In July 2026 the National Data Administration stated at its big data expo press conference that it encourages business-model innovation based on token applications, explores new trading models such as token trading, cultivates a market consensus of “paying for quality data”, and deepens the integration of token applications with industry, healthcare and finance.
This is a statement oriented towards industrial policy, but at the company level it raises a series of concrete compliance questions.
From “pay per call” to “pay per token”
In the past, AI services were mostly priced by number of calls or by seat. Token pricing refines the billing granularity to the model’s input and output volume. Its advantage is that it is closer to actual consumption, and it lets “data quality” be priced directly for the first time — quality data produces better model performance, which in turn supports a higher token value.
The “data flywheel” effect the National Data Administration mentions describes exactly this cycle: model applications pull data supply, and data in turn empowers model iteration.
Four contractual issues raised by token trading
Pricing and settlement clauses. Who provides the statistics on token consumption, how they are verified, and whose data prevails in a dispute must be written into the contract.
Data authorisation chain. If token trading involves the transfer of training or inference data, whether the authorisation covers secondary use and whether it can be sub-licensed are core risk points. See our Data Compliance service page.
Service-level undertakings. Under token billing, response latency, rate limiting and availability have a large impact on cost, and service-level clauses must match the pricing method.
Termination and data return. After cooperation ends, how accumulated data and fine-tuning results are handled should be arranged in advance.
Two easily overlooked compliance points
First, the legality of data sources. Paying for quality data presupposes that the data itself has clear provenance. Building a dataset from crawled, purchased or user-uploaded data requires checking the scope of authorisation.
Second, cross-border scenarios. If token trading involves cross-border data flows and cross-border settlement, it triggers both cross-border data transfer and foreign-exchange management. See our Multi-Jurisdiction Compliance service page.
Three points for AI companies
First, write token pricing into standard contracts early, clarifying statistical basis and dispute resolution. Second, build a data-source and authorisation ledger to provide a compliant foundation for “data pricing”. Third, watch tax treatment — as a new trading model, the invoicing and revenue-recognition method for token trading needs to be confirmed with tax advisers.
For support on contract design for token trading, data authorisation and cross-border arrangements, see our Commercial Contracts and Data Compliance services.
This article is based on public statements by the National Data Administration in July 2026 on business-model innovation for token applications. It is for reference only and does not constitute legal advice.