US cleared a Chinese firm with an apparently empty office to handle Nvidia H200 chips

The US Commerce Department approved a Beijing company called New Matrix to assemble Nvidia's H200 AI chips into servers. According to The New York Times, the office the company lists as its address is empty: "a hollow gray shell with wires dangling from the ceiling where lights should be."

The finding matters because the H200 licensing system rests on the idea that the US government knows who receives these chips and what they use them for. A license holder that cannot be found at its own address suggests that the checks may be thinner than the policy assumes.

What the Times found

New Matrix was created 17 months ago, and very little about it can be found online. The Times reports that the company directory at its listed building does not include it. At the nearby property management agency for the industrial park, a receptionist and a manager both said they had no record of the firm.

Chinese corporate records link New Matrix's owners to Wise Road, a private equity fund the US government placed on its entity list in 2024. Two analysts told the Times that New Matrix may be a front for firms tied to that fund. The Times did not establish who ultimately controls the company. New Matrix did not respond to requests for comment.

It is not the only case. The Times reports that in September it found the listed office of another approved firm, Maginfra, apparently deserted. Analysts told the paper they believe Maginfra is an offshoot of Inspur, a server maker on the entity list that works with the Chinese military. Two more licensees, Zhejiang Yindun Cloud Technology and Shanghai Jiatang, have owners with ties to firms that serve the Chinese military, according to the Times. The paper describes this as a pattern, not a single lapse.

The Times did not establish that any H200 chips have been diverted to the Chinese military.

How the licensing works

Nvidia was restricted from selling advanced AI chips to China under the Biden administration. The Trump administration allowed H200 sales to China in December 2025, on the condition that the US government receives a cut of the revenue. Nvidia CEO Jensen Huang had lobbied intensively for the change.

Under the system, the Commerce Department approves specific Chinese companies to buy or handle the chips. People familiar with the approvals told the Times that about two dozen Chinese firms have been cleared, including Baidu, BYD and Tesla's Shanghai unit. Each was approved to buy tens of thousands of chips. The full list has not been published.

So far the volumes are small. H200 shipments to China-based customers made up less than 1 percent of Nvidia's data center revenue in the quarter that ended in July, according to the reporting. Nvidia says Beijing has limited H200 imports to favour domestic chipmakers.

What Commerce and Nvidia say

A spokesperson for the Bureau of Industry and Security, the part of the Commerce Department that runs export controls, said it carefully reviews every application for national security risk. The bureau pointed to $324 million in export-violation penalties imposed in 2025 and more than 90 enforcement officers hired this year.

Nvidia spokesman John Rizzo told the Times that "it makes no sense for the Chinese military to depend on American technology." Nvidia has not publicly explained how New Matrix was vetted.

Liza Tobin, a former China director on the White House National Security Council, told the Times she has "a ton of concerns" that neither the Commerce Department nor Nvidia is doing the due diligence needed to make sure licensing terms are followed.

The report adds to other scrutiny of Nvidia's China business. The Verge notes that Huang was apparently targeted by China's Ministry of State Security in a spying plot, and three Republican House committee chairs have asked Huang to brief them on November 18.

What the case shows about access to compute

Export controls on AI chips work as a gate. They decide which organisations get large amounts of computing power and which do not. The gate has two parts: the rule that says who qualifies, and the checks that confirm an applicant is who it says it is. The New Matrix case concerns the second part. A rule that bars entity-listed firms has little effect if a newly created company with the same owners can pass review on paper.

The practical result is that access to advanced compute may depend less on what a company does and more on how it presents itself in an application. Firms with real operations and public records, such as Baidu or BYD, are visible to reviewers. A company that exists mainly as a registration is harder to assess, and the Times reporting suggests it can still be approved.

What it means for work

The work affected here is verification: confirming that a license applicant has real premises, real staff and owners who are not on a restricted list. The Times did much of this with a site visit, a building directory and a call to a property manager, steps that are slow and manual but apparently were not decisive in the licensing process. For export compliance staff at Commerce and at chipmakers, the case points to more on-the-ground checks and closer tracing of ownership through corporate records. For the AI developers who depend on these chips, it means the line between who is allowed to build large models and who is not is drawn partly by paperwork, and that paperwork can be gamed. What remains open is who controls New Matrix, whether any chips reached restricted users, and whether Commerce will revisit the licenses it has granted.