Nvidia to Launch AI Chipset for China Amid Export Curbs

Nvidia is set to release a new AI chipset for the Chinese market at a significantly lower price compared to the H20 models. Mass production is scheduled to begin in June, according to Reuters, citing sources.

The agency reports that the graphics processing unit (GPU) will be part of the latest generation of chips with Blackwell architecture, priced between $6500 and $8000. In comparison, H20 prices range from $10,000 to $12,000.

The lower cost reflects reduced technical specifications and simpler manufacturing requirements.

The chip is based on the RTX Pro 6000D—a server-class GPU. It uses standard GDDR7 memory instead of high-bandwidth memory, limiting the device’s potential.

The new product does not employ TSMC’s advanced Chip-on-Wafer-on-Substrate packaging technology.

“Until we choose a new design and receive approval from the US government, we are effectively barred from entering the Chinese data center market, which is valued at $50 billion,” commented an Nvidia representative.

New export restrictions on processors to China involve limits on GPU memory bandwidth—a crucial parameter for AI workloads. According to investment bank Jefferies, the rules cap the figure at 1.7–1.8 TB/s. In comparison, the H20 can deliver 4 TB/s.

Reuters also learned of another chip under development with Blackwell architecture for China, expected to be produced in September. Specifications for the processor are not yet known.

In April, the US government required Nvidia to obtain a license to export H20 to China and several other countries. The company estimated losses from this decision at $5.5 billion.