Investigation Raises Questions About Licenses for Chinese H200 Chip Firms Despite Shipments Below 1%
Key Facts
An investigation by The New York Times, whose details were reported by The Business Times, raised questions about U.S. scrutiny of Chinese companies licensed to handle Nvidia’s H200 chips. It said Washington approved New Matrix to assemble the chips into AI servers, although its registered Beijing office appeared empty when a newspaper reporter visited. The investigation also said New Matrix and Maginfra are licensed to assemble the chips into servers for other customers, making verification of the ultimate beneficiary a central issue in the export process. These questions arise amid U.S. restrictions requiring compliance procedures, customer screening and independent product testing. The investigation, however, did not establish that any company diverted the chips to a military entity or violated its license terms.
In New Matrix’s case, the company website listed no employees, job postings or information about its history as of November 6. The investigation linked its owners to corporate records connected to Wise Road Capital. The U.S. Commerce Department added Wise Road to the Entity List in 2024, saying the company had helped China acquire strategically important companies and semiconductor technologies. These corporate ties put the ownership background under scrutiny, but do not in themselves prove that New Matrix violated its license terms. The regulatory question centers on whether U.S. authorities can verify the licensed company and the customer that will receive the servers.
As for Maginfra, the investigation said analysts believe it is an extension of the Inspur group, noting that the two companies share employees and patents and are both state-owned. This underscores the importance of examining links between the license recipient and entities that may help operate or benefit from the products. The investigation provided no evidence that the alleged connection led to H200 chips being transferred to a prohibited entity. Under the terms reported by The Associated Press, China may not use the chips for military purposes, and its imports may not exceed 50% of the chips sold to U.S. customers. These restrictions require verification of the end user as well as screening of the intermediary company.
The investigation linked Zhejiang Yindun Cloud Technology to the Ring Intelligent Computing Technology group, saying some of Yindun’s operations are run from a data center owned by Ring and that several Yindun officials are relatives of the group’s founder. The report also said Shanghai Jiatang originated from state-owned Datang Telecom. It described Datang as a chipmaker that provides technology to the Chinese military. These remain details about ownership and operations, not proof that H200 chips were used in a military program or diverted to a prohibited entity. Shanghai Jiatang, Yindun Cloud and Maginfra did not respond to requests for comment, the investigation said.
The U.S. Commerce Department’s Bureau of Industry and Security announced on January 13, 2026, that export applications for H200 chips and similar products to China would be reviewed on a case-by-case basis. The bureau requires exporters to show that shipments will not reduce production capacity available to U.S. customers, and requires Chinese buyers to adopt compliance procedures and customer-screening measures. It also requires products to undergo independent third-party testing in the United States. These conditions put end-user verification at the heart of the review, particularly when chips are assembled into servers for another customer. The tests and checks give Washington tools to monitor compliance, while their implementation remains dependent on information about the parties in the supply chain.
Nvidia said in a regulatory filing that restrictions imposed by the Chinese government had prevented it from selling all H200 products covered by U.S. licenses. The company recorded a cost of 0.4 billion dollars related to excess inventory and purchase obligations for H200 chips during the first half of its fiscal 2027. In the latest quarter it reported, H200 shipments under the licensing program accounted for less than 1% of data center revenue. Nvidia also said the chips are inspected in the United States before shipment. The figures show the program’s limited contribution to data center revenue so far, as stalled Chinese approvals leave the company with costs tied to inventory and purchase obligations.
On the demand side, Reuters reported on November 5 that about 600,000 Nvidia AI chips were stored in Hong Kong awaiting permission to be sent to the mainland, as Beijing delayed approval for large-scale imports. The agency said major Chinese internet companies had each received about 10,000 H200 chips in previous months, pointing to limited deliveries even as other shipments remained stuck. By the end of October, a mainland server containing 8 H200 chips was listed for more than 6 million yuan, about three times its price a year earlier. The gap therefore persists between reported demand and what import procedures allow. Beijing’s approval of imports remains crucial to determining when the stored shipments will reach customers.