“It’s a terrible mistake to trade off national security for advantages in trade,” said Eric Hirschhorn, a former Commerce Department official, as Washington announced the dismantling of one of the biggest illicit technology export schemes in recent years. The Department of Justice’s “Operation Gatekeeper” has seized a network accused of funneling at least $160 million worth of Nvidia H100 and H200 Tensor Core GPUs to China and other restricted destinations, bypassing stringent U.S. export controls designed to safeguard America’s lead in artificial intelligence hardware.
Court filings reveal that Alan Hao Hsu, 43, of Missouri City, Texas, and his company Hao Global LLC pleaded guilty to smuggling and unlawful export activities. From October 2024 to May 2025, Hsu and others allegedly created false shipping documents for the goods, misclassified the nature of the merchandise, and hid their ultimate destinations. The GPUs-powering generative AI, large language models, and high‑performance scientific computing-went out to China, Hong Kong, and other prohibited destinations. More than $50 million in wire transfers from China were traced by investigators to finance the operation.
Two others charged are Fanyue “Tom” Gong, a Chinese citizen living in New York, and Benlin Yuan, a Canadian citizen residing in Ontario. Authorities said both conspired separately with a Hong Kong logistics company and a China-based AI firm to get around U.S. controls. Prosecutors said Gong arranged for straw purchasers and U.S. warehouses that stripped Nvidia labels from the GPUs, then rebranded them as “SANDKYAN,” and then shipped them overseas. Yuan allegedly hired inspectors and then directed them to conceal Chinese destinations; created cover stories to release shipments detained by the relevant authorities and discussed providing false statements to law-enforcement authorities.
The stakes go well beyond the courtroom. The H100 and H200 GPUs are not even Nvidia’s newest designs, but they rank among the most sophisticated processors legally barred from export to China under the export controls administered by the Commerce Department’s Bureau of Industry and Security. Their architecture provides enormous parallel processing power, high‑bandwidth memory, and specialized tensor cores optimized for AI workloads.
U.S. export controls have moved swiftly since the October 2022 rules that brought in country‑wide restrictions on high‑end AI chips to China. BIS has expanded tools like the Foreign Direct Product Rule to capture foreign‑made semiconductors built with U.S. technology, adding in end‑use and end‑user controls aimed at Chinese military‑civil fusion efforts. These are measures designed to choke off access to chokepoint technologies that underpin advanced AI model training and inference. But as Operation Gatekeeper shows, illicit networks exploit loopholes–routing hardware through third countries, mislabeling shipments, using shell companies to disguise end‑users.
The geopolitical backdrop is amplifying. President Donald Trump’s recent decision to allow Nvidia’s H200 chips to be sold to “approved customers” in China, with a 25% fee to the U.S. government, has drawn sharp criticism from lawmakers who warn it could bolster Beijing’s AI and military capabilities. Compared to the downgraded H20 chips previously permitted for China, the H200 is almost six times as powerful, according to the Institute for Progress, and would be helping meet surging demand for AI compute power in the country. Although Nvidia’s latest Blackwell architecture outclasses the H200, its performance, in conjunction with China’s rapid progress in domestic AI chips from Huawei, Alibaba, Baidu, and Cambricon, has the potential to narrow the gap.
Meanwhile, China’s AI hardware ecosystem is already adapting. Huawei’s Ascend 910C, Alibaba’s PPU, Baidu’s Kunlun P800, and Cambricon’s MLU 590 are all designed to reduce reliance on U.S. suppliers, with advances in FP8 precision formats, multi‑chiplet designs, and rack‑scale supercomputing clusters. These domestic efforts, however, still lag Nvidia’s top‑tier GPUs in raw performance, memory bandwidth, and software ecosystem maturity. Export‑controlled H100 and H200 chips remain attractive targets for illicit procurement because they offer proven capabilities and compatibility with existing AI frameworks like CUDA.
The globalized semiconductor supply chain-only heightens the enforcement challenge, as inputs can cross borders dozens of times before reaching end customers. Lacking coordinated controls from key technology export control allies like Japan, the Netherlands, and South Korea, unilateral U.S. measures face erosion over time. BIS has driven for closer alignment, but differences in licensing policies and enforcement capacity persist, leaving room for diversion.
Assistant Attorney General John A. Eisenberg drove that home: “This advantage isn’t free, but rather the result of our engineers’ and scientists’ hard work and sacrifice.” As Operation Gatekeeper continues, with Hsu facing up to 10 years in prison and Yuan up to 20, the case has emerged as a sobering reminder that in the U.S. China technology race, advanced AI chips are both the currency and the battlefield.
