Nvidia Chips Reach Chinese Firms With Ties to Blacklisted Entities

An Address That Leads Nowhere

New Matrix, a company approved to purchase Nvidia's H200 AI chips, lists an address in China that leads to an empty office. Visitors find bare walls, exposed wiring where lighting fixtures should hang, and no company nameplate in the building directory. Staff at the nearby property management office for the industrial park say they have no record of the firm.

The company's online footprint is similarly sparse. Created 17 months ago, New Matrix has left almost no digital trace beyond basic registration documents. Yet it appears on the list of Chinese entities cleared by the US Commerce Department to acquire advanced AI processors, hardware subject to export controls designed to prevent military or surveillance applications.

The gap between approval and operational reality points to a broader pattern. According to investigations into the approved buyer list, New Matrix is not an isolated case. Multiple firms granted permission to purchase Nvidia chips maintain connections to entities that US authorities have placed under national-security-related trade restrictions since 2024.

Ownership Trails and Blacklist Connections

Available corporate records link New Matrix's ownership structure to a Chinese private equity firm that has been on the US government's Entity List for two years. That designation prohibits American companies from conducting business with the firm without special licences, reflecting concerns about technology transfer or end-use in restricted sectors.

The approval of firms with such ties suggests gaps in the vetting process. Export control mechanisms typically require end-user checks, verification of physical premises, and reviews of corporate ownership to ensure compliance with restrictions. When a registered office sits vacant and ownership chains lead back to blacklisted investors, those checks appear incomplete.

Industry observers note that shell companies and nominee structures have long been used to circumvent trade controls in the semiconductor sector. The pattern documented in recent months indicates that enforcement may not have kept pace with the volume of applications or the complexity of corporate structures in China's tech ecosystem.

Policy Shifts Under the Current Administration

Nvidia faced strict limits on chip sales to China during the previous administration, part of a wider strategy to slow Beijing's progress in artificial intelligence and high-performance computing. Those restrictions covered the company's most advanced processors, including the A100 and H100 lines, and later extended to modified versions designed specifically for the Chinese market.

The current administration relaxed those constraints after Nvidia agreed to terms that include directing 15 per cent of China sales revenue to the US government. The arrangement, finalised earlier this year, reopened a market that had accounted for billions of dollars in annual revenue before the export curbs took effect.

Jensen Huang, Nvidia's chief executive, led a sustained advocacy effort in Washington, arguing that unilateral restrictions would cede the Chinese market to competitors without meaningfully slowing AI development. The lobbying campaign coincided with reports that China's Ministry of State Security had targeted Huang in an intelligence operation, though details of that incident remain limited.

The Consequence of Hollow Oversight

The discovery of phantom buyers on the approved list undermines the rationale for the policy shift. If the objective was to allow legitimate commercial activity while maintaining safeguards against military or surveillance use, then approving entities that exist only on paper defeats that purpose.

Chips routed through shell companies can be redirected to end users that would not themselves qualify for export licences. Once hardware enters China through an approved channel, tracking its final destination becomes difficult. The H200, Nvidia's latest data-centre GPU, offers performance improvements over earlier generations and is particularly suited to training large language models and other AI workloads that have both civilian and dual-use applications.

Commerce Department procedures require companies to verify the bona fides of buyers, but the burden of proof and the depth of investigation vary. Nvidia, for its part, has not publicly detailed what steps it takes to confirm that a firm on the approved list is a functioning business rather than a registered address with no staff or operations.

The issue extends beyond a single approval or one vendor. Export controls are effective only when the entities cleared to receive controlled goods are genuine, traceable, and subject to meaningful accountability. When approvals flow to companies that cannot be located at their stated addresses, the control framework becomes a formality rather than a safeguard.

What Comes Next

The Commerce Department has not commented on the specific case of New Matrix or outlined plans to audit the approved buyer list. Nvidia has likewise remained silent on its vetting protocols and whether it intends to suspend shipments to firms whose credentials are now in question.

Congressional committees with oversight of export controls may request briefings or initiate reviews, particularly if additional cases come to light. The 15 per cent revenue-sharing arrangement that underpinned the policy change could face renewed scrutiny if it becomes clear that the relaxation of restrictions enabled chip flows to entities that would have been denied licences under the previous framework.

For now, the approved list remains in force. Companies seeking to purchase Nvidia's high-end processors can apply through established channels, and the absence of a verified office or documented business operations does not appear to be disqualifying. Whether that changes will depend on how regulators and lawmakers respond to evidence that approvals have been granted to entities that exist, at best, only on paper.