A third Nvidia chip-smuggling network charged in six months, as a research group maps how the chips actually get to China

Federal prosecutors arrested Greg Lui, the 38-year-old owner of a City of Industry, California server reseller, on October 1 on a three-count indictment accusing him of smuggling more than $300 million worth of export-controlled computer servers containing Nvidia AI chips to China. A federal grand jury had returned the indictment two days earlier, on September 29. It is the third publicly charged chip-smuggling network since March -- and set against a research group's new map of how restricted chips actually travel and a bipartisan Senate letter asking why Nvidia's own compliance keeps missing it, the pattern now looks less like a crackdown reaching its peak than a chronic, ongoing leak.

The scheme prosecutors describe, run through Lui's company Earthmade Computer Inc. between 2023 and 2024, follows a now-familiar shape: buy high-end servers from US manufacturers under false paperwork naming buyers in countries that don't require an export license for Nvidia's top parts, route the hardware to Singapore or Malaysia, then forward it on to China. One cited purchase order covers 27 servers with Nvidia H100 GPUs for roughly $7.6 million, routed through Kuala Lumpur; another describes 100 H100-equipped servers worth $22 million. Between January and October 2024 alone, prosecutors say Earthmade received more than $176 million from two Malaysia-based shipping companies as part of the scheme. Lui faces up to 20 years each on the export-control and money-laundering conspiracy counts, and up to 10 on outbound smuggling.

Nvidia itself is not a defendant in the case and says the pattern doesn't reflect badly on its controls. “Smuggling is a nonstarter,” the company has said, adding that it primarily sells to well-known partners, including the original equipment manufacturers who help ensure sales comply with export rules -- and that reported diversion represents under 0.5% of its total shipments. That figure is Nvidia's own estimate; no independent audit of it is cited in any of this week's reporting.

Lui's case is not the first of 2026, or even the largest. In March, federal prosecutors in the Southern District of New York charged Supermicro co-founder Wally Liaw along with Steven Chang and Willy Sun with conspiring to route $2.5 billion worth of servers to a Southeast Asian reseller, which allegedly repackaged roughly $510 million worth of them -- carrying restricted H200 and B200 GPUs -- onward to China. Surveillance footage cited in that case shows the alleged method: replica "dummy" servers substituted for real ones during audits, with genuine units' serial numbers peeled off using a hair dryer and transferred to the fakes. And in July, Singapore authorities seized a $42 million mansion tied to Wei Zhaolun, chief executive of server reseller Aperia Group, as part of a money-laundering case connected to a separate US investigation into whether chips reached China's DeepSeek through third-party buyers.

Three criminal cases are a count of prosecutions, not a measure of scale -- for that, the research group C4ADS published a report on September 9, titled Covert Compute, mapping how restricted chips move independent of any single indictment. Its conclusion: “China has demonstrated a sustained and adaptive ability to acquire restricted U.S.-designed chips, exploiting gaps in jurisdictional coordination and enforcement regimes.”

Reading those three figures together is the point: a $4.6 billion import total, a $13.4 million transshipment estimate, and the $1.7 million in chips C4ADS traced into university contracts are three different kinds of number, and treating them as comparable is exactly the kind of error this component exists to prevent. What C4ADS actually establishes is structural -- three distinct, independently operating pathways, each requiring a different fix -- not a single aggregate dollar figure for how much restricted compute has reached China.

Nvidia's own response, underway since before Lui's arrest, has been to tighten who it sells to rather than contest the cases individually: the company built a stricter buyer whitelist this year and removed more than half of its previously authorized Asian AI-chip buyers from it, with intensified due-diligence requirements specifically in Singapore, Malaysia, and Japan -- the same jurisdictions named in both the Lui and Supermicro cases.

None of this week's reporting says whether Lui, the Supermicro defendants, or the Singapore network actually succeeded in getting working chips into a Chinese AI lab's hands, as opposed to into a reseller's warehouse -- that distinction matters and remains unresolved in the public record. What's established is narrower and still significant: the handoff points keep being the same two or three Southeast Asian jurisdictions, the paperwork trick keeps being the same false end-user declaration, and -- for now -- Nvidia's fix is a buyer list it has not published the size of, rather than a change to how the hardware physically leaves its supply chain.