The Linux Foundation announced the Tokenomics Foundation on August 4, with roughly thirty founding members and a straightforward premise: nobody can currently compare what two AI providers charge, because nobody agrees on what is being counted. The group will work alongside the FinOps Foundation and its FOCUS billing specification, extending a discipline built for cloud spend into token based pricing. Founding members include IBM, Oracle, SAP, ServiceNow, Broadcom, Lenovo, Accenture, JPMorganChase and BNY. If you have ever tried to forecast a model's cost across a quarter and given up, this is aimed squarely at that problem.
The short answer
The Linux Foundation launched the Tokenomics Foundation on August 4 to build vendor neutral standards for measuring the cost and return of AI. It works in close partnership with the FinOps Foundation, extending a discipline built for variable cloud spend into token based pricing. The roadmap includes AI value frameworks, vendor neutral cost models, standard cost to serve methods, and token cost telemetry inside the FOCUS billing specification. Founding members run from IBM and Oracle to JPMorganChase and BNY.
Ask anyone running AI in production what a feature costs per user per month and watch what happens. You get a number, then a pause, then a series of caveats about caching, about which model version was in use that week, and about whether the reasoning tokens are in that figure or not.
On August 4, the Linux Foundation announced the Tokenomics Foundation, backed by around thirty initial members, to do something about exactly that. The framing in the announcement is about return on investment and value measurement, which sounds like a finance department problem. Underneath it is an engineering problem, and it is one most of us have hit.
Why AI bills resist comparison
Everyone prices in tokens. That is where the agreement ends.
Tokenisers differ, so the same text is a different number of tokens depending on whose model you send it to. Cached input is priced differently from fresh input, and the rules for what stays cached differ. Reasoning output may be billed at the visible output rate or not counted the same way at all. Batch processing has separate pricing. Rate limit tiers change effective cost. None of this arrives in a shape you can normalise without writing provider specific code.
The result is that the simplest question a finance team asks, whether you are paying more or less per unit of useful work than the alternative, is genuinely hard to answer with confidence. And the question engineering teams ask, which is whether a prompt change made things cheaper, is answered by a spreadsheet that somebody maintains by hand.
The part worth watching is FOCUS
Most of the announced roadmap is definitional: shared language for AI value, frameworks for measuring business impact, vendor neutral models for total cost, standard methods for cost to serve, education and certification. Those are the outputs you expect from a new industry body, and they are worth roughly what the participation behind them is worth.
One item is different. The roadmap includes delivering token cost telemetry inside the FOCUS specification, and the group is described as working in close partnership with the FinOps Foundation.
FOCUS, the FinOps Open Cost and Usage Specification, is the open standard that lets billing exports from different cloud providers land in one schema. It is the reason a cost tool can show AWS, Azure and Google spend in the same table without a bespoke parser for each. It already exists, it already has adoption, and it already has tooling built on it.
If token usage and cost get expressed in FOCUS, every tool that already reads FOCUS gets AI spend without a new integration. That is a real deliverable with a testable outcome, as opposed to a framework document. It is also the item that requires the model providers to participate, which is the open question here.
What the membership list tells you
Around thirty organisations signed on at launch: IBM, Oracle, SAP, ServiceNow, Broadcom, Lenovo, Hitachi, Accenture, GoDaddy, JPMorganChase, BNY, and cost management specialists including Cast.ai, DoiT, Finout, Flexera and Kion, alongside channel players such as SHI and WWT.
Read that list carefully and a pattern appears. It is dominated by organisations that buy AI capacity, resell it, or sell tools for managing what it costs. Those are exactly the parties with an incentive to make pricing comparable. They are not the parties who set it.
That is not a criticism, it is a description of where the leverage sits. Standards efforts driven by buyers can work, and the FinOps precedent is encouraging: cloud providers did eventually adopt FOCUS, because enough of their customers were asking for it in the same words. But the timeline for that was years, not quarters.
How we would treat this today
As a bookmark rather than a plan.
Nothing shipped on August 4. What launched is an intent, a member list and a roadmap. If you are trying to get a grip on AI spend right now, the practical work is unchanged: instrument your own token usage at the application layer rather than relying on provider billing exports, tag requests by feature so you can attribute cost to something a product decision can act on, and keep your own record of which model version served which traffic, because provider billing will not reconstruct that for you later.
Do that and you are not waiting on a standard. You are also in a good position to adopt one, because the data you need to map into a schema will already exist.
The thing we will be watching for is a FOCUS revision that includes token cost telemetry, and model providers emitting it. That is the moment this stops being a press release and starts being infrastructure. Until then, it is a serious group of companies agreeing that the current situation is untenable, which is a reasonable place for a standards effort to start.
Sources and further reading
- Linux Foundation launches the Tokenomics Foundation to define the economics and ROI of AI value
- Linux Foundation announces the intent to launch the Tokenomics Foundation
- Linux Foundation announces Tokenomics Foundation, Phoronix, August 4, 2026
- Linux Foundation targets AI's cost-management problem with Tokenomics Foundation, CIO
- The FOCUS specification, FinOps Foundation
Frequently asked questions
What problem is this actually solving?
That you cannot compare two AI bills. Every provider prices in tokens, but what counts as a token differs by tokeniser, cached input is billed differently from fresh input, reasoning output may or may not be charged the same as visible output, batch pricing has its own rules, and none of it is expressed in a shape you can normalise. Put two invoices side by side and you cannot answer the simplest question a finance team will ask, which is whether you are paying more or less per unit of work than the alternative. The Tokenomics Foundation's stated goal is a vendor neutral way to express that, and standard methods for measuring cost to serve.
What is FOCUS and why does it keep coming up?
FOCUS is the FinOps Open Cost and Usage Specification, an open standard for billing data that lets cost and usage records from different cloud providers be normalised into one schema. It is the reason a multi cloud cost tool can put AWS, Azure and Google spend into the same table without a bespoke parser per vendor. The Tokenomics Foundation's roadmap includes delivering token cost telemetry inside FOCUS, which is the concrete deliverable to watch. If token usage lands in FOCUS, existing FinOps tooling gets AI spend for close to free, rather than everyone building a parallel stack.
Is this a real standards effort or a marketing consortium?
The honest answer is that it is too early to tell, and the membership list argues both ways. Thirty members including IBM, Oracle, SAP, ServiceNow, Broadcom, Lenovo and several large consultancies is a serious convening. It is also, notably, a list heavy on buyers and integrators. The Linux Foundation announcement is framed as an intent to launch, and the near term output is described as frameworks, shared language and certification programmes. Our test would be simple: does token cost telemetry actually land in the FOCUS specification, and do the model providers themselves emit it? Frameworks without provider participation are documentation.
Does this affect me if I am not running AI at enterprise scale?
Indirectly, and mostly through tooling. If a normalised way to express token cost gets into FOCUS, it eventually reaches the cost dashboards, the cloud billing exports and the open source FinOps tools that smaller teams also use. Today, if you want to know what a feature costs to run per user per month, you build that calculation yourself from provider specific billing exports and you rebuild it every time a provider changes its pricing model. A shared schema removes that work. That is a boring benefit, and it is the same reason standardised cloud billing data mattered a decade ago.
What does the roadmap actually commit to?
Five things, per the announcement: defining tokenomics and the value metrics for AI return on investment, creating AI Value Frameworks for measuring business impact, building vendor neutral models for the full cost of AI, establishing standard methods for measuring cost to serve, and delivering token cost telemetry in the FOCUS specification, plus practitioner education and certification. Four of those are definitional work. The fifth is a technical deliverable against an existing specification, which is why we would judge the effort by that one.