In the history of technology, we have never seen a line item explode this fast.
Goldman Sachs projects token consumption growing 24x by 2030. Gartner has worldwide AI spending up 47% this year alone. Neither of those is a forecast about adoption. They are forecasts about a bill, and most organizations have no agreed method for reading and controlling it.
On 4 August the Linux Foundation formed the Tokenomics Foundation with 30 member organizations. PointFive is a Premier member on the Governing Board. This post is about why that matters more than a membership announcement usually does.
AI cost is not cloud cost with different labels
The instinct across the industry has been to treat AI spend as a new tab in an existing FinOps practice. That instinct is wrong, and it is costing people.
Managing AI cost is a fundamentally different discipline. The data sets are different. The technology underneath is different. It scales at a rate cloud never did. And the stakeholders are different. The people who authorize AI spend are frequently not the people who authorize infrastructure spend, and neither group is reliably the one watching the meter.
The measurement gap shows up immediately. McKinsey's Enterprise AI FinOps survey puts only 20-25% of companies at a mature AI FinOps practice. So spending is compounding at 47% a year against a discipline three-quarters of the market has not built yet, using definitions nobody has agreed on.
That is the position most organizations are in. Two forecasts about a bill, 24x token growth and 47% spend growth, and no shared method for reading either one. You cannot benchmark against that. You cannot budget against it either.
Standards set by vendors are not standards
Here is the part I feel strongly about.
How you handle AI cost should not be dictated by vendors. The average practitioner should not have to abide by what one company decides to do, and then rebuild their entire measurement approach when they switch tools or when that company changes its mind.
Practitioners need to come together, decide how this problem gets solved, and set the standards that determine how the industry operates and how tools get built. That order matters. Standards first, tools second. Every vendor in this space, PointFive included, should be building against a definition it did not write alone.
That is what a neutral body is for. The Foundation's opening position makes the scope clear: tokenomics is about the value of all AI spending, not counting tokens. The roadmap follows from it. Definitions and value metrics for AI ROI, AI Value Frameworks that tie spend to business impact, vendor-neutral models for the full cost of AI, standard methods for cost to serve, token cost telemetry in the FOCUS specification, and the education and certification practitioners actually need.
None of that is work any single company should be doing by itself.
What is actually happening in the field
The year started with tokenmaxxing. Adopt as much as possible, spend as much as possible, worry later. Then the bills arrived, and the question changed from "how fast can we adopt this" to "how do we handle this without it running away from us."
The evidence backs up the mood. The same McKinsey survey found 93% of organizations surveyed had already exceeded their AI budget. Put that next to the maturity figure above and the picture is complete: almost everyone is over budget, and almost nobody has the practice in place to do anything structural about it.
What our own research keeps finding is that the waste is not where teams look for it. PointFive Labs found that 50-80% of a typical enterprise AI bill sits outside the in-platform spend controls teams are relying on. Outside Unity AI Gateway, outside Cortex, running direct against Anthropic, OpenAI or Bedrock, where no single console sees the whole picture. In one environment, a Bedrock workload was tracking 40-50% over budget for a single reason: high-complexity tasks were defaulting to a frontier model that a model 15x cheaper handled identically. No cost export would have surfaced that.
That is the argument for research as a first-class activity rather than a marketing byproduct. You cannot standardize what nobody has measured properly.
Nobody owns AI spend yet
Ask an enterprise who owns AI cost and the answer is unclear. There is an AI transformation leader. There are engineering leaders. There is a CIO. There is platform engineering and infrastructure. It is still genuinely unclear who owns what.
Within the next year, establishing who owns what and who is accountable for AI cost will be mandatory. Not a best practice. Mandatory. Without it there is no mechanism by which this stays under control.
We have seen what happens when ownership is explicit. Nubank runs one of the largest digital banks in the world, with more than 10,000 engineers and 45+ business units, each with its own Cost Champion. Separating price from usage and giving each an owner is what took resolution time on unattached EBS volumes from 210 days to 16, optimized 3,000 DynamoDB tables through one central platform team, and delivered 8-12% of total cloud spend saved, verified against the bill, not modeled. It reached 92% of business units.
That is a cloud example, because cloud is where the practice is mature enough to produce numbers like these. The point is the mechanism, and the mechanism transfers.
What PointFive owes this
Our role is not to define the standard on the community's behalf. It is to provide the tools practitioners need to solve this: control and governance over AI spend, and the ability to measure the ROI on AI investment, so teams can be confident there are no surprises and that the return is real.
AI has moved from experiment to infrastructure, and the industry is only beginning to build the discipline to account for what it consumes. The Foundation gives us a shared, open place to define how token spend should be measured and managed, so that efficiency becomes a standard every organization can reach rather than an advantage held by a few.
We are excited to be releasing a major AI research paper tomorrow, "Token Reduction is Not Cost Reduction." This is the first in what will be a series of research papers from our AI Lab that will help shape the direction of the industry, the Foundation, and our ability to provide the intelligence for organizations to actively make their AI and cloud run efficiently.
There is a lot of work to do.
The Tokenomics Foundation launched on 4 August 2026 with 30 founding member organizations. The first Tokenomicon + FinOps X joint event is in Amsterdam, 22 and 23 September 2026, at Muziekgebouw.
Sources: Linux Foundation launch announcement (Goldman Sachs and Gartner figures) · McKinsey Enterprise AI FinOps survey, May 2026 · PointFive Labs, July 2026 · Nubank, FinOps X 2026.