Editor’s Note — DCC.
On 2 July 2026 the State Administration for Market Regulation and the Standardization Administration of China issued two recommended national standards for data held as an asset — GB/T 47949-2026 Classification and Codes for Data Assets and GB/T 47950-2026 Guidance for Data Assets Registration. Both take effect on 1 September 2026. They were drafted by the Ministry of Finance’s Asset Management Department and the China National Institute of Standardization under the National Technical Committee on Asset Management Standardization (SAC/TC 583).
This is DCC’s own reading of the two standards. The single thing to carry away: both use the word 登记 (“registration”), and they are not the same act as the registration in the National Data Administration’s Data Property Rights Registration Work Guide (Trial) — finalized on 1 July 2026, which DCC covered in the draft-to-Trial diff. One registers data as an asset on your own books; the other registers rights in data with an outside institution that issues a certificate. Conflating them is the central error, and it is easy to make because the Chinese term is identical.
— Not legal advice.
The same word, twice
Chinese data policy now contains two things called 数据…登记, and a foreign reader who meets them a week apart can be forgiven for assuming they are one regime described twice. They are not. They come from different ministries, answer different questions, and produce different pieces of paper.
- 数据资产登记 — data-asset registration. The subject of the two new standards. It is an act of asset management: an organization records a data resource it has confirmed to be an asset onto its own asset register (资产台账) and into its own accounting books. Nothing is filed with anyone. The MOF/SAC standards tell you how.
- 数据产权登记 — data property-rights registration. The subject of the NDA’s Trial Guide. It is an act of rights confirmation: an applicant asks an NDA-catalogued registration institution to review who holds which of the three data rights (hold / use / operate), which then publicly announces the claim, preserves the evidence, and issues a certificate.
Put side by side:
| Data-asset registration (GB/T 47950) | Data property-rights registration (NDA Trial Guide) | |
|---|---|---|
| Who registers | The organization itself, internally | An NDA-catalogued registration institution |
| Question answered | Is this an asset of ours, and what is it worth? | Who holds which right over this data? |
| Output | Data asset card, asset register, accounting books | A publicly announced, evidence-preserved rights certificate |
| Governing instrument | GB/T 47950 + GB/T 47949; MOF data-asset policy | Data Property Rights Registration Work Guide (Trial) |
| Legal effect | Internal management, accounting, reporting | External proof of attribution in transactions, financing, disputes |
| Trigger | Follows from asset-management duties | Voluntary application |
They are not rivals — they interlock. The data asset card defined by GB/T 47950 has a field for whether ownership has been registered, and the thing that fills that field is precisely the NDA certificate. An organization pursuing full data-asset capitalization will generally end up doing both: confirm and book the asset under the MOF/SAC track, and confirm the rights under the NDA track. But they are two projects, run by two internal owners, against two rulebooks.
What the two standards actually do
GB/T 47949 gives data assets a code. It slots them into China’s existing asset-classification system as intangible assets: the whole population of data assets now lives under the block A0806020000 — intangible assets (08) → information-and-data intangible assets (06) → data (02). Below that, the standard’s Clause 6 table splits data assets into three classes by structure — structured, semi-structured, unstructured — and fourteen sub-classes, each with a prescribed unit of measure (megabytes, records, images, tracks, and so on). It classifies by what the data technically is, not by sector or use, and it is careful to say it does not change how data assets are defined or classified for accounting purposes.
GB/T 47950 gives them a process. It sets three principles — security, compliance, traceability — routes the Data Security Law and PIPL into the registration process as a precondition (a data asset whose source is non-compliant is not registrable), and allocates the work across four internal functions: the data business owner, asset management, accounting, and — notably — the data-technology/informatization function, which runs the compliance review at every stage. Registration is dual: an asset register built from a data asset card, and the accounting books, kept in parallel. Classification and units of measure on the card come straight from GB/T 47949.
The one genuinely new thing: an asset code for training data
Buried in the GB/T 47949 table is A0806020307 — AI-training multimodal data: pre-training and fine-tuning corpora, image-text / video-text / audio-text aligned samples, synthetic labelled data, multi-source heterogeneous data. So far as DCC is aware, this is the first time a Chinese national asset-classification standard has given AI training corpora a dedicated asset code — and its unit of measure is megabytes and tokens.
That token count is not decoration. If a Chinese counterparty is capitalizing its training data, this is the code line its asset register will carry and the quantity the register expects to see. For any cross-border AI deal where training data is part of the consideration or the diligence, “how many tokens, under which code, on whose books” is now a question with a standardized answer format.
Why 入表 and 确权 keep getting mixed up
The reason the two registrations are so easy to confuse is that the underlying concepts they serve are also routinely confused: 入表 (balance-sheet entry) and 确权 (rights confirmation).
- 入表 is an accounting question — can this data resource be recognized as an asset, and at what value? It is governed by the Ministry of Finance’s Interim Provisions on the Accounting Treatment of Enterprise Data Resources (财会〔2023〕11号) and the data-asset management notices. GB/T 47950 is the operating manual for the register that sits behind 入表.
- 确权 is a rights question — who owns or holds what in the data. It is governed by the Data 20 Articles three-rights framework and, operationally, by the NDA Trial Guide.
You can do one without the other, and the gap between them is exactly where data-asset deals get fragile. DCC’s brief on Xu Ke’s account of when data becomes an asset works through the definitional side of 入表 — the narrow-sense test (owned or controlled, generates economic benefit, reliably measurable) that GB/T 47950’s “data asset confirmation” (数据资产确认) now codifies as the gate to initial registration. The registration standards handle the asset half of that sentence; the NDA guide handles the rights half. Neither resolves the deeper contest — whether a data right is a property right at all — that the Data 20 Articles deliberately left open.
The quiet innovation: the subsidiary record
One feature of GB/T 47950 deserves a flag because it is genuinely useful: the data asset subsidiary record (数据资产备查簿). The standard’s process is gated on confirmation — a data resource that meets the asset-confirmation conditions goes to initial registration; one that does not yet meet them is entered into the subsidiary record instead of being lost.
That gives organizations a defined place to park data that is real and managed but not yet bookable — data whose ownership is unsettled, whose economic benefit is not yet demonstrable, or whose value cannot yet be measured reliably. Without it, the practical choice would be premature recognition or invisibility. For a firm building toward capitalization, the subsidiary record is the on-ramp: it is where a data resource waits while the 确权 question is resolved on the NDA track and the 入表 test is worked toward on this one.
For overseas counsel
Three practical points:
- Do not treat a data-property certificate and an asset entry as the same assurance. An NDA certificate answers who holds the rights; a GB/T 47950 asset register answers what the counterparty booked. Diligence that wants both must ask for both — and check whether the asset card’s “ownership registered?” field points to an actual certificate.
- The compliance gate is real and it is upstream. GB/T 47950 makes source compliance — DSL, PIPL, important-data rules — a precondition of registration, mirroring the source-compliance review the NDA guide imposes on the rights side. Data with a defective provenance is registrable on neither track.
- Watch the training-data code. A0806020307 standardizes how AI training corpora appear as assets. If your matter involves Chinese training data as value — licensing, contribution, securitization — expect it to surface in a standardized asset form, measured in tokens, from 1 September 2026 onward.
Both standards are recommended (GB/T), not mandatory (GB) — they bind through adoption in asset-management and procurement practice rather than by force of law. But given that the Ministry of Finance drafted them and its own administrative-institution data-asset notice (财资〔2024〕1号) already directs public bodies to register and report data assets, adoption in the state sector is not really optional, and the enterprise sector tends to follow the MOF register format regardless.
— Not legal advice.