Banned Nvidia chips power China AI models

New research indicates that some of China's most advanced AI models were trained using Nvidia chips that U.S. export rules prohibit from being sold to China, according to a TechCrunch report. The finding is adding pressure to an already heated debate over whether export controls are slowing Beijing's AI progress or merely pushing supply chains into less visible channels. Supporters of the restrictions say the rules remain essential because advanced chips underpin frontier AI development. Critics argue that the research shows the measures are leaky and that high-end hardware can still reach restricted buyers through intermediaries. The report points to questions about chip smuggling, third-party transshipment, and the difficulty of tracking advanced semiconductors after they are sold. Once hardware leaves the original buyer, following its movement across borders and through cloud providers or resellers can be extremely hard. That uncertainty is fuelling political pressure on Nvidia and prompting lawmakers to ask whether companies are doing enough to prevent diversion. It also complicates the broader debate over how to measure the effectiveness of export controls. If Chinese labs can obtain banned chips, the policy may need stronger enforcement, better intelligence, or new technical safeguards. If the problem is limited, officials may argue that the controls are working as intended. The research does not settle that argument, but it gives critics and supporters new evidence to cite. For Nvidia, the issue is especially sensitive because China is a major market and because the company is already navigating congressional scrutiny over its global sales. For policymakers, the central challenge is enforcement: designing rules that can be verified in practice rather than only announced on paper. The report suggests that the next phase of the AI chip debate will focus less on whether controls exist and more on whether they can be monitored, enforced, and trusted.