Nvidia developing new AI chip for China, surpassing H20 model: Reuters
4 Min Read
Nvidia is developing a new AI chip for China, based on its Blackwell architecture, that will outperform the current H20 model.
-
The Nvidia headquarters is seen in Santa Clara, California, U.S., May 25, 2022. (AFP Photo)
Nvidia is developing a new artificial intelligence chip for the Chinese market that is expected to outperform its current H20 chip, Reuters reports, citing sources familiar with the matter.
The upcoming product is based on Nvidia’s latest Blackwell architecture and marks the company’s most significant effort yet to retain a foothold in China’s chip industry amid strict US export restrictions.
The new chip, tentatively named the B30A, will use a single-die design and deliver roughly half the computing power of Nvidia’s flagship dual-die B300 accelerator card. Despite its scaled-down configuration, it is expected to surpass the Nvidia H20 chip, which was built on the older Hopper architecture.
The chip will include high-bandwidth memory and Nvidia’s NVLink technology, similar to the H20, but with performance enhancements designed to meet demand in China. Nvidia aims to deliver test samples to Chinese clients as early as next month, sources told Reuters.
Nvidia confirmed 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.”
Read more: US secretly putting trackers in AI chip shipments to China: Reuters
How the new model compares to the H20
The Nvidia H20 chip was developed specifically for China in 2023 after US regulators imposed export restrictions. However, sales were halted earlier this year before permission was restored in July. Critics, including US lawmakers, have argued that even scaled-down models risk giving China too much access to advanced AI technology.
US President Donald Trump recently suggested that Nvidia may be permitted to sell lower-powered versions of its next-generation chips in China. He indicated that future designs could be “30% to 50% off” in computing power compared to Nvidia’s flagship models, while calling the H20 “obsolete.”
The potential rollout of the B30A depends heavily on US regulatory approval. Washington remains wary of giving Chinese companies access to cutting-edge AI technology, even in reduced form. China accounted for around 13% of Nvidia’s revenue in the past financial year, making it a critical but politically sensitive market.
The US government has also negotiated a deal requiring Nvidia and rival AMD to give 15% of revenue from advanced chip sales in China to Washington, further complicating future product strategies.
Read more: Intel secures $2 billion SoftBank investment amid turnaround efforts
Competition from Huawei in the chip industry
Nvidia’s challenges in China are compounded by growing competition from Chinese chip maker Huawei. Analysts say Huawei’s latest chips are beginning to rival Nvidia in raw computing power, although they still lag in software ecosystem support and memory bandwidth capabilities.
Meanwhile, Chinese state media have raised concerns about Nvidia’s products, warning that the company’s chips could pose security risks, claims Nvidia denies.
Reuters further states that Nvidia is also preparing to release another China-specific chip, currently called the RTX6000D, aimed at AI inference tasks. The chip uses conventional GDDR memory and offers bandwidth just below the threshold set by U.S. restrictions, ensuring it complies with export rules. Deliveries to Chinese clients are expected to begin in September.
Industry experts note that Nvidia’s strategy reflects a balancing act: meeting US regulatory requirements while keeping Chinese clients engaged in its software ecosystem to prevent a shift toward domestic alternatives. Despite uncertainty, Nvidia continues to adapt its chip manufacturing and product roadmap to navigate one of the most challenging periods in the global chips industry.
