NeuroAI NEUROAINEUROAI.SITE
ESC

China opens its public-data vault with a new rulebook

Since March 1, 2025, China runs a unified public-data registration and authorized-operation system, turning government-held data into a managed market asset.

2026-10-07 · 810 words · NeuroAI
China opens its public-data vault with a new rulebook

Every time you pay a utility bill, ride public transit, or file a permit, a government body or state-run enterprise records something. Multiplied across a country of 1.4 billion, that is an enormous pile of "public data" (公共数据) — and for years it mostly sat unused, locked inside the agency that produced it. On 1 March 2025, China switched on a national system designed to change that, without simply handing the data away.

The "1 + 3" architecture

The backbone is a 2024 opinion from the Party and State leading offices, Opinions on Accelerating the Development and Utilization of Public Data Resources (《关于加快公共数据资源开发利用的意见》), published on 9 October 2024. That is the "1." The "3" are three pillars issued together by the National Development and Reform Commission (国家发展改革委) and the National Data Administration (国家数据局):

  • Interim Measures for Public Data Resource Registration (《公共数据资源登记管理暂行办法》), document number 发改数据规〔2025〕26号.
  • Implementation Specification for Public Data Resource Authorized Operation (Trial) (《公共数据资源授权运营实施规范(试行)》).
  • A notice establishing a pricing-mechanism for authorized public-data operation.

All three were dated 8 January 2025 and took effect on 1 March 2025, with a five-year validity. On that same day, the National Public Data Resource Registration Platform (国家公共数据资源登记平台, sjdj.nda.gov.cn) went live.

What counts as public data

The rules define public data resources (公共数据资源) as the valuable data sets generated when party and government organs, enterprises and institutions perform their duties or provide public services (依法履职或提供公共服务过程中产生的具有利用价值的数据集合). Note the word "generated," not "collected" — the distinction matters, because it covers data produced in the course of serving the public rather than data actively harvested.

How authorized operation works

The core mechanism is "authorized operation" (授权运营): a local government or national ministry can authorize a qualified operating entity to govern and develop a defined slice of public data and offer it to the market as data products and services. The rules are deliberately structured:

  • Two roles. An implementing body (实施机构) runs the authorization; an operating entity (运营机构) does the development. The operator may not re-develop the delivered products itself, but others are encouraged to build on top of them.
  • Open selection. Operators are chosen through public tender, invitation-to-tender, or negotiation — not back-room assignment.
  • Term limits. Authorization runs at most five years, and the registration result is valid for three years, renewable.
  • Disclosure. Both the implementing body and the operator must periodically publish what was authorized, to whom, and how it is used.

Registration, step by step

The registration measures lay out a clean lifecycle: application → acceptance → formal review → public notice → code assignment. First registration is due within 20 working days of starting operation; operations already running before the rules took effect had 30 working days to register. The national platform issues a unified query code, and registration results are meant to be interoperable across provincial platforms. A registration result is valid for three years and can be renewed in three-year blocks.

Why it matters for AI

Public data is among the highest-quality, widest-coverage data China has — weather, transport, healthcare, utilities, permits. For AI teams starved of clean training and evaluation data, a compliant channel to access and build on public data is a big deal. The rules explicitly encourage privacy-computing (隐私计算) and other secure, traceable circulation technologies, which is the technical assurance that makes market use feasible without exposing raw records.

Reading the trend

According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the registration-plus-authorization design is best read as turning public data from a free-by-default resource into an accounted-for asset class — which is what makes cautious, large-scale AI use politically permissible.

Honest limitations

This article describes the rule framework as published by the NDRC and the National Data Administration; we have not tracked how many provinces or ministries have stood up local registration platforms or signed first operators, and those rollout numbers are not yet consolidated in a single public dashboard. The pricing-mechanism notice sets principles rather than fixed prices, so the economics for operators will vary case by case. The rules also leave sector-specific scope (what data is in or out of authorization) to implementation plans vetted through local "major decision" (三重一大) procedures, so the practical data availability differs by region. Treat the system as launched and operative, but still maturing in coverage.

What readers can do now

  • If your work touches public-sector data, check whether the relevant province or ministry has published its authorization implementation plan and registration entry on the national or provincial platform.
  • AI and data-vendor firms should watch the open tenders for operating entities — the rules require competitive selection, creating a real procurement window.
  • For foreign-invested or cross-border teams, pair this with China's separate cross-border data provisions; authorized public-data products still sit inside the domestic compliance perimeter.
  • Follow the National Data Administration (国家数据局) for the evolving standard set and platform-interoperability notices that determine how portable a registration is across regions.

Related coverage

More in “Policy & Governance” → · Back to home · Markdown version