--- title: "Cold Water on 'Token Trading' — Wang Qinglan on the NDA's High-Quality Data Set Initiative" author: "DCC Editorial" published: 2026-04-24T01:00:00.000Z url: https://datacompliancechina.com/posts/qinglan-token-trading-cold-water/ description: "In March 2026, the National Data Administration released the *Implementation Plan for Promoting High-Quality Industry Data Set Construction (Draft for Public Consultation)*, which explores a 'token (词元) based value system' and 'token trading as a new transaction mode' for high-quality data sets. The Chinese AI policy community immediately heralded the move as 'revolutionizing data trading.' Wang Qinglan pours cold water: token is a measuring unit, not a magic transformer. AI tokens are not crypto tokens. The bottleneck in China's data-element market isn't measurement — it's supply, rights clarity, compliance cost, and data silos." tags: ["tokens", "ai-training-data", "data-trading", "national-data-administration", "commentary"] laws_cited: ["data-foundation-system-opinions", "common-data-terms-batch-1", "common-data-terms-batch-2"] domains: ["data-economy", "ai-governance"] account: "qinglan-data" original_title: "给\"词元交易\"泼一盆冷水" original_author: "王青兰 (Wang Qinglan)" original_publication: "青兰数据观察" original_url: "https://mp.weixin.qq.com/s/0Nbcam7GbrYx8d31JmTGGA" source_language: "zh" --- > *Editor's Note — DCC.* > > In March 2026, the National Data Administration released the > *Implementation Plan for Promoting High-Quality Industry Data Set > Construction (Draft for Public Consultation)*, introducing "token > (词元) based value system" and "token trading" as exploratory concepts. > Chinese AI policy commentary immediately escalated: token trading > would "revolutionize" data trading, "subvert digital economy rules," > "reconstruct value systems." Wang Qinglan — head of compliance at a > Chinese data exchange and one of the closer observers of the > data-element market — pushed back. This brief tracks her argument. > DCC's framing emphasizes the policy context for overseas readers who > see this concept appearing in Chinese policy documents and need to > understand what it actually means. ## First — AI tokens are not crypto tokens The piece opens with a disambiguation that overseas readers in particular need. The word "token" in Chinese AI policy refers to the **AI processing unit** — the unit by which large language models segment text for processing. *Not* the crypto-asset *Token* that can be traded, hoarded, and speculated on. The official Chinese rendering, **词元** (literally, "word element"), was formally adopted by the National Data Administration in March 2026. Wang's clean separation: - **AI tokens (词元)** = a measuring unit for AI compute consumption. Analogous to *kWh* for electricity, *gallons* for water, or *gigabytes* for data transmission. AI service providers charge by token consumed. - **Crypto tokens (通证)** = digital assets. Can be bought, sold, and speculated on. Subject to crypto-market dynamics — completely unrelated to AI. If a reader confuses the two, the entire token trading discussion becomes incoherent. Wang's first move is to insist on the distinction. ## The NDA's March 2026 policy move The *Implementation Plan for Promoting High-Quality Industry Data Set Construction (Draft for Public Consultation)* (March 2026) introduces two notable phrases: > *"Explore a token (词元) based value system."* > > *"Explore new transaction modes such as token trading, building a quantifiable, priceable data set value system based on tokens."* The doctrinal move: extending the role of tokens from *AI compute pricing* (output side) to *data set pricing* (input side). The reasoning: if AI consumes data measured in tokens, perhaps the data sets it consumes should also be priced in tokens. Wang's analogy: "previously, 'kilograms' was just a unit the mill used to charge for processing. Now someone proposes using kilograms to *sell the wheat itself.*" In principle, the move is reasonable. In practice, it conflates different problems. ## The scale metaphor Wang's central conceptual move: **the token is a scale, not a transformer**. The scale weighs things. It can weigh wheat (raw data), flour (processed data resources), and cakes (refined data products). But: - A scale doesn't grow more wheat. - A scale doesn't change the quality of the flour. - A scale doesn't make cakes appear from the oven. It just measures. This is the lens Wang applies to the over-claims about token trading. ## Three over-claims and their flaws ### Over-claim 1 — "Token trading will revolutionize data trading" The claim: token-based pricing creates a unified standard for the data market, making everything more efficient and transparent — *revolutionizing* the market. Wang's response: a more precise scale does help, *but only for standardized goods*. Token-based pricing makes sense for *standardized AI training data sets* — the equivalent of *flour, a homogeneous commodity that can be sold by weight*. But the data market is bigger than that. It also includes: - **Raw data** — wheat. Token pricing is wrong; raw data is sold by other dimensions. - **Deeply processed data products** — cakes and pastries. Industry insight reports, market analytics, custom data products. These are priced by *creativity, brand, scarcity, and value*, not by *weight*. A scale that can only measure flour, applied to wheat and cakes, distorts pricing. *"Revolutionizing data trading" is overreach.* ### Over-claim 2 — "Token-based pricing solves the data-element market" The claim: with token-based pricing, the data-element market will function. Wang's response: the bottleneck isn't measurement. *"What is China's biggest data-trading-market problem today? A missing scale? A missing unit? No."* The real bottlenecks: - **Insufficient high-quality data supply.** The shelves are nearly empty. - **Unclear data rights attribution.** Buyers don't trust they can use what they buy. - **High compliance cost.** Risk-averse sellers hold back. - **Data silos.** Data that exists isn't shared. A more precise scale doesn't solve any of these. *"Like giving a starving restaurant a more elegant spoon and claiming it will revolutionize the food industry."* ### Over-claim 3 — "AI's token algorithm transfers directly to data trading" The claim: AI's token-counting method should be the basis for data trading pricing. Wang's response: AI tokens are not language-neutral. They're optimized for the AI's compute efficiency — high-frequency word combinations merge into single tokens; low-frequency characters split into multiple tokens. The consequence: - A given semantic unit in English typically requires ~80 tokens. - The same semantic unit in Chinese requires ~120–150 tokens. - Lower-frequency languages can double or triple the count. If AI's token algorithm becomes the basis for data set pricing, **language difference, not data value, determines price**. A 1,000-token Chinese data set would be objectively worth less in token terms than a 1,000-token English data set — even if the Chinese data set is more valuable in content. The fix would require a *new* tokenization algorithm specifically for data-set pricing, decoupled from AI compute optimization. The NDA's draft doesn't yet specify what that algorithm would look like. ## What Wang's piece doesn't say but DCC notes The 国家数据局 (NDA) policy move is, on its own, not unreasonable. There IS a real benefit to having a unit of account for AI training data sets. Standardization helps. The criticism Wang directs at is *not* the NDA's policy — it's the *secondary commentary* claiming token trading will revolutionize the market. In DCC's reading, this is a useful primer for overseas observers because: - The "token trading" concept will appear in Chinese AI policy discussions for the next 12–24 months. Foreign readers need a sober framing. - The conflation of AI tokens and crypto tokens is real — and is being deliberately exploited by some commentators to attract attention. - The bottleneck Wang identifies — **high-quality data supply**, **rights clarity**, **compliance cost**, **data silos** — is the genuine constraint on the Chinese data-element market. Foreign teams advising on Chinese data deals should focus on these problems, not on the token trading hype. ## Why this matters for overseas teams Three operational takeaways: - **The "high-quality industry data set" framework is real and matters.** The NDA's draft Implementation Plan is the real policy direction — token trading is just one (overhyped) element of it. The plan's emphasis on industry-specific high-quality data sets, with standardized formats and quality, is the operational lever that will shape Chinese AI data supply over the next 2–3 years. Multinational AI providers active in China should align with the high-quality data set standards as they emerge. - **Token trading is a unit of account, not a market revolution.** When evaluating Chinese partner pitches that invoke "token trading," strip out the hype. The token is a measuring unit. The question for the deal is the same as it was before: what data is being acquired, what rights attach to it, what compliance has been done. Token-pricing is downstream of those questions, not a substitute for them. - **The four real bottlenecks are the operational risk map.** *Insufficient high-quality data supply* — sourcing risk. *Unclear data rights attribution* — title risk. *High compliance cost* — friction risk. *Data silos* — integration risk. A multinational's China data strategy will rise or fall on how it addresses these four. The token trading discourse is mostly a distraction from this map. The deeper observation in Wang's piece is that **policy hype cycles in the Chinese data-element market need a skeptical Chinese-language voice to puncture them**. Wang plays that role within the data exchange community. For overseas readers, having access to a sober Chinese perspective — not the hype, not the foreign critic — is one of the more useful things DCC's editorial mandate is built to provide. --- — Wang Qinglan (王青兰), *给"词元交易"泼一盆冷水* (Pouring Cold Water on "Token Trading"), 青兰数据观察 WeChat Official Account, April 24, 2026. [Original article (Chinese).](https://mp.weixin.qq.com/s/0Nbcam7GbrYx8d31JmTGGA) *Not legal advice. The above is DCC's structured summary of Wang's commentary; not a verbatim translation. The author's views are her own and do not represent her employer.*