ChainCatcher reported that AI Proem author Grace Shao said in an Oct. 9 post that China has built up a massive pool of data over many years, but has not yet produced a large independent AI data company on the scale of U.S. firms such as Surge AI, Scale AI, and Mercor.
Shao wrote that several factors sit behind that gap. Frontier model development and upstream data markets formed earlier in the United States. In China, major technology companies have long kept data within closed ecosystems. She also said data across enterprises and traditional industries is fragmented, while state-owned enterprises are not willing to easily hand it over to external suppliers.
China’s data output and structured data ratio
The post cited figures from the National Data Administration showing that China generated 52.26 ZB of data in 2025. Of the data already stored, only 22.13% was classified as structured.
How concentrated the labeling market is
Shao also referenced an interview published by DataYuan with Lin Zhenya, CTO of Benyuan Zhishu. According to that interview, China has about 2,000 to 3,000 data-labeling companies, and the top three account for roughly 20% to 30% of the market.
Lin estimated that the top three U.S. companies hold about 70% to 80% combined. Shao noted that this was an industry estimate, not an independently calculated market-share figure.
Funding, acquisition, and public company disclosures
The Information reported in September 2025 that Beijing-based UniPat had completed a financing round at a $1 billion pre-money valuation. Its rival Humanlaya reached a valuation close to $1 billion after completing its own financing.
According to the post, both companies received investment from HSG and were said to serve Alibaba, ByteDance, Moonshot AI, and DeepSeek.
A Hong Kong Stock Exchange filing from Thousand Baidu showed that the company acquired about 90.91% of Benyuan Zhishu in June 2026 for RMB 300 million.
Haitian Ruisheng said in its 2025 annual report that revenue rose 59% year over year to RMB 377 million.