrid estate is running.

That matters, because it means you don't have to choose between understanding your AI spend and understanding your infrastructure spend. They're the same question, answered by the same platform, tied back to the same work, outcomes, and strategic objectives — not two separate tools giving you two separate, incomplete pictures.

The First Rule, Revisited

So yes — don't count tokens. But don't stop at "measure outcomes" as an abstract idea, either. Measure the whole cost of getting there: the tokens, the infrastructure underneath them, and the work they were both in service of. That's what ValueOps AIPM, with Tokenomics built in, is built to show you.

Frequently Asked Questions

How should AI costs be measured?

Tokens account for roughly 25% of overall costs, while the remaining 75% comes from supporting infrastructure like compute, memory, and databases. True AI cost measurement requires tracking direct token usage alongside underlying infrastructure expenses and business outcomes using platforms like ValueOps AIPM.

What is the 25/75 problem?

Only about 25% of total AI spending goes toward direct token usage, while the other 75% is absorbed by infrastructure, compute, memory, databases, and supporting labor.

Why are token measurements insufficient?

Token metrics vary by model, provider, and workload, making them an unstable unit that overlooks three-quarters of the underlying infrastructure costs.

What does ValueOps AIPM track?

ValueOps AIPM connects token consumption, hybrid estate infrastructure costs, and application workloads to overall business outcomes and strategic goals.