While the U.S.
government has restricted China’s access to advanced AI chips, no similar prohibitions currently exist for the training data used to develop the models themselves.
Industry experts suggest this multi-hundred-million-dollar trade is helping Chinese AI labs close the performance gap with their American rivals.
By purchasing "off-the-shelf" datasets—standardized packages of human-graded information—Chinese firms gain access to the same sophisticated reasoning structures and quality-control rubrics used by leading U.S.
developers.
Nathan Lambert, a former researcher at the Allen Institute for AI, told *Forbes* that high-quality data is the most significant factor for making models viable in challenging domains like financial modeling or coding, allowing buyers to bypass months of trial and error.
The scale of this commerce is significant, with estimates suggesting the top six Chinese AI labs spend approximately $500 million annually with American data vendors.
This creates a complex landscape where the same startups providing data to the U.S.
military and federal government are simultaneously fueling the AI capabilities of foreign competitors.
While some industry leaders view these sales as a national security risk, others argue that restricting data exports could inadvertently damage global open-source innovation and create domestic monopolies.