Feds bust four for smuggling Nvidia AI chips to China

The real estate company didn't sell a single home. Instead, it was allegedly moving one of the most heavily restricted commodities on the planet: top-tier Nvidia AI chips.

Federal prosecutors have charged four people with illegally smuggling Nvidia GPUs and HP supercomputers into China, according to a court filing uncovered by Court Watch. The scheme allegedly routed dozens of cutting-edge processors through a network of shell companies and falsified shipping manifests.

The four individuals charged are Mathew Ho, Bryan Curtis Raymond, Tony Li, and Harry Chen. They face counts of smuggling, conspiracy, and money laundering. As of now, only one suspect has been arrested.

Fake leases and real H200s

Court filings outline an operation that kicked off in late 2023. Without obtaining export licenses, the group allegedly managed to ship 50 prized Nvidia H200 GPUs along with multiple batches of older H100s to China.

At the center of the pipeline was Janford Realtor LLC. Despite the name, prosecutors say the firm never touched any real estate. It existed purely as an intermediary to buy and export advanced American hardware.

Ho, a US citizen, served as the registered agent, while Li, a Chinese national, was identified as a manager. Funds were wired directly from bank accounts in China to buy hardware from domestic vendors, masking the trail with forged shipping papers and bogus sales contracts.

Around $2 million flowed directly to an Alabama entity owned by Raymond. Raymond publicly lists himself as CEO of Bitworks, an AI infrastructure firm offering Nvidia and AMD sales. He recently claimed to be hired as CTO at AI cloud provider Corvex, but Corvex quickly denied this, stating his job offer was rescinded and the company had zero involvement in the case.

Nvidia pushed back hard against the idea of illicit clusters. Company spokesperson John Rizzo said cobbling together datacenters from smuggled gear is a technical and economic nonstarter. Modern datacenters are complex systems, and Nvidia provides zero support or repairs for restricted hardware.

Why it matters

This case pulls back the curtain on how hardware actually moves when geopolitical fences go up. Ever since DeepSeek released its R1 model, Silicon Valley has debated how much frontier compute actually sits behind the trade barrier. Scale AI CEO Alexandr Wang previously argued that China holds far more H100s than official tallies reflect. Cases like this show precisely how that hardware slips through the cracks.

Yet running a few dozen H200s is very different from training a frontier model. Without official driver maintenance, replacement parts, and cluster networking support, running smuggled chips at scale is an operational nightmare. It works for specialized lab research or fine-tuning, but it cannot sustain a hyperscale fleet.

We think this bust will trigger an aggressive compliance crackdown across the entire secondary hardware market. Independent GPU brokers and regional system builders will face far deeper background checks, higher legal costs, and intense scrutiny over every commercial invoice. Smugglers might grab headlines, but the immediate result will be more friction and overhead for legitimate AI startups trying to source hardware.