NVIDIA is preparing to release a more affordable artificial intelligence (AI) chipset for the Chinese market. The new graphics processing unit (GPU) will be part of NVIDIA's latest generation Blackwell-architecture AI processors and is expected to be priced between $6,500 and $8,000. This is significantly lower than the previously restricted H20 model, which sold for $10,000 to $12,000.

The upcoming AI chipset will be based on NVIDIA's RTX Pro 6000D, a server-class graphics processor. Unlike the H20 model, the new GPU will use conventional GDDR7 memory instead of high bandwidth memory. Additionally, it will not include TSMC's advanced Chip-on-Wafer-on-Substrate (CoWoS) packaging technology. These design choices are intended to make the new chipset cheaper and easier to manufacture.

Due to US export restrictions that began in 2022, NVIDIA's market share in China has decreased from 95% to 50%. China accounted for 13% of NVIDIA's sales in the last financial year. Huawei, with its Ascend 910B chip, is NVIDIA's primary competitor. NVIDIA CEO Jensen Huang suggests that if US export curbs continue, more Chinese customers may opt for Huawei chips.

NVIDIA had to write off $5.5 billion in inventory and lost $15 billion in sales due to the H20 ban. The latest export restrictions have imposed new limits on GPU memory bandwidth, which is a critical factor for AI workloads involving extensive data processing. Jefferies, an investment bank, estimates that these regulations limit memory bandwidth to 1.7-1.8TB per second, compared to the 4TB per second that the H20 could handle.