Nvidia new China AI chip will enter mass production by June, according to multiple insiders. The GPU belongs to Nvidia’s latest Blackwell-architecture family and complies with new U.S. export restrictions. It will sell for much less than the banned H20 model.
The chip’s price ranges between $6,500 and $8,000, while the H20 cost $10,000–$12,000. That lower price reflects its reduced performance and simpler production process. Nvidia built it using the RTX Pro 6000D and chose GDDR7 memory over more advanced high-bandwidth memory. The chip does not use Taiwan Semiconductor Manufacturing Co’s (TSMC) CoWoS packaging technology.
Sources confirmed the production timeline, architecture, and pricing. They requested anonymity since they lack authorization to speak publicly.
An Nvidia spokesperson said the company continues to evaluate its options. “Until we finalize the design and secure U.S. government approval, we cannot access China’s $50 billion data center market,” the spokesperson said.
China remains critical for Nvidia, contributing 13% of total sales in the last financial year. This marks the third redesign Nvidia has created for China, responding to U.S. export restrictions aimed at slowing Chinese tech development.
Despite reduced computing power, the Nvidia new China AI chip helps the company maintain its position in China. Nvidia competes directly with Huawei, which offers the Ascend 910B chip. Industry experts believe Huawei could match Nvidia’s downgraded GPUs within two years.
“Nvidia’s edge lies in its ability to integrate AI clusters through the CUDA platform,” said Nori Chiou, a semiconductor expert at White Oak Capital Partners. CUDA allows developers to build and optimize AI software specifically for Nvidia’s GPUs, forming a loyal developer base.
Jensen Huang, Nvidia’s CEO, recently said their Chinese market share fell from 95% in 2022 to 50% now. He warned that ongoing export controls could push more customers toward Huawei.
Nvidia also plans to release another Blackwell-based GPU for China by September. Details about this upcoming chip remain undisclosed for now.
In April, the U.S. banned the H20, forcing Nvidia to abandon $15 billion in potential sales and write off $5.5 billion in inventory. Huang said they initially tried modifying the H20, but the older Hopper architecture no longer met export rules.
New regulations limit GPU memory bandwidth to 1.7–1.8 TB/s, compared to H20’s 4 TB/s. Analysts expect the upcoming chip to hit that cap using GDDR7, just staying within the allowed limits. GF Securities speculated that Nvidia might name it 6000D or B40, though the company has not confirmed this.
The Nvidia new China AI chip may not match its predecessor’s power, but it ensures Nvidia stays in the game. With strong software integration and loyal developers, the company still holds a valuable position in China’s rapidly evolving AI market.
