NvidiaĀ is developing aĀ newĀ AIĀ chipĀ forĀ ChinaĀ based on its latest Blackwell architecture that will be more powerful than theĀ H20Ā model it is currently allowed to sell there, two people briefed on the matter said.
U.S. President Donald Trump last weekĀ opened the doorĀ to the possibility of more advancedĀ NvidiaĀ chips being sold inĀ China. But theĀ sourcesĀ noted U.S. regulatory approval is far from guaranteed amid deep-seated fears in Washington about givingĀ ChinaĀ too much access to U.S. artificial intelligence technology.
TheĀ newĀ chip, tentatively known as the B30A, will use a single-die design that is likely to deliver half the raw computing power of the more sophisticated dual-die configuration inĀ Nvidia’s flagship B300 accelerator card, theĀ sourcesĀ said.
A single-die design is when all the main parts of an integrated circuit are made on one continuous piece of silicon rather than split across multiple dies.
TheĀ newĀ chipĀ would have high-bandwidth memory andĀ Nvidia’s NVLink technology for fast data transmission between processors, features that are also in theĀ H20Ā – aĀ chipĀ based on the company’s older Hopper architecture.
TheĀ chip’s specifications are not completely finalized butĀ NvidiaĀ hopes to deliver samples to Chinese clients for testing as early as next month, said theĀ sourcesĀ who were not authorized to speak to media and declined to be identified.
NvidiaĀ said in a statement: “We evaluate a variety of products for our roadmap, so that we can be prepared to compete to the extent that governments allow.”
“Everything we offer is with the full approval of the applicable authorities and designed solely for beneficial commercial use,” it said.
The U.S. Department of Commerce did not respond to a Reuters request for comment.
Flashpoint
The extent to whichĀ China, which generated 13% ofĀ Nvidia’s revenue in the past financial year, can have access to cutting-edgeĀ AIĀ chips is one of the biggest flashpoints in U.S.-Sino trade tensions.
NvidiaĀ only received permission in July to recommence sales of theĀ H20. It was developed specifically forĀ ChinaĀ after export restrictions were put in place in 2023, but the company was abruptly ordered to stop sales in April.
Trump said last week he might allowĀ NvidiaĀ to sell a scaled-down version of its next-generationĀ chipĀ inĀ ChinaĀ after announcing an unprecedented deal that will seeĀ NvidiaĀ and rival AMDĀ give the U.S. government 15% of revenue from sales of some advancedĀ chips inĀ China.
AĀ newĀ NvidiaĀ chipĀ forĀ ChinaĀ might have “30% to 50% off”, he suggested in an apparent reference to theĀ chip’s computing power, adding that theĀ H20Ā was “obsolete”.
U.S. legislators, both Democratic and Republican, have worried that access to even scaled-down versions of flagshipĀ AIĀ chips will impede U.S. efforts to maintain its lead in artificial intelligence.
ButĀ NvidiaĀ and others argue that it is important to retain Chinese interest in itsĀ chips ā€” which work withĀ Nvidia’s software tools ā€” so that developers do not completely switch over to offerings from rivals like Huawei.
Huawei has made great strides inĀ chipĀ development, with its latest models said to be on par withĀ NvidiaĀ in some aspects like computing power, though analystsĀ sayĀ it lags in key areas such as software ecosystem support and memory bandwidth capabilities.
ComplicatingĀ Nvidia’s efforts to retain market share inĀ China, Chinese state media have also in recent weeks alleged that the U.S firm’sĀ chips couldĀ pose security risks, and authoritiesĀ have cautionedĀ Chinese tech firms about purchasing theĀ H20.Ā NvidiaĀ says itsĀ chips carry no backdoor risks.
NvidiaĀ is also preparing to start delivering a separateĀ newĀ China-specificĀ chipĀ based on its Blackwell architecture and designed primarily forĀ AIĀ inference tasks, according to two other people familiar with those plans.
ReutersĀ reported in MayĀ that thisĀ chip, currently dubbed the RTX6000D, will sell for less than theĀ H20, reflecting weaker specifications and simpler manufacturing requirements.
TheĀ chipĀ is designed to fall under thresholds set by the U.S. government. It uses conventional GDDR memory and features memory bandwidth of 1,398 gigabytes per second, just below the 1.4 terabyte threshold established by restrictions introduced in April that led to the initialĀ H20Ā ban.
NvidiaĀ is set to deliver small batches of RTX6000D to Chinese clients in September, said one of the people.
