Nvidia Developing Faster AI Chip for China

Nvidia is preparing to introduce the B30A, a new AI chip built on its advanced Blackwell architecture, designed specifically for the Chinese market. 
 
The move highlights Nvidia’s effort to protect its presence in the world’s second-largest AI economy, even as U.S. export restrictions tighten.
 
 
The B30A is positioned as a successor to the H20, a chip tailored to meet earlier trade rules. 
 
Offering higher performance than the H20, it still falls within the thresholds imposed by Washington, ensuring regulatory compliance.

 
The chip features a single-die design, providing about half the raw computing power of Nvidia’s flagship dual-die B300 accelerator. 
 
Despite this limitation, it incorporates high-bandwidth memory and NVLink interconnect technology, ensuring faster communication between processors.
 
Its reliance on conventional GDDR memory with 1,398 GB/s bandwidth allows Nvidia to balance strong performance with compliance requirements, avoiding restrictions on advanced memory technologies.
 
 
However, the launch faces geopolitical hurdles. 
 
The U.S. continues to limit China’s access to cutting-edge AI chips, citing risks tied to military and surveillance use.
 
Adding further complexity, the U.S. President Donald Trump has suggested he might permit scaled-down chip sales in China, while a new U.S. policy requires Nvidia and AMD to surrender 15% of Chinese chip revenues.
 
For Nvidia, the B30A underscores a delicate balancing act—preserving a multi-billion-dollar Chinese business while navigating shifting U.S. policies and competition from domestic players like Huawei.