One of the Chinese companies the US cleared to handle Nvidia’s H200 AI chips appears to have no real office. The Verge AI reports that a New York Times investigation found the approved firm, New Matrix, lists an address that turned out to be “a hollow gray shell with wires dangling from the ceiling where lights should be.” This raises a hard question about whether the Commerce Department and Nvidia are actually checking who buys these chips.
⚡ Key Takeaways
- The firm: New Matrix is on the list of Chinese companies approved to handle Nvidia’s H200 chips. It was created 17 months ago.
- The office: The Times found the listed address empty. The building’s company directory doesn’t include the firm.
- The paper trail: There’s very little about the company online. A receptionist and a manager at the industrial park’s property management agency both said they had no record of it.
- The red flag: New Matrix’s owners have ties to a Chinese private equity firm that the US government has blacklisted since 2024.
- The bigger issue: According to the Times, this isn’t a one-off. It’s part of a pattern of companies tied to entities under national-security trade restrictions getting approved to buy Nvidia chips.
🔍 What Actually Happened
The Times went looking for New Matrix and found almost nothing. The office was empty. The building directory didn’t list it. The people managing the industrial park had never heard of it.
What little the reporters did find makes it worse. The owners are linked to a private equity firm that’s been on a US blacklist since 2024. A company that sits one step away from a restricted entity ended up on an approved buyer list for some of the most valuable hardware in AI.
That’s the core problem here. It’s not just that one shell company slipped through. It looks like nobody checked closely enough to notice.
📜 How We Got Here
This story only exists because US chip policy toward China changed course.
- Under Biden: Nvidia faced restrictions on selling advanced AI chips to China. The aim was to slow China’s access to the computing power behind frontier AI models.
- Under Trump: The administration loosened those limits after Nvidia agreed to give the government a 15 percent cut of all sales.
- The lobbying push: Nvidia CEO Jensen Huang ran an intense lobbying campaign around this. He was also reportedly targeted by China’s Ministry of State Security in a spying plot.
So export control shifted from “don’t sell” to “sell, but only to vetted buyers.” That approach only works if the vetting is real. The New Matrix case suggests it may not be.
🧠 Why This Matters
What stands out to me is the incentive problem. When the government takes a 15 percent cut of every sale, both Washington and Nvidia earn more when more chips ship. Neither side has a strong financial reason to slow things down with deep background checks.
The H200 isn’t Nvidia’s newest chip, but it’s still serious hardware for training and running large AI models. If approved buyers turn out to be shell companies, chips can flow through them to the restricted entities that export rules are meant to block. A buyer list is only as strong as its weakest entry.
This matters well beyond Nvidia:
- For policymakers: Expect pressure from Congress and national security hawks for audits of the approved buyer list.
- For Nvidia: The company could face questions about its own due diligence, not just the government’s. “Commerce approved it” may not be enough cover.
- For the AI industry: Export policy is shaky. Rules that loosened quickly can tighten just as quickly if more cases like this come out.
- For competitors and investors: Any company betting on steady China revenue from AI chips should treat that revenue as politically fragile.
👀 What to Watch Next
The obvious next step is whether the Commerce Department reviews or pulls approvals, starting with New Matrix. Watch for congressional letters and hearings, and for any change to how buyers get vetted, like site visits or ownership checks.
If the Times is right that this is a pattern, the whole “sell with a revenue share” model could come under real scrutiny. You can find the full details in the original reporting from The Verge AI and The New York Times.