Nvidia H200 Sales Approved for Blacklist-Tied Firm with Empty Office

The US is approving Nvidia H200 AI chip sales to Chinese firms tied to blacklisted entities, including a company with an empty office. The pattern raises compliance and vetting questions for AI operators.

What happened

Concerns are being raised that neither the Commerce Department nor Nvidia are doing their due diligence when vetting Chinese firms cleared to handle H200 AI chips. According to The New York Times, New Matrix is on the list of approved firms, yet its offices are described as “a hollow gray shell with wires dangling from the ceiling where lights should be.”

The paper could find little evidence that the company exists in any meaningful way. New Matrix was created 17 months ago. At the building it lists as its address, a company directory does not include the firm. At the nearby property management agency for the industrial park, a receptionist and a manager both said they had no record of the company.

What little information exists about New Matrix is concerning. Its owners have ties to a Chinese private equity firm that has been blacklisted by the US government since 2024. The Times reports this is not a rare lapse but a troubling pattern of companies connected to entities with national-security-related trade restrictions being approved to purchase Nvidia chips.

Key facts

• New Matrix is approved to handle Nvidia H200 AI chips, but its listed office is empty and described as “a hollow gray shell with wires dangling from the ceiling where lights should be.”
• The firm was created 17 months ago; a company directory and property management staff have no record of it.
• New Matrix’s owners have ties to a Chinese private equity firm blacklisted by the US government since 2024.
• Nvidia had been restricted from selling chips to China during the Biden presidency, but the Trump administration relaxed those limits after Nvidia agreed to give the government a 15-percent cut of all sales.
• CEO Jensen Huang launched an intense lobbying campaign and was apparently targeted by China’s Ministry of State Security in a spying plot.

Our analysis

The reported approval pattern suggests export enforcement is shifting from a strict national-security filter to a transaction-based model. The 15-percent cut gives the US government a direct financial interest in more sales, which likely changes how aggressively it verifies buyers. For AI content operators, that could mean more H200-class compute becomes available through Chinese-linked channels, but also more uncertainty about who actually controls the hardware.

If a buyer is a shell entity with ties to a blacklisted firm, any creator platform, model provider or agency using that compute chain could face disruption if a future enforcement action freezes the entity or its partners. The reported spying plot involving Jensen Huang also signals how strategically China treats advanced AI chips; teams working across borders should assume hardware access and vendor ownership are monitored, not just code or data.

This is not only a Washington policy story. It changes the practical risk calculus for operators sourcing AI video, image and text generation capacity from resellers or cloud providers with opaque China-linked supply chains.

What it means for operators

• Verify physical and ownership records before buying AI compute. If a provider or reseller lists a China-based entity, check the actual address, company directory, property management records and blacklist ties; an empty office is a red flag.
• Add export-policy clauses to vendor contracts. Require proof of where H200-class chips are physically hosted and ask for advance notice if a supplier’s ownership or approval status changes.
• Prepare a fallback compute path. Because approval rules can shift quickly and a key vendor could be sanctioned, keep non-China-linked alternatives for critical AI content workflows.
• Screen crypto community partners with the same lens. Global teams that use Chinese counterparties for GPU-based AI or node infrastructure should treat opaque ownership as a legal and operational risk, not just a technical one.