Nvidia H200 assembler New Matrix lists an empty Beijing address

The Trump administration approved a Chinese company called New Matrix earlier this year to assemble Nvidia’s advanced H200 chips into AI computers. According to The New York Times, the address the company lists is an empty building.

It sits in a recently built industrial park on Beijing’s northwest edge. The Times describes a hollow gray structure with wires hanging from the ceiling where light fixtures should be.

What the approval covers and why it matters

New Matrix’s role is assembly. It was cleared to take Nvidia’s H200 processors and build them into the computer systems used to train and run artificial intelligence models.

The H200 became legally available to Chinese buyers after President Trump allowed its sale to China late last year. The policy shift came on December 8, 2025. Under its terms, H200 exports could proceed with a 25% cut of the revenue going to the US.

The process then became more formal. Beginning in mid-January 2026, the US Commerce Department moved to case-by-case reviews of Nvidia’s chip exports. Each license, in theory, is supposed to involve scrutiny of who is buying and what they plan to do.

The H200 pipeline has been clogged on both ends

In January 2026, Beijing conditionally approved around 400,000 H200 chips for companies including Alibaba, ByteDance, and Tencent. Large-volume import requests since then have reportedly continued to run into obstacles on the Chinese side.

As of early November 2026, an estimated 600,000 H200 chips are reported to be stockpiled in Hong Kong, waiting on Chinese import approvals that have not materialized at scale.

The financial picture for Nvidia reflects the bottleneck. H200 sales in China accounted for less than 1% of the company’s data-center revenue in the July 2026 quarter.

Meanwhile, prices for AI servers on the gray market have reportedly been climbing.

Background: a policy built on trust and verification

The December 2025 decision marked a notable turn in US policy toward AI chips. Rather than a blanket prohibition on advanced Nvidia hardware reaching China, the administration chose a licensed, revenue-sharing model with selective approvals for specific Chinese firms.

Beijing, for its part, has its own priorities. The Chinese government has increasingly emphasized building domestic semiconductor capabilities.

What this means for Nvidia, regulators, and the AI chip market

For Nvidia investors, the direct revenue stakes are currently small. With H200 sales in China below 1% of data-center revenue in the July 2026 quarter, the business already contributes little.

The 600,000 chips reportedly parked in Hong Kong are worth watching closely. If Beijing loosens import approvals, that inventory could convert to sales relatively quickly.

The key things to watch next: whether US officials respond to the Times report or review New Matrix’s approval, whether Beijing approves larger H200 import volumes for firms like Alibaba, ByteDance, and Tencent, and whether Nvidia’s next reporting shows any movement in China data-center sales.