Nvidia Forges Ahead with Powerful New AI Chip for China, Surpassing Restricted H20

Nvidia has begun work on a new artificial intelligence chip tailored for the Chinese market, drawing from its advanced Blackwell architecture to deliver performance that exceeds the H20 model currently permitted under U.S. export rules. This step reflects the ongoing tug-of-war in U.S.-China tech relations, where American companies navigate strict regulations while trying to maintain a foothold in a lucrative overseas market. China accounted for 13 percent of Nvidia’s revenue in the last fiscal year, making it a critical battleground despite escalating trade tensions.

The forthcoming chip, tentatively dubbed the B30A, adopts a single-die configuration, which is expected to provide roughly half the raw computing power of Nvidia’s flagship B300 accelerator that uses a dual-die setup. It will incorporate high-bandwidth memory and Nvidia’s NVLink for rapid data transfer between processors—features shared with the H20, which relies on the older Hopper architecture. Nvidia aims to ship samples to Chinese customers for evaluation as soon as next month, although final specifications remain under wraps.

This initiative follows Nvidia’s recent agreement with the Trump administration on a revenue-sharing arrangement, under which the company and competitors will remit 15 percent of earnings from certain advanced chip sales in China to the U.S. government. President Trump has signaled openness to permitting a diluted version of Nvidia’s next-gen chip for China, potentially with “30% to 50% off” in computing power relative to the H20, which he dismissed as “obsolete.” Yet, securing regulatory clearance from Washington remains uncertain, driven by concerns over bolstering China’s access to cutting-edge AI capabilities.

The H20 itself emerged as a response to Biden-era export curbs imposed in 2023, which barred more potent chips from reaching China. Nvidia gained approval to resume H20 sales only in July, after a suspension in April. Critics argue these restrictions have inadvertently spurred Chinese firms like Huawei to accelerate domestic alternatives, though Huawei’s offerings trail in software integration and memory bandwidth. A case in point: Chinese AI firm DeepSeek abandoned Huawei chips for its R2 model due to “persistent technical issues,” scrapping a planned May rollout and switching back to Nvidia.

Chinese state media have raised alarms about alleged security vulnerabilities in Nvidia’s products, including claims of backdoors in the H20, prompting warnings to local tech companies against purchases. Nvidia has firmly rejected these accusations, insisting no such risks exist. In a statement, the company affirmed: “We evaluate a variety of products for our roadmap, so that we can be prepared to compete to the extent that governments allow.” It added, “Everything we offer is with the full approval of the applicable authorities and designed solely for beneficial commercial use.”

Beyond the B30A, Nvidia is gearing up to supply the RTX6000D, another Blackwell-based chip optimized for AI inference tasks. Priced below the H20 with reduced specs, it employs standard GDDR memory and a bandwidth of 1,398 gigabytes per second—just under the 1.4 terabytes threshold that triggered the H20’s initial prohibition. Limited quantities are slated for Chinese delivery in September.

These developments underscore the challenges American tech giants face in balancing innovation, market demands, and national security imperatives. Conservatives have long warned that overly restrictive policies under previous administrations risk ceding ground to adversaries, allowing China to close the gap in critical technologies. With the Trump team’s revenue-sharing model, there’s potential for U.S. firms to compete abroad while funneling resources back home, but vigilance remains essential to prevent unintended tech transfers that could undermine American leadership.