Surge AI, Mercor Reportedly Sell Training Data to Chinese Labs

Interesting angle from Forbes this week that reframes what US-China AI competition looks like on the ground. Chip export controls have been the loud policy story for years, but according to a Forbes investigation by Anna Tong, the same Silicon Valley data-labeling shops feeding OpenAI, Anthropic and in some cases US federal entities are also selling training datasets to China's top labs.

The specifics, as Forbes reports them: AfterQuery reportedly pulls at least $50 million in recurring revenue from Chinese AI labs. Mercor booked roughly 2% of its Q2 revenue from Chinese labs, off an annualized run rate that crossed $2 billion in June, and has hired an engagement manager specifically for China. Surge AI, whose CEO Edwin Chen founded the company in 2020 and whose 2024 revenue Forbes puts at $1.2 billion (from customers including Google and Anthropic), has actively courted the region, with Chen traveling to China to meet lab executives directly. Internal project documentation reviewed by Forbes suggests Turing has worked with ByteDance. In aggregate, Forbes estimates the top six Chinese AI labs are spending roughly $500 million a year with American data labelers.

The Forbes framing shifts where the leverage sits in this contest. US policy has spent the last few years assuming that gating advanced chips was the load-bearing lever. Expert-annotated training data turns out to be a parallel and largely unregulated lever, and the same vendors that built the pipelines pushing US frontier labs forward are, per Forbes, selling comparable pipelines into Beijing. That closes a gap chips alone were supposed to hold open.

Where Forbes' evidence has gaps: the reporting leans on documentation, internal communications and audio transcripts from lab employees rather than on-record confirmations, and ByteDance, Alibaba, Moonshot, Tencent and Ant Group did not respond to Forbes. The vendors' own framing of the work is not in the piece, and the article does not spell out whether any of this crosses an existing US export-control line or only an unwritten one. What is likely to shift regardless is the policy conversation: if data pipelines start getting treated the way chips are, Mercor, Surge AI, Turing and AfterQuery move into a very different regulatory posture than the growth-stage story they have been telling investors.