Chinese data vendors emerging and trying to go global – focus on healthcare data

China amassed vast amounts of data over a decade but did not spawn big independent AI data vendors earlier because much labeling and data work was done in-house or through domestic outsourcing, and dominant tech platforms and state-owned enterprises hoarded their information in walled gardens and fragmented systems. As frontier models evolved they began demanding specialist knowledge (doctors, lawyers, engineers), multimodal inputs (images, audio, video) and realistic task environments rather than sheer volume. That shift created opportunity for firms that already control expert networks or industry relationships to package structured, high-quality training data and simulated workplaces. Names now surfacing in China’s nascent data sector include UniPat, Humanlaya, 本原智数 and 海天瑞声, which seek to convert proprietary industry access into sellable datasets, task designs and evaluation frameworks despite persistent “data island” fragmentation.

The U.S. data boom offers a blueprint: many leading vendors started as recruiting or software businesses that monetized networks and expertise. Surge AI generated about $1.2 billion in revenue in 2024 and was discussed at a $25 billion valuation in 2025; Scale reported roughly $870 million in 2024. Mercor raised $350 million at a $10 billion valuation in October 2025 and reached a reported ~$2 billion gross annualized run rate, paying 60-70% to contractors (implying roughly $600-800 million retained pre-costs). Snorkel and others are building reinforcement-learning environments and standardized processes; acquisitions like Mercor’s purchase of Deeptune illustrate the push to combine expert labor, simulated environments and measurement. Chinese entrants face both opportunity and limits as labs can internalize parts of this stack and data ownership remains highly guarded.