U.S. Startups Sell $500 Million in AI Training Data to Chinese Labs
Updated
Updated · Forbes · Aug 5
U.S. Startups Sell $500 Million in AI Training Data to Chinese Labs
1 articles · Updated · Forbes · Aug 5
Summary
$500 million a year in AI training data is flowing from U.S. labeling startups to China’s top six AI labs, according to documents, messages and industry estimates reviewed by Forbes.
The trade persists because Washington restricts advanced chips but not training data, letting Chinese buyers purchase custom and off-the-shelf datasets built with the same expert networks, rubrics and quality controls used for OpenAI and Anthropic.
Tencent, Alibaba, Ant Group and ByteDance were identified as buyers or counterparties, while U.S. suppliers cited in the report included Surge AI, Mercor, AfterQuery and Turing; one source said Mercor got 2% of Q2 revenue from Chinese labs.
The sales give Chinese labs a faster route to narrow the performance gap with U.S. rivals by buying packaged human expertise instead of building data pipelines from scratch, raising national-security concerns even as vendors defend the business as part of open-source AI.
Could restricting data exports to protect U.S. dominance accidentally destroy the open-source ecosystem that built the AI industry in the first place?
While blocking AI chips, are Silicon Valley startups quietly handing competitors the exact blueprints needed to win the global AI race?
The $500 Million AI Data Pipeline: U.S.-China Trade, Regulatory Loopholes, and the Global Open-Source Shift
Overview
The global AI market is shaped by a massive, largely unregulated trade in training data and services, with U.S. companies like Appen relying on Chinese clients even as their stock values fall. While the U.S. focuses on restricting hardware exports, a regulatory gap allows data and software to flow freely, enabling Chinese labs to build competitive models by optimizing data use and bypassing chip restrictions. This has led to industrial-scale distillation of U.S. AI models by Chinese firms, sparking security concerns and prompting U.S. developers to restrict access. Meanwhile, Chinese open-source models offer a strong cost advantage, leading 80% of U.S. startups to use them, and global backlash grows against closed U.S. models after abrupt shutdowns. As the EU enforces strict transparency rules for AI, the world faces increasing regulatory fragmentation and the risk of splintered AI ecosystems.