The companies training America’s military AI are also cashing checks from Beijing’s top artificial intelligence labs. US data labeling firms collectively pull in roughly $500 million per year from China’s six leading AI operations, a commercial relationship that sits awkwardly alongside Washington’s escalating efforts to throttle Chinese access to advanced technology.
The numbers behind the dual-client model
Surge AI reported $1.2 billion in revenue in 2024, built on providing expert-annotated datasets to both US military branches and Chinese AI laboratories. The annotations are complex, domain-specific datasets used to train sophisticated AI models, the kind of work that requires genuine expertise and commands premium pricing.
Mercor has been growing even faster, crossing an annualized revenue run rate of $2 billion by June 2026. Chinese contracts accounted for 2% of its Q2 2026 revenue.
AfterQuery pulls in at least $50 million in recurring annual revenue from Chinese AI labs alone. Turing rounds out the group of notable US firms participating in this trade.
On the buying side, the client list includes ByteDance, Alibaba, Ant Group, and Tencent. The Pentagon has designated Tencent as a Chinese military company, which means a US defense-adjacent data firm is simultaneously servicing an entity that the Department of Defense considers part of China’s military apparatus.
Why data annotation matters more than chips
The datasets these American firms produce require English-fluent domain experts, often with advanced degrees, who can provide the nuanced annotations that teach AI systems to reason, code, and analyze. This is exactly the type of human expertise that’s hardest to replicate domestically in China, which is precisely why Chinese labs are willing to pay American firms hundreds of millions for it.
Companies like Anthropic also rely on these same annotation services. So the supply chain serving America’s most security-conscious AI developers is, in practice, the same pipeline feeding Chinese competitors.
The regulatory gap
The situation exposes a significant blind spot in US technology export policy. While the Commerce Department’s Bureau of Industry and Security has built increasingly sophisticated restrictions around chip exports, chip-making equipment, and even cloud computing access, there is no equivalent framework governing the sale of annotated training data.