JAKARTA - Nvidia will launch a new artificial intelligence (AI) chipset for China at a much cheaper price than the recently restricted H20 model. The US company plans to start mass production as soon as June.
The GPU (graphic processing unit) is part of Nvidia's latest generation AI processor with Blackwell architecture and is expected to be sold for between 6,500 and 8,000 US dollars, well below the H20 price sold for between 10,000 and 12,000 US dollars.
This lower price reflects weaker specs and simpler manufacturing needs. The chip will be based on the Nvidia RTX Pro 6000D, a server class graphics processor, and using conventional GDDR7 memory, not a more sophisticated high bandwidth memory, according to two sources.
They added that this chip will not use advanced Chip-on-Wafer-on-Substrate (CoWoS) technology from Taiwan Semiconductor Manufacturing Co (TSMC).
A Nvidia spokesman said the company was still evaluating their "limited" choice. "Until we set a new product design and got approval from the US government, we could not effectively access China's 50 billion data center market," the spokesman said.
China remains a big market for Nvidia, accounting for 13% of last financial year's sales. This is the third time Nvidia has had to adjust the GPU to the world's second-largest economy after restrictions from US authorities to hinder China's technological development.
Following the US ban on H20 in April 2025, Nvidia initially considered developing a version of the H20 that was passed down specification specifically for China, but the plan failed.
Nvidia CEO Jensen Huang said last week that the older flaring architecture used by H20 could no longer be modified further due to current US export restrictions.
China's broker, GF Securities, said in a note Tuesday that the new GPU is likely to be named 6000D or B40, although it does not disclose the price or source of information.
According to two sources, Nvidia is also developing another Blackwell architectural chip for China which is scheduled to start production in September. Reuters has not been able to confirm the specifications of the variant.
"Nvidia's market share in China dropped dramatically from 95% before 2022, when US export restrictions came into effect, to 50% at the moment," Huang told reporters in Taipei this week. The main competitor was Huawei with the Ascend 910B chip.
SEE ALSO:
Huang also warned that if US export restrictions continued, more Chinese subscribers would switch to Huawei chips.
The H20 ban forced Nvidia to record a supply loss of 5.5 billion US dollars, and Huang told the Stratechery podcast on Monday that the company should also release a potential sale of 15 billion US dollars.
Recent export restrictions introduce new limits on the metric GPU memory bandwidth which measures the speed of data transmission between major processors and memory chips. This capability is very important for AI workloads that require massive data processing.
Jefferies' investment bank estimates a new regulation limits memory bandwidth to 1.7-1.8 terabytes per second, compared to H20's ability to reach 4 terabytes per second.
GF Securities projects that the new GPU will reach around 1.7 terabytes per second using GDDR7 memory technology, just according to export control limits.
The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language. (system supported by DigitalSiber.id)
