NYT: The US Cleared a Chinese Firm to Assemble Nvidia H200 Servers. Its Listed Office Is Empty.
The US government is supposed to vet every Chinese company that touches Nvidia’s H200 AI chips. According to a New York Times investigation published this morning, at least one of the firms that passed that review has a listed office with nobody in it.
What the Times found
The company is called New Matrix. The Times reports that the Commerce Department approved it earlier this year to assemble H200 chips into the servers that train and run AI models. When reporters visited the address New Matrix lists, in a new industrial park on Beijing’s northwest edge, they describe finding an empty shell with wires hanging where lights should be.
- The building’s company directory didn’t list New Matrix, and the park’s property managers told the Times they had no record of it.
- New Matrix was created about 17 months ago, and the Times found little evidence online that it does business in China.
- Chinese corporate records cited by the Times tie its owners to a Chinese private equity fund the US placed on a trade blacklist in 2024. Secondary coverage names that fund as Wise Road.
- Two analysts told the Times they believe New Matrix may be a front for firms connected to that fund. That’s their assessment, not a finding by any government.
It isn’t the first case. The Times reported in September that the listed office of another approved firm, Maginfra, appeared deserted. Analysts have described Maginfra as a possible offshoot of Inspur, a server maker already on the US Entity List.
How H200s got to China in the first place
President Trump allowed H200 sales to China in December 2025, with the US government taking a 25% cut. In January, Commerce moved to case-by-case licensing, which was meant to screen each buyer and assembler for national security risks. The Times reports that roughly two dozen Chinese companies have been cleared so far. The full list isn’t public.
The chips aren’t exactly flooding in, either. Reuters reported on November 5 that about 600,000 H200s are sitting in Hong Kong waiting on Beijing’s import approval, citing two people familiar with the matter. H200 shipments to China made up under 1% of Nvidia’s data center revenue in the quarter that ended in July.
What Nvidia and Commerce say
According to the Times, a Commerce Department spokesperson said the agency carefully reviews every application for national security risk, and pointed to $324 million in export-violation penalties in 2025 and more than 90 enforcement officers hired this year. An Nvidia spokesperson argued it makes no sense for China’s military to depend on American technology. New Matrix and Maginfra did not respond to the Times’ requests for comment.
No one has been charged with anything, and there’s no public evidence yet that H200s approved for New Matrix ended up somewhere they shouldn’t. The story is about the screening, and whether an empty office should have passed it.
Why this lands harder this week
Nvidia is already under a spotlight in Washington. Yesterday we covered the FT report on US officials’ suspicions about a woman linked to Jensen Huang, and lawmakers have been pressing Huang for months about China sales. Expect this report to come up in those conversations.
PCGeek’s take
The H200 deal was sold as a controlled opening: the US gets a cut, and Commerce vets every company that handles the chips. A vetting process is only as good as its checks, and a reporter walking into an empty building is about the most basic check there is. If the Times is right, the problem isn’t one sketchy firm, it’s that the review didn’t look.
What we can’t confirm: who actually runs New Matrix, whether it has received any H200s yet, or whether Commerce will revisit its approval. We’ll update if any of that changes.
Should the US publish the full list of Chinese companies cleared to handle Nvidia’s AI chips? Tell us in the comments.
Sources: The New York Times (Nov 7, 2026), Reuters via FinChannel (Nov 5–6, 2026), TechCrunch (Dec 2025).
Follow @pcgeekinfo for more.
