The new GPU featuring Blackwell architecture is expected to be priced between $6,500 and $8,000, according to sources. In comparison, the earlier H20 model for China was priced at $10,000 to $12,000. The new chip will use conventional memory and won’t require TSMC’s advanced packaging, sources added.
Nvidia to Launch Lower-Cost Blackwell AI Chip for China Amid U.S. Export Restrictions
Nvidia plans to launch a new artificial intelligence (AI) chipset specifically for the Chinese market, priced significantly below its recently restricted H20 model, according to sources familiar with the matter. Mass production could begin as early as June.
The upcoming GPU, part of Nvidia’s latest Blackwell architecture AI processor lineup, is expected to be priced between $6,500 and $8,000—a notable decrease from the H20’s $10,000 to $12,000 range. The reduced price reflects the chip’s more modest specifications and simplified manufacturing process.
Sources say the new chip will be based on the RTX Pro 6000D, a server-grade GPU, and will utilize GDDR7 memory instead of more advanced high-bandwidth memory (HBM). It will also forgo Taiwan Semiconductor Manufacturing Co.’s (TSMC) advanced Chip-on-Wafer-on-Substrate (CoWoS) packaging technology, further streamlining production.
The pricing, technical details, and production timeline of the new GPU have not been publicly disclosed until now. Three industry sources, who requested anonymity due to media restrictions, confirmed the information.
An Nvidia spokesperson commented that the company is still reviewing its limited options. “Until we finalize a new product design and receive U.S. government approval, we remain effectively shut out of China’s $50 billion data center market,” the spokesperson said.
TSMC declined to comment.
Nvidia Faces Market Share Drop in China Amid Continued U.S. Export Restrictions AI Chip for China
China continues to be a critical market for Nvidia, contributing approximately 13% of its revenue in the past fiscal year. However, due to ongoing U.S. export restrictions aimed at curbing Chinese technological advancement, Nvidia has once again been forced to redesign its GPUs for the region—marking the third such instance.
Following the U.S. ban on the H20 GPU in April, Nvidia initially explored creating a scaled-down version of the chip for the Chinese market, sources said. However, that effort was ultimately abandoned. According to CEO Jensen Huang, the older Hopper architecture, which the H20 is based on, cannot support further modifications under current U.S. regulations.
While the final name of Nvidia’s upcoming GPU remains unconfirmed, Chinese brokerage GF Securities suggested in a note that it may be branded as the 6000D or B40, though it did not disclose sources or pricing details.
In addition, two sources revealed that Nvidia is working on another Blackwell-based GPU for China, with production expected to begin as early as September. Reuters has not confirmed the full specifications of this upcoming variant.
Nvidia’s market share in China has dropped sharply—from 95% in 2022 before the U.S. imposed export controls, to around 50% today, according to Huang. The company’s main rival in the region is Huawei, which manufactures the Ascend 910B AI chip. Huang also warned that if export curbs persist, more Chinese customers may shift to Huawei’s alternatives.
The H20 ban had major financial consequences for Nvidia. The company was forced to write off $5.5 billion in inventory and, according to Huang’s interview on the Stratechery podcast, had to abandon $15 billion in projected sales.
The latest U.S. regulations introduced strict limitations on GPU memory bandwidth, a key factor in AI performance. While the H20 offered bandwidth of 4 terabytes per second, new rules cap this at 1.7 to 1.8 TB/s. GF Securities predicts that the new Nvidia chip will reach the maximum allowable threshold of 1.7 TB/s by using GDDR7 memory, keeping it compliant with export restrictions.