The White House has informed other federal agencies that it will not permit Nvidia to sell its latest scaled-down AI chips to China, The Information reported on Thursday, citing three people familiar with the matter.
Nvidia has provided samples of the chip to several of its Chinese customers, according to the report.
The chip, known as the B30A, can be utilized to train large language models when efficiently arranged in large clusters, a capability many Chinese companies require, the report added.
An Nvidia spokesperson told Reuters that the company has “zero share in China’s highly competitive market for datacenter compute, and do not include it in our guidance.”
White House did not immediately respond to Reuters’ request for a comment.
Nvidia is working on modifying the B30A’s design in hopes that the U.S. administration will reconsider its stance, The Information report said, citing two company employees.
The California-based company, however, has also been facing regulatory headwinds in China.
Beijing has recently issued guidance requiring all new data center projects that receive any state funding to use only domestically developed chips, Reuters reported on Wednesday, citing sources familiar with the matter.
Data centers that are less than 30% complete will have to remove all installed foreign chips, or cancel plans to purchase them, while projects in a more advanced stage will be reviewed on a case-by-case basis, the sources added.
The guidance effectively shuts out Nvidia and its AI chips from a lucrative market segment, including advanced models under U.S. export controls that are nevertheless available in China via grey market channels.
