Nvidia H200 Approved for Sale to China, US Review Accused of Laxity: Shell Company Reportedly on the List

The US review process allowing Nvidia to export H200 AI chips to China is facing scrutiny. According to a New York Times report, a Chinese company called New Matrix appears on the list of approved recipients of H200 chips, yet its office space is completely empty, described as "an empty gray shell." The discovery has raised concerns about whether the US Department of Commerce and Nvidia have been negligent in their due diligence.

A "Shell" Company on the Approved List

According to the New York Times investigation, New Matrix was included on the list of approved companies permitted to handle Nvidia's H200 AI chips. However, when reporters visited its office location, they found it completely empty with no signs of actual operations—nothing but "an empty gray shell." The scene is starkly at odds with the infrastructure and staffing one would expect from a company capable of handling high-end AI chips.

The H200 is Nvidia's high-performance chip for AI training and inference, and its export is strictly controlled by the US government. Any Chinese company approved to receive such chips should in theory have been vetted by the US Department of Commerce to ensure its end use and end user comply with export control requirements. New Matrix's "shell" status directly undermines the credibility of that vetting mechanism.

The H200 is an AI accelerator chip based on Nvidia's Hopper architecture, released in 2024 and positioned between the H100 and Blackwell-architecture products. It features 141GB of HBM3e high-bandwidth memory with roughly 4.8TB/s of memory bandwidth, offering significant gains in memory capacity and bandwidth over the H100 for large-model training and inference workloads. Precisely because its compute and memory specs fall within the high-performance range of export control concern, H200 exports to China require an export license from the US Department of Commerce's Bureau of Industry and Security (BIS) and are subject to end-use and end-user verification. This also means that companies on the approved list should in theory possess compute infrastructure, data center conditions, and professional teams commensurate with the chips' intended use.

A Double Failure in the Review Process

The report points out that the problem may lie at both ends of the chain: the US Department of Commerce and Nvidia.

On one hand, the Commerce Department is responsible for issuing export licenses and conducting background checks on approved entities. If a company with an empty office and no trace of actual business activity can pass review, then the review standards and enforcement clearly have gaps. On the other hand, Nvidia, as the chip supplier, has a responsibility to conduct due diligence on customers and confirm that they have the capability and qualifications to use the chips legitimately.

The New York Times report suggests that neither government-level regulatory review nor corporate-level customer verification worked as it should in the case of New Matrix. This is not a mere procedural flaw but a substantive issue concerning the effectiveness of export controls.

"End-use checks" are a key component of the export control system, typically carried out by the Bureau of Industry and Security (BIS) or through export control officers stationed overseas, to confirm that approved entities actually exist, have the capacity to receive and use controlled items, and that the items will not be resold or diverted to restricted purposes. Such checks often rely on site visits, document reviews, and follow-up tracking. When an approved entity shows no signs of actual operations, the reliability of the verification mechanism is directly challenged. At the same time, chip suppliers' customer due diligence typically includes understanding the nature of the customer's business, ownership structure, and end-use statements, but in cross-border transactions, suppliers have limited ability to verify customers' actual operating conditions, relying more on contractual commitments and government approval as endorsement.

Cracks in Trust in Export Controls

Nvidia's H200 exports to China are themselves a sensitive issue. While allowing the exports, the US government seeks to manage risk through approval lists and end-use checks. However, the New Matrix case shows that the approval list on paper may be disconnected from reality.

If an approved entity can be a shell company, then the chips' actual destination may drift outside regulatory oversight. This possibility not only weakens the practical effect of export controls but also creates compliance and reputational risks for Nvidia. For those watching the AI chip supply chain and geopolitical tech competition, this incident raises a direct question: in the absence of on-site verification and ongoing supervision, to what extent can export approvals ensure that technology is not misused?

Currently, neither the US Department of Commerce nor Nvidia has made a public response regarding the specific situation of New Matrix. But the New York Times report has brought a structural problem to the fore: when the names on an approval list are disconnected from actual operating conditions, cracks appear in the foundation of regulatory trust.