Nvidia is developing a China-specific artificial-intelligence accelerator, codenamed B30A, according to people familiar with the matter. The processor is based on the company’s latest Blackwell architecture and is designed to stay within U.S. export thresholds while still outperforming the Hopper-based H20 model that is currently approved for sale in China. The single-die B30A is expected to deliver roughly half the raw computing power of the dual-die B300 flagship but will retain high-bandwidth memory and NVLink interconnect technology. Specifications are still being finalized, but the chip is intended to offer higher performance than the H20 without breaching restrictions that bar Nvidia’s most advanced products from the Chinese market. Nvidia hopes to send sample units to Chinese clients for testing as early as September, the sources said. Final shipments will depend on clearance from the U.S. Department of Commerce, which last month allowed the company to resume H20 sales after striking a deal that gives Washington a 15% share of revenue from certain China chip sales. President Donald Trump has signaled he may allow scaled-down Blackwell parts to be exported, though bipartisan lawmakers remain wary. China generated about 13% of Nvidia’s revenue in the last fiscal year, making continued access to the market critical as domestic rivals such as Huawei advance their own AI silicon. Nvidia is also preparing the RTX6000D, a lower-spec Blackwell card aimed at AI inference workloads, as it seeks to defend market share while navigating tightening geopolitical controls over cutting-edge semiconductors.