New research published this week indicates that China’s most advanced artificial intelligence models were trained using Nvidia semiconductors subject to U.S. export controls, raising fresh questions about the effectiveness of Washington’s technology restrictions. The findings, detailed in a study examining the hardware behind several leading Chinese AI systems, suggest that despite tightened licensing rules imposed since 2022, developers in China have continued to access high-performance chips capable of supporting frontier model training. The research did not identify specific circumvention routes, but analysts note that restricted hardware has long been available through third-party markets and transshipment channels.
The disclosure is likely to intensify debate in Washington over whether current controls are sufficiently enforced or whether further tightening is required. Nvidia, which has consistently said it complies with all applicable export regulations, has seen its China-focused revenue fluctuate as successive rule changes reshaped its product lineup. The study’s authors caution that their findings are preliminary and call for broader verification, while Chinese AI developers have not publicly commented on the research. The episode underscores the difficulty of monitoring the physical supply chains underpinning AI development and is expected to feature in upcoming congressional hearings on export policy.
Independent reporting synthesized from primary wires and global correspondents.