---
title: "China's first national data-labeling plan targets 20%+ growth a year"
date: 2026-11-12
category: Policy & Governance
site: NeuroAI
canonical: https://neuroai.site/a/na-policy-data-labeling-industry
language: en
---

# China's first national data-labeling plan targets 20%+ growth a year

> Four Chinese ministries issued the country's first national opinion on the data-labeling industry on January 13, 2025, setting a 20%-plus annual growth target through 2027.

A self-driving car learns to tell a pedestrian from a shadow because a human tagged millions of frames. A medical AI recognizes a tumor because someone outlined thousands of scans. None of that happens without data labeling — the unglamorous labor behind every model. On January 13, 2025, China turned that back-office work into a national industrial priority.

## Key takeaways

- Four ministries — the National Development and Reform Commission (国家发展改革委), the National Data Administration (国家数据局), the Ministry of Finance, and the Ministry of Human Resources and Social Security — jointly issued the opinion (document No. 发改数据〔2024〕1822号, dated December 26, 2024), which was made public on January 13, 2025.

- It is the first national-level, systematic plan for the data-labeling industry (数据标注产业), covering demand, innovation, ecosystem, and support systems.

- The opinion sets a target that the industry's compound annual growth rate exceed 20% through 2027, with significantly stronger specialization, intelligence, and innovation capability.

- Seven cities, including Chengdu, Shenyang, Hefei, and Changsha, had already been designated as data-labeling bases before the opinion was issued.

- The plan leans on public-data labeling catalogs, fiscal and financial support, and local "data / algorithm / compute vouchers" to lower costs for labeling firms.

## Why labeling became a national issue

High-quality data, not just bigger models, is the bottleneck Chinese policymakers now name openly. State media quoted experts calling the shortage of high-quality data the central bottleneck in today's large-model (大模型) development. The opinion frames data labeling — filtering, cleaning, classifying, annotating, tagging, and quality-checking data — as the industry that produces the datasets models need to "read, learn, and train well."

The 2023 scale of China's data-labeling industry was estimated at around 80 billion yuan (≈ US$11B / HK$88B). That figure comes from industry estimates cited in state media, not a published official census, so treat it as a ballpark. The point the opinion makes is directional: the country wants that base to grow fast and get smarter, not just bigger.

## What the opinion actually commits to

The document lays out 13 specific measures across four areas. On demand, it tells governments to release public-data labeling needs, compile public-data labeling catalogs (编制公共数据标注目录), and push labeling into government services, city governance, and rural revitalization. On innovation, it backs key technology work such as cross-domain, cross-modal semantic alignment, 4D labeling, and large-model labeling (大模型标注), plus intelligent labeling and review tools.

On ecosystem, it aims to grow leading labeling enterprises and encourage consolidation through integration and mergers. On support, it promises fiscal and financial backing, pulls data-labeling services into government procurement, and calls for national occupational standards for AI trainers and data-labeling roles.

## The voucher link

One practical lever the opinion names is local vouchers. It tells localities to make full use of the data, algorithm, and compute vouchers (数据券、算法券、算力券) they already issue to cut labeling firms' costs, and to invite industrial and social capital into the sector. That connects this opinion to the broader local-subsidy playbook already running in places like Shenzhen — vouchers as a way to subsidize AI's input costs without the state owning the infrastructure.

## Where the seven bases fit

Before the opinion, China had already picked seven cities to host data-labeling bases — state media named Chengdu, Shenyang, Hefei, and Changsha among them. The opinion builds on that footprint, asking for a batch of distinctive, effective labeling bases and a relatively complete industry ecosystem by 2027. The bases are meant to concentrate talent, tooling, and quality control so that labeled data becomes a repeatable commodity rather than a scattered craft.

## Why labeling is now strategic

Labeling has always been the hidden tax on AI progress, but framing it as a national industry changes who pays it. By routing public-data catalogs and local vouchers toward labeling firms, the opinion shifts part of the cost off individual model developers and onto a shared base — the same logic behind China's broader compute- and data-voucher experiments. The bet is that cheaper, higher-quality labeled data compounds into better domestic models without the state having to own the factories.

## Honest limitations

This article describes the four-ministry opinion, made public on January 13, 2025 (the document is dated December 26, 2024); it summarizes, rather than reproduces, all 13 measures. The 20%-plus CAGR and the 2027 capability targets are forward-looking goals set by the opinion, not outcomes already achieved. The ~80 billion yuan (≈ US$11B / HK$88B) 2023 scale is an estimate cited in state-media reporting, not an official statistical release, and should be read as approximate. The opinion is a guiding document, not a binding regulation with penalties, and we did not track how each of the seven base cities or provincial governments have since implemented it.

## What readers can do now

- If you run a data or AI team in China, check whether your province offers data, algorithm, or compute vouchers — the opinion explicitly points firms to them.

- If you build labeling tooling or operate a labeling workforce, watch the public-data labeling catalogs and government-procurement openings the opinion calls for.

- If you train or hire for AI roles, note the push for national occupational standards in data labeling and AI training — credentials in this field may formalize over the next two years.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-data-labeling-industry
Free to quote with attribution and a link to the original.
