U.S. Officials Say Nvidia Is Falling Short on China Chip Smuggling

U.S. government officials say Nvidia is not doing all it could to stop illegal smuggling of its AI chips to China.

The allegation, reported by Engadget, focuses on due diligence. Officials and industry critics point to gaps in how large orders were vetted and question how smuggling red flags went unnoticed. They estimate that 115,000 or more restricted Nvidia processors have already been diverted to China.

At the center of the current criticism is a $2.5 billion transaction involving OBON, a Bangkok-based company. The AI servers were sold ostensibly for AI infrastructure in Thailand. U.S. officials say they were instead shipped to end users in China.

Officials argue Nvidia should have caught the discrepancy. OBON lacked the 100 megawatts of data center space needed to host the volume of servers it purchased. That mismatch, they say, should have triggered additional review before shipment.

Federal authorities have paired that criticism with enforcement action. Greg Lui was arrested for allegedly smuggling servers containing $300 million worth of AI chips to China in violation of U.S. export laws. In connection with smuggling cases, officials released surveillance video depicting a person using a red hair dryer to peel serial number stickers off Nvidia server boxes.

Nvidia says it has abided by export rules and has never been accused of abetting smuggling.

The cases described this week build on a longer enforcement record. In an action called Operation Gatekeeper, U.S. authorities said they disrupted a trafficking network and seized more than $50 million in advanced GPUs destined for China and other destinations, according to the Justice Department. Separately, prosecutors alleged a smuggling ring attempted to export at least $160 million worth of Nvidia AI chips between October 2024 and May 2025, as reported by CNBC. Three people linked with a U.S. technology supplier were later charged over an alleged plot to smuggle American AI chips to China, the BBC reported.

Investigators have also turned to technical countermeasures. U.S. authorities secretly placed location tracking devices in targeted shipments of advanced chips to catch diversions to China, Reuters reported. A U.S. lawmaker planned to introduce legislation to verify the location of artificial-intelligence chips like those made by Nvidia to target smuggling, Reuters reported.

Transshipment through third countries remains a concern cited by officials. U.S. officials believe Malaysia is among a handful of countries where organized smuggling operations have worked to get U.S. chips into China. A U.S. lawmaker asked the Trump administration to help Malaysia crack down on chip smuggling, according to Reuters. Bloomberg reported in a feature titled "China Chip Smuggling Cases Expose Nvidia's Blind Spots" that Nvidia's AI chips keep making their way to China despite U.S. curbs, and that the company faces questions over those cases.

The broader context here is worth spelling out for practitioners who manage infrastructure supply chains every day. Export compliance for high-value compute has traditionally stopped at the point of sale to a vetted distributor or integrator. That model assumes serial numbers, invoices and end-user attestations create a reliable chain of custody. The tactics described by officials, from peeled labels to infrastructure orders without power capacity to support them, are intended to break that chain after the initial sale.

In my view, the dispute is less about whether Nvidia violated written rules and more about where responsibility for ongoing diligence should sit when the product is concentrated, easily diverted and used for AI buildouts. Vendors can check power capacity and order scale without operating as law enforcement. Enforcement agencies, in turn, can expand seizure operations and embedded tracking, but those tools scale poorly against 115,000 diverted units. Location verification would shift the control point, though deployment and circumvention tradeoffs would need careful design.

What remains workable is unglamorous. Tighter checks on large-scale orders, preservation of serial-level traceability through distribution, and faster information sharing between vendors and investigators would not end diversion. They would raise its cost and narrow the channels through which large clusters can be assembled unnoticed.