There is now a standards body for token costs.
August 15, 2026
On 4 August 2026 the Linux Foundation launched the Tokenomics Foundation with 29 founding members, including JPMorgan Chase and IBM, announced alongside the FinOps Foundation at FinOps X. Its remit is the thing token pricing arrived without: a vendor-neutral definition of what enterprise AI costs and what that cost is worth. For a finance function, the launch itself changes nothing this quarter. What it signals is worth acting on now.
The problem it exists to fix
Every AI provider meters differently. Input, output, cached, and reasoning tokens price on separate scales, the ratios move whenever a model changes, and no two invoices express the same unit. The result is a cost line that finance owns and cannot compare to anything - not to another vendor, not to a peer company, not to last year.
That became expensive in the first half of 2026. Companies reported running 3x over their entire annual token budget by May. Uber gave Claude Code to roughly 5,000 engineers in December 2025 and had spent its full annual AI budget by April, at $500 to $2,000 per engineer per month. Those were not overspending stories so much as measurement stories: nobody had a unit that could carry a forecast.
What it plans to define
| Workstream | What finance gets out of it |
|---|---|
| Tokenomics and value metrics | A shared vocabulary for AI ROI, so cost per outcome means the same thing across two vendors and two companies. |
| AI Value Frameworks | A structured way to state business impact, not just spend - the half of the case a cost dashboard cannot make. |
| Full-cost models | Vendor-neutral models covering what the token meter never sees: GPU reservations, evals, vector stores, human review. |
| Education and certification | A hiring signal, mirroring how FinOps certification became one for cloud. |
The framing in the launch is worth repeating verbatim to anyone who thinks the token line is the AI bill: tokens are described as the most easily metered layer of AI spend, accounting for a portion of it, not all of it. Manage only what is easy to meter and the rest of the cost keeps moving unobserved.
Three things this changes for the finance function
Comparability, which is leverage. A standards body backed by JPMorgan Chase and IBM is large buyers organising around a pricing model they did not design. Consistent units are what make two providers comparable on price, and comparability is the entire basis of a renegotiation. Until then, provider quotes are structurally hard to score against each other. See AI vendor negotiation and provider arbitrage.
Auditability, eventually. A vendor-neutral full-cost model is the kind of reference an auditor eventually asks a policy to point at, particularly where AI costs are being capitalised. Finance functions that can show a documented, reconciled derivation today will map onto whatever ships. Ones working from provider invoices alone will be rebuilding. See AI capitalisation, capex vs opex.
Vendor claims become checkable. Once cost per outcome has a definition, a tool that reports it differently has to explain why. If you are buying AI cost tooling this year, ask whether the vendor is a member and how it intends to map its metrics onto the frameworks. Either answer tells you something.
What not to wait for
Standards bodies publish on a multi-year cadence, and the first deliverables here are frameworks and definitions rather than tooling. The FinOps Foundation's own cloud work took years to reach the point where a spec changed what vendors shipped. Nothing in this launch justifies deferring a decision.
It also arrives against a market that has not solved the underlying problem: in the State of FinOps 2026 survey of 1,192 practitioners, granular AI spend monitoring - tokens, requests, GPU utilization - was the single most-requested capability in the entire survey. A capability at the top of that list is one nobody has shipped. See State of FinOps 2026 for CFOs.
The work that makes you standards-ready
It is the same work that pays for itself now, which is the only reason to do it on this timeline:
- A tagging contract at the gateway. Feature, environment, team, customer on every model call. Every value framework that ships will be computed from something like these four fields, and a request made without them cannot be attributed afterwards.
- Monthly reconciliation to the provider invoice. An internal figure nobody has checked against a real bill will not survive a framework that assumes it was.
- Consumption budgets alongside dollar budgets. Unit prices keep falling while bills rise; only a token-denominated budget shows the volume growth doing the work. See falling prices, rising bills.
- The unmetered layers priced explicitly. GPU reservations, evals, vector stores, human review, in the same model as tokens even if roughly. Full-cost modelling is a named workstream; arriving with only a token number means starting over.
- One unit metric per workflow. Cost per resolved ticket, per merged PR, per sourced deal. This is the denominator every value framework needs. See proving AI ROI.
Emit the underlying telemetry using OpenTelemetry GenAI conventions rather than a private schema. Standards land on standards.
The honest read
A launch with 29 members and a roadmap is a direction, not a deliverable. But the direction is unambiguous: AI cost is being pulled from a vendor-defined number into a buyer-defined one, and the finance functions that already have their own reconciled derivation will adopt the standard as a rename. The ones that do not will adopt it as a migration.
Related
- State of FinOps 2026 for CFOs
- 73% of agentic AI projects went over budget
- Proving AI ROI
- How to build an LLM CFO function
- AI governance framework
FAQ
What is the Tokenomics Foundation?
An open governance body hosted by the Linux Foundation, launched 4 August 2026 with 29 founding members including JPMorgan Chase and IBM, working alongside the FinOps Foundation. Its remit is vendor-neutral standards for measuring, allocating, and justifying enterprise AI spend - the accounting layer that token pricing arrived without.
Why does a standards body matter to a CFO?
Because comparability is the precondition for negotiation and for audit. Without shared units, two AI cost figures from two vendors cannot be compared, an internal AI cost cannot be benchmarked against a peer, and no auditor can test how a capitalised AI cost was derived. Standards make all three possible.
What will it standardize?
Four workstreams: a shared definition of tokenomics and value metrics for AI ROI, AI Value Frameworks for measuring business impact, vendor-neutral models for the full cost of AI beyond the token line, and practitioner education and certification.
Should we wait for the standards before building AI cost reporting?
No. Standards bodies publish on a multi-year cadence and the first deliverables are frameworks, not tooling. The work that makes you standards-ready - a tagging contract, invoice reconciliation, one unit metric per workflow - is the work that pays for itself immediately.
Who founded it and why does the membership matter?
29 founding members including JPMorgan Chase and IBM. The membership matters because it signals large buyers organising around a pricing model they did not design. Consistent units are what make providers comparable, and comparability is what gives a buyer leverage.
Does this change how AI spend gets capitalised?
Not yet, but it is the direction. A vendor-neutral full-cost model is exactly the kind of reference an auditor eventually asks to see a policy tied to. Finance functions with a documented, reconciled derivation today will map onto it; ones working from provider invoices alone will not.