Nvidiaย is developing aย newย AIย chipย forย Chinaย based on its latest Blackwell architecture that will be more powerful than theย H20ย model it is currently allowed to sell there, two people briefed on the matter said.
U.S. President Donald Trump last weekย opened the doorย to the possibility of more advancedย Nvidiaย chips being sold inย China. But theย sourcesย noted U.S. regulatory approval is far from guaranteed amid deep-seated fears in Washington about givingย Chinaย too much access to U.S. artificial intelligence technology.
Theย newย 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 inย Nvidia’s flagship B300 accelerator card, theย sourcesย 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.
Theย newย chipย would have high-bandwidth memory andย Nvidia’s NVLink technology for fast data transmission between processors, features that are also in theย H20ย – aย chipย based on the company’s older Hopper architecture.
Theย chip’s specifications are not completely finalized butย Nvidiaย hopes to deliver samples to Chinese clients for testing as early as next month, said theย sourcesย who were not authorized to speak to media and declined to be identified.
Nvidiaย 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 whichย China, which generated 13% ofย Nvidia’s revenue in the past financial year, can have access to cutting-edgeย AIย chips is one of the biggest flashpoints in U.S.-Sino trade tensions.
Nvidiaย only received permission in July to recommence sales of theย H20. It was developed specifically forย Chinaย 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 allowย Nvidiaย to sell a scaled-down version of its next-generationย chipย inย Chinaย after announcing an unprecedented deal that will seeย Nvidiaย and rival AMDย give the U.S. government 15% of revenue from sales of some advancedย chips inย China.
Aย newย Nvidiaย chipย forย Chinaย might have “30% to 50% off”, he suggested in an apparent reference to theย chip’s computing power, adding that theย H20ย was “obsolete”.
U.S. legislators, both Democratic and Republican, have worried that access to even scaled-down versions of flagshipย AIย chips will impede U.S. efforts to maintain its lead in artificial intelligence.
Butย Nvidiaย and others argue that it is important to retain Chinese interest in itsย chips โ which work withย Nvidia’s software tools โ so that developers do not completely switch over to offerings from rivals like Huawei.
Huawei has made great strides inย chipย development, with its latest models said to be on par withย Nvidiaย in some aspects like computing power, though analystsย sayย it lags in key areas such as software ecosystem support and memory bandwidth capabilities.
Complicatingย Nvidia’s efforts to retain market share inย China, Chinese state media have also in recent weeks alleged that the U.S firm’sย chips couldย pose security risks, and authoritiesย have cautionedย Chinese tech firms about purchasing theย H20.ย Nvidiaย says itsย chips carry no backdoor risks.
Nvidiaย is also preparing to start delivering a separateย newย China-specificย chipย based on its Blackwell architecture and designed primarily forย AIย inference tasks, according to two other people familiar with those plans.
Reutersย reported in Mayย that thisย chip, currently dubbed the RTX6000D, will sell for less than theย H20, reflecting weaker specifications and simpler manufacturing requirements.
Theย chipย 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 initialย H20ย ban.
Nvidiaย is set to deliver small batches of RTX6000D to Chinese clients in September, said one of the people.
