Nvidia Develops B30A GPU for China Amid US Export Limits

Nvidia is developing a new AI graphics processing unit (GPU), codenamed B30A, specifically for the Chinese market. This component aims to balance compliance with US export regulations with the need to offer competitive performance, all as part of a strategy to maintain the company's position in a key market.

The decision to develop the B30A chip, originally reported by Reuters, is in direct response to Nvidia's need to compete effectively in the Chinese market while still circumventing restrictions imposed by the US government. According to the same report, the company is looking to offer a substantially more powerful product than its currently authorized offerings for China, without violating performance limits that would trigger an export ban.

"We evaluated a wide variety of products for our roadmap so that we can be prepared to compete to the extent that governments allow. Everything we offer has the full approval of the relevant authorities and is designed exclusively for beneficial commercial use," Nvidia says in a press release.

This stance highlights the delicate balance the company must maintain. It must adhere to the US national security directives, under the context of the Chips Security Act, as well as not cede a critical market to local competitors, mainly Huawei, which has gained ground due to restrictions imposed on Western hardware.

From Prohibition to Limited Trade 

The situation is the result of several changes in US export control policy. Initially, former US President Joe Biden’s administration implemented a broad ban on the sale of advanced AI chips to China.

According to Graham Webster, Researcher, Stanford University, this move was based on the belief that AI was approaching a tipping point that could give a nation significant military and economic advantages. The goal was to prevent China from reaching these technological thresholds first by controlling access to high-performance semiconductors.

This policy led Nvidia to design chips with reduced capabilities, such as the H800 and later the H20, to continue operating in China. However, the US strategy changed under President Donald Trump’s administration. Following meetings between Jensen Huang, CEO, Nvidia, and Trump during last July 2025, an agreement was reached allowing the sale of the H20 chip. In return, the US government would receive 15% of the revenue from these sales.

President Trump himself called the H20 "obsolete," but acknowledged its market value, thus justifying the trade agreement. This policy change partially reopened the market for Nvidia, but also highlighted the need for a more advanced product than H20. The goal was to meet the computing demand of Chinese technology companies without matching the performance of its most powerful architecture, Blackwell, whose export remains restricted. Trump indicated that he would not allow the export of Blackwell technology unless its performance was reduced by 30%-50%.

Technical Specifications of the B30A

The B30A is designed to be significantly more powerful than the H20, but will be positioned below the B100 and B200 GPUs in the Blackwell architecture. According to unnamed sources cited by Reuters, the B30A will be based on a single-chip design, as opposed to the dual-chip design used in more powerful models.

However, it will inherit key features of the new architecture, such as support for the NVLink high-speed interconnect and high-bandwidth memory (HBM). These features will improve data transmission and efficiency in multi-GPU systems. This technical profile positions it as an attractive solution for Chinese companies that require more computing power to train AI models but cannot access top-tier hardware.

Future Opportunities

Nvidia's success is not solely dependent on Washington's approval. The company faces growing technology nationalism in China. For example, China has incentivized companies to avoid Nvidia's H20 chips, urging them to consume local alternatives, reports Bloomberg. Chinese officials have expressed concern about features of Nvidia's chips, such as location tracking and remote shutdown capabilities, labeling them as a potential risk. Nvidia has responded that its commercial products are not intended for military or government infrastructure and that China has a sufficient supply of local chips for such operations.

The United States continues to debate the export of chips to China. Figures such as Dario Amodei, CEO of Anthropic, and Matt Pottinger, former Deputy National Security Advisor, advocate strict export controls. They argue that controls are the only means of preventing China from developing AI capabilities that could threaten US leadership.

Others such as David Sacks, an AI advisor to the Trump administration, argue that "overly restrictive" controls are counterproductive. Sacks says that these restrictions cede market share to Chinese companies and allow them to bolster their own technology ecosystem.