Nvidia plans B30A AI chip for China under US export limits
Nvidia, the world’s most valuable semiconductor company, is pressing ahead with efforts to retain China as a crucial growth market despite stringent US restrictions on advanced technology exports. The company is developing a new AI graphics processor, the Blackwell B30A, designed specifically for China in compliance with Washington’s rules that prohibit the sale of top-tier AI chips to the country.
The B30A chip is understood to deliver half the performance of Nvidia’s flagship Blackwell B300 Ultra GPU while offering more power and memory than the restricted Hopper H20. Industry reports indicate the B30A will feature a single-die design and include up to 144GB of HBM3E memory, compared to the H20’s 96GB. By contrast, the B300 Ultra uses a dual-die configuration, achieving substantially higher throughput and up to 288GB of memory. The reduced specifications of the B30A reflect Nvidia’s attempt to strike a balance between US export compliance and the demands of Chinese customers.
The move follows months of turbulence surrounding the H20 chip. In April, the US government imposed a ban on H20 sales, which led Nvidia to record a $5.5 billion write-off. The ban was reversed three months later, allowing renewed shipments under a framework that required Nvidia to remit 15 percent of China sales revenue to the US government. Despite this reprieve, Chinese state media criticized the H20 as outdated and alleged that it contained security risks, discouraging domestic firms from adopting the chip. Nvidia has since asked suppliers to scale down H20 production as it prepares its Blackwell-based alternatives.
Reports suggest Nvidia has already presented the B30 and B30A variants to US officials for approval. The chips are said to offer up to 80 percent of the performance of a full Blackwell GPU but remain capped at levels Washington is prepared to authorize. The design incorporates features such as high-bandwidth memory and NVLink interconnects but deliberately excludes certain scaling capabilities that would enable multi-GPU clustering at the highest performance levels.
The regulatory negotiations have coincided with broader geopolitical dynamics. The US government has linked its chip export concessions to Chinese cooperation on rare-earth mineral supplies, which are critical for advanced technology manufacturing. At the same time, Beijing’s cool response to the H20 is seen by analysts as part of a strategy to promote local competitors such as Huawei, which has made progress in developing domestic accelerators but still lags behind Nvidia in software support and supply volume.
Market observers believe the B30A could secure demand in China in the short term, given its improved performance over the H20 and Nvidia’s entrenched ecosystem. However, the long-term outlook is uncertain as regulatory volatility, political mistrust, and the rise of state-backed rivals pose challenges to Nvidia’s ability to sustain its position in the market. The company is reportedly preparing a successor to the B30 series, known as the Rubin R30, scheduled for 2028, but its availability in China will ultimately depend on the trajectory of US-China technology relations.
