Nvidia is set to launch a new artificial intelligence chip tailored for the Chinese market as early as June, according to sources familiar with the matter. The move comes in response to escalating U.S. export restrictions that have barred Nvidia from selling high-performance GPUs like the H20 in China.
The upcoming chip will be part of Nvidia’s new Blackwell-architecture lineup but will feature significantly scaled-down specifications compared to the H20. Priced between $6,500 and $8,000, it undercuts the H20’s $10,000–$12,000 range, reflecting its simpler design and reduced capabilities.
Built on the RTX Pro 6000D platform, the chip will utilize standard GDDR7 memory rather than advanced high-bandwidth memory (HBM), and will forgo Taiwan Semiconductor Manufacturing Co’s (TSMC) Chip-on-Wafer-on-Substrate (CoWoS) packaging. These choices help Nvidia comply with U.S. export rules that limit GPU performance and memory bandwidth.
The new chip is expected to deliver up to 1.7 terabytes per second of memory bandwidth—just within U.S. export control limits. Although weaker in computing power than the H20, it aims to help Nvidia maintain a foothold in China’s $50 billion data center market, which accounted for 13% of its revenue last year.
Nvidia’s market share in China has dropped from 95% before 2022 to around 50%, according to CEO Jensen Huang. He warned that continued restrictions could drive more customers to competitors like Huawei, whose Ascend 910B chip is quickly improving.
Nvidia is also reportedly developing another Blackwell-based chip, expected to enter production in September. These efforts are part of the company’s broader strategy to navigate the regulatory landscape while leveraging its CUDA platform—a major advantage in AI development ecosystems.
The company has yet to receive U.S. government approval for the new product. TSMC declined to comment, and Nvidia said its options for China remain “limited” under current restrictions.
