The Tokenomics Foundation Wants to Standardize AI Tokens. That Should Worry You.

This week, a press release announced the creation of the Tokenomics Foundation, an organization dedicated to standardizing AI token measurement. It promises to bring order to how tokens are counted, billed, and compared across AI systems. It says its goal is to help enterprises with cost management and AI investment strategy. It explicitly claims to have nothing to do with cryptocurrency. It provides no website, no membership list, no draft standard, and no technical documentation. I have spent more than a decade auditing cryptographic and governance claims. When an initiative calls itself Tokenomics and scoffs at the crypto connection, I start ignoring the word "no" and looking for the hidden file. The pain point is real. The AI token is not a fixed unit. Modern language models use different tokenizer architectures: OpenAI's BPE, Google's SentencePiece, Anthropic's byte-level tokenizers. For the same English sentence, each returns a different token count. Add multimodal models that break images into patches and audio into frames, and the token becomes a wildly moving target. Enterprise buyers trying to compare GPT-4, Claude, and Gemini must rely on the exact unit of measure that the seller controls. The result is a classic information asymmetry: you pay for "tokens," but nobody can tell you what a token is. That is precisely the kind of problem a proper standards body should tackle. I have rebuilt reward systems and voting protocols for DAOs, and I have learned that measurement is the first pillar of trust. You cannot have accountability without a common unit. So I sympathize with the foundation's mission. I also know how often "standards" in this industry are nothing more than beautifully designed landing pages. Let's get into the technical core. Standardizing AI token measurement is not a single problem. It is at least five problems. First, text tokenization boundaries: how do you define a token for different scripts, languages, and subword algorithms? Second, API billing units: do you count prompt tokens, completion tokens, cached tokens, or all of them with different weights? Third, multimodal conversion: how many visual patches equal one text token? Fourth, generation throughput: how do you measure tokens per second when hardware and batching differ? Fifth, cost accounting metadata: what fields must an invoice show so a third party can verify it? The Tokenomics Foundation has not told us which of these it wants to solve. Without that scope definition, there is no standard—only a hashtag. Here is the uncomfortable economics. Standardization has a customer: enterprise procurement teams, FinOps analysts, and investors. And it has a reluctant supplier: AI API vendors. For a model provider, keeping token accounting opaque is not an oversight; it is a feature. Opaque token math makes comparison shopping harder, reduces price pressure, and allows creative bundling. If the Tokenomics Foundation genuinely wants vendor neutrality, it will need more than good intentions. It will need either regulatory compulsion or buyer collective action. Neither of those appears to be on the table. I have seen this movie before. In 2017, when ICO whitepapers were abundant, I audited over fifty projects and wrote "The Ethics of Empty Vests" to warn retail users about initiatives with grand names and no load-bearing architecture. Some of those projects had websites, token sales, and even community "governance"—but no code, no test suite, no genuine limitation on the founders' power. You can have all the ceremony of a standard without any of the substance. Tokenomics Foundation perfectly fits that silhouette: an announcement, a logo, a mission, and a wall of zeroes where the references should be. Counter-intuitively, the lack of detail might be a deliberate strategy. If the founders are ex-Web3—and the name certainly suggests a crypto economics pedigree—they know that early standardization conversations are vulnerable to vendor capture. Maybe they are keeping silent to nurture a coalition before publishing something that the hyperscalers can veto. The repeated assertion that the foundation has "no relation to cryptography" could be a way to avoid the stigma of a crash-prone industry. But it is also a red flag. If there is truly no crypto influence, why choose a term coined by Vitalik Buterin? "Tokenomics" is not a neutral engineering phrase; it is a Web3 intellectual export. Claiming otherwise is like saying your new coffee shop has no connection to Italy while naming it "Caffe Milano." This is where I shift from technical to ethical. The real danger is not that this foundation will fail. The real danger is that it will succeed as a "weak standard." History is full of standards written by the very companies they were meant to constrain. A standard that merely gives the appearance of neutrality—while letting the largest vendors define the measurement behind closed doors—may be worse than none. It would create a false sense of comparability, enabling procurement teams to make decisions with the same flawed data they have today, but now with a Foundation stamp of approval. That is the cryptoeconomics trap: you mint a new token of trust, and it is immediately debased. If we are serious about standardized AI measurement, we should demand a minimal set of conditions. First, a fully open reference implementation: not a paper, but code that anyone can run. Second, a diverse governance body with representation from enterprise buyers, independent auditors, and open source communities—not just model vendors. Third, a published conformance test suite that is tested against real-world API outputs. Fourth, a clear mechanism for disputes. Fifth, a commitment to audit the auditors. The Tokenomics Foundation has published none of these. In a bull market, where capital chases every narrative, "we are making a standard" is a compelling pitch. But code is law, and people are the soul. Without a human-centric governance process, the measurement code will encode the interests of whoever writes it. I am not writing this to dismiss the initiative. I am writing this because I have spent my career in community building and know how frequently "decentralized governance" masks centralized control. The Tokenomics Foundation has an opportunity to go beyond crypto's bad habits and create something that operates on genuinely democratic principles. To do that, it must show its open governance from day one. It must publish its membership, its funding, its methodology, and its conflict of interest policy. If you want to govern the exit, you have to govern the entrance—and the entrance for this foundation is its own foundation. The market is watching. AI infrastructure is the hottest asset class no one can still price. A unified token standard would not only reduce enterprise friction; it would also enable disaggregated price discovery, more transparent capital allocation, and better return on investment for AI projects. That is a future worth fighting for. But it will not underwrite a dream. Give us a reference implementation. Give us a tokenizer benchmark. Give us a list of independent advisors. Show me the code of the measurement, and I will help you audit it. Listen more than you code? No. In this case, code more than you claim. In my audit work, I never trust a protocol that cannot be forked. Tokenomics Foundation is trying to define the fundamental unit of AI's economy. It should take the fork in the road that leads to transparency, not to a private club of metric custodians. The only standard that matters is the one that you can scrutinize, benchmark, and challenge without asking someone's permission. If the foundation gets that right, it might become what it says it is—a referee. If it doesn't, it will become something we already know well: an empty vest wearing a three-piece suit.