Skip to content
AI Usage3 min read

US Arrests Tech CEO Over $300M Nvidia Chip Smuggling

Federal authorities arrest Earthmade Computer CEO Greg Lui for allegedly smuggling $300 million in export-controlled Nvidia AI GPUs into China.

AB

Author

AUG Bot

Published

Digital representation of Nvidia AI chips and international supply chain export controls

US Arrests Tech CEO Over $300M Nvidia Chip Smuggling

Fraudulent paperwork masked high-end AI server shipments to China

Federal authorities have arrested Greg Lui, CEO of Earthmade Computer, for allegedly executing a $300 million scheme to smuggle export-controlled Nvidia AI servers into China. The illicit diversion of Nvidia A100 and H100 graphics processing units highlights the intense global premium on advanced AI compute hardware and the escalating costs of bypassing international trade restrictions.

Key details

According to an indictment filed by the Federal Bureau of Investigation, 38-year-old Greg Lui used falsified purchase orders and identity theft to bypass U.S. export controls between October 2023 and August 2026. The Department of Justice alleges that Earthmade Computer purchased servers loaded with Nvidia A100 and H100 GPUs under the guise of shipping them to entities in Malaysia and Singapore.

The hardware was subsequently re-routed through transshipment hubs in South Asia and Hong Kong before arriving at AI development firms in Hangzhou, China. Evidence cited by federal prosecutors includes a 2024 order where Lui purchased 27 export-restricted servers for approximately $7.61 million, which co-conspirators transferred directly to Chinese buyers. In another transaction, a $22 million shipment of 100 servers containing Nvidia H100 chips was exported using stolen identity documents under the name "Jackie Lui." In 2024 alone, Earthmade Computer allegedly received over $176 million from Chinese entities as part of the $300 million smuggling operation.

Why this matters

The case illustrates how severe chip shortages and U.S. export limits have created a massive grey market for AI hardware, inflating compute procurement costs for global organizations. When hyperscale chips like the Nvidia H100—which command tens of thousands of dollars per unit—are diverted through illicit supply chains, black-market markups drive up the baseline capital expenditure required to secure frontier AI infrastructure worldwide.

Context

As artificial intelligence models scale in complexity, access to high-density GPU clusters has become a primary bottleneck for tech infrastructure. U.S. trade restrictions aim to curb China's access to advanced AI hardware, but surging demand for model training and inference has incentivized complex supply-chain workarounds. Similar component shortages have pushed enterprise hardware lead times to 40 weeks and driven workstation GPU pricing up by 55%, reinforcing hardware procurement as a major cost driver for AI scaling.

Risks and open questions

The arrest underscores persistent vulnerabilities in the global hardware supply chain, as transshipment networks in intermediate countries continue to facilitate illicit chip flows. It remains uncertain whether tightening trade enforcement will successfully restrict access or simply increase the black-market price premium on Nvidia GPUs, further distorting international hardware valuations.

What happens next

Federal prosecutors are seeking the full forfeiture of all funds and assets generated by Earthmade Computer through the illegal export scheme. U.S. trade regulators and law enforcement agencies are expected to increase scrutiny on freight-forwarding firms in Southeast Asia to disrupt secondary routing channels for advanced AI accelerators.


Source: Ars Technica Published on AI Usage Global, author: AUG Bot

Related

Read more

More posts that expand on the topics, companies, and AI trends covered in this story.

High-voltage transmission lines and electrical grid substation powering a hyperscale AI data center
AI Usage

MISO Proposes Fast-Track Interconnection for 200 MW+ AI Loads

MISO files a proposal with FERC for a 120-day Large Load and Associated Resource Study process to fast-track grid connections for 200 MW+ AI data centers paired with co-located power.

NetworkOcean floating solar panels in San Francisco Bay powering an AI data center
AI Usage

NetworkOcean Tests Floating Solar AI Compute in San Francisco Bay

Startup NetworkOcean powers an Nvidia H100 GPU in San Francisco Bay using a 20kW floating solar array, targeting 1GW of off-grid sea-based AI compute by 2030.

Digital representation of a data center facility in a rural landscape
AI Usage

Rural Data Centers Are in for a Big Federal Tax Break

The One Big Beautiful Bill Act introduces major federal tax breaks for rural AI data centers starting next year, though hyperscalers remain cautious amid grid limits.