What is the Tokenomics Foundation?
Published 15 August 2026
The Tokenomics Foundation is an open governance body hosted by the Linux Foundation, launched on 4 August 2026 with 29 founding members including JPMorgan Chase and IBM. Announced alongside the FinOps Foundation at FinOps X 2026, its remit is the thing nobody owns today: a vendor-neutral, shared definition of what enterprise AI actually costs and what that cost is worth.
Why a standards body, and why now
Token billing arrived before the accounting for it did. Every provider meters differently. No two invoices use the same units. A "token" means one thing on an OpenAI statement, something adjacent on an Anthropic one, and something else again inside a Bedrock or Azure line item. There is no agreed answer to a question every finance team now asks out loud: what does a token cost this business, and what did it buy.
The gap stopped being academic in the first half of 2026. Companies reported being 3x over their entire 2026 token budget by May. The most-cited example is Uber, which rolled Claude Code out to roughly 5,000 engineers in December 2025; by April the company's whole annual AI budget was spent, with monthly per-engineer costs landing somewhere between $500 and $2,000. Nothing there was a failure of engineering. It was a failure of measurement: nobody had a unit that could carry the forecast.
That is the shape of problem standards bodies exist for. When a cost is new, every vendor invents its own accounting, and buyers cannot compare, forecast, or negotiate until somebody neutral defines the units.
What it plans to standardize
The published roadmap runs to four workstreams:
| Workstream | What it means in practice |
|---|---|
| Tokenomics and value metrics | A shared vocabulary for AI ROI, so "cost per outcome" means the same thing across two companies and two vendors. |
| AI Value Frameworks | Structured ways to measure business impact, not just spend. The half of the equation most dashboards leave out. |
| Full-cost models | Vendor-neutral models for total AI cost, including the layers the token meter never sees. |
| Education and certification | A practitioner track, mirroring how the FinOps Foundation built its certification path for cloud. |
The point buried in the announcement: tokens are not the whole bill
The foundation's own framing is worth reading carefully. Tokens are described as the unit of measure that has become a pricing method - and as the most easily metered layer, accounting for a portion of AI spend rather than all of it.
That distinction is the single most useful thing in the launch. A token meter is easy to read, so it becomes the number people manage. But the token line sits on top of costs that never appear in it:
- GPU reservations and idle capacity held for latency headroom.
- Vector store writes, reads, and the reindexing that follows every embedding-model change.
- Eval harnesses, which routinely cost more than the feature they grade.
- Human review and annotation on anything customer-facing.
- Egress, orchestration, and the retry traffic that agents generate by design.
Manage only the metered layer and you optimize the part you can see while the rest drifts. This is the same trap as reading an invoice total without attribution underneath it: a real number that answers the wrong question.
How this fits with what already exists
Three efforts are converging on AI cost, and they operate at different layers. Confusing them wastes time.
- OpenTelemetry GenAI semantic conventions standardize the telemetry: what attributes a model call emits, named consistently across SDKs. See the conventions in detail. This is the wire format.
- FOCUS and the FinOps Foundation standardize the billing data: a common schema for provider cost exports, which AI spend is now being pulled into. See LLM FinOps standards.
- The Tokenomics Foundation targets the layer above both: what the measured thing means, and what it is worth.
You need all three, and they land in that order. Telemetry produces the data, billing schemas make it comparable, and value frameworks make it arguable in a budget meeting.
What this changes for practitioners in 2026
Honestly: not much, yet. This is an intent-to-launch turning into a launch, with 29 members and a roadmap. The first deliverables are frameworks and definitions, not tooling. Standards bodies publish on a multi-year cadence, and the FinOps Foundation's own cloud work took years to reach the point where a spec changed what vendors shipped.
What it does change is the direction of travel, and that is worth acting on early for two reasons:
- Vendor claims are about to get checkable. Once "cost per outcome" has a definition, a tool that reports it differently has to explain why. If you are evaluating AI cost tooling now, ask whether the vendor is a member and how it plans to map its metrics onto the frameworks. The answer is informative either way.
- The 29 founding members are your negotiating peers. A standards body backed by JPMorgan Chase and IBM is buyers organizing around a pricing model they did not design. Consistent units are the precondition for comparing providers, which is the precondition for moving workloads between them.
What to do before the standards land
The work that makes you standards-ready is the work that pays for itself now:
- Enforce a tagging contract at the gateway. Every request carries feature, environment, team, and customer. Everything a value framework will eventually ask for is derived from those tags. Start at cost attribution.
- Track consumption, not just cost. Unit prices are falling while bills rise; a budget denominated in dollars hides that, a budget denominated in tokens does not. See cost trends.
- Reconcile to the invoice monthly. An internal estimate nobody has checked against a real bill will not survive contact with a framework that assumes it was. See invoice reconciliation.
- Price the unmetered layers explicitly. Put GPU reservations, evals, vector stores, and human review in the same model as tokens, even roughly. Full-cost modelling is a named workstream; showing up with only a token number means starting over.
Teams already doing this will adopt the standard as a rename. Teams that are not will adopt it as a migration.
Related
- LLM FinOps standards - the existing schema work this builds on.
- OpenTelemetry GenAI conventions - the telemetry layer underneath.
- What is LLM FinOps? - the discipline this sits inside.
- How to budget for AI spend - the practice the Uber case failed at.
- Agent economics - why coding agents blow through annual budgets in four months.
Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →
FAQ
What is the Tokenomics Foundation?
The Tokenomics Foundation is an open governance body hosted by the Linux Foundation, launched on 4 August 2026 with 29 founding members including JPMorgan Chase and IBM. It works alongside the FinOps Foundation to build vendor-neutral standards for measuring, allocating, and justifying enterprise AI spend.
Why was it created?
Because token billing arrived before the accounting for it did. Every provider meters differently, no two invoices use the same units, and there is no shared definition of what a token costs a business. Companies reported running 3x over their entire 2026 token budget by May.
What will the Tokenomics Foundation 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 education and certification for practitioners.
Is a token the same as AI cost?
No. Tokens are the most easily metered layer of AI spend, not the whole of it. GPU reservations, vector storage, egress, eval harnesses, and human review all sit outside the token meter. Treating the token bill as the AI bill understates real cost, often by a wide margin.
How does this relate to OpenTelemetry GenAI conventions?
They solve different halves of the same problem. OpenTelemetry GenAI semantic conventions standardize the telemetry emitted by a request. The Tokenomics Foundation is aimed at the layer above: what that telemetry means in business terms, and what a token is worth once it is measured.
What should we do before the standards ship?
Do not wait. Standards bodies publish on a multi-year cadence and the first deliverables are frameworks, not code. Build a tagging contract, reconcile to the invoice monthly, and track consumption alongside cost. Teams already doing that will adopt the standard as a rename; teams that are not will adopt it as a migration.