Reuters reported on August 19, 2025, that Nvidia is developing a new specialized AI chip for the Chinese market with the working title B30A. This accelerator is built on the companys latest and most powerful architecture, Blackwell. The goal of the B30A is to offer Chinese customers significantly higher performance and compute density compared to the previous "export" model, H20, while remaining within the strict limitations set by the U.S. Department of Commerce. According to sources, Nvidia engineers pursued this by maintaining the chips high computational power and designing it as a single-die package; the article notes the inclusion of NVLink and high-bandwidth memory, but Reuters does not explicitly state that NVLink speed was intentionally capped. This move highlights the ongoing cat-and-mouse game between Nvidia and U.S. regulators, as well as the strategic importance of the Chinese market, for which the company is willing to create unique, finely-tuned products.
Nvidia Develops New B30A AI Chip for Chinese Market on Blackwell Architecture
