AI News Monitor

Nvidia is preparing to release a new artificial intelligence chip for the Chinese market at a significantly lower price than its previously restricted H20 model, according to three people familiar with the matter, as the U.S. semiconductor giant seeks to navigate tightening export controls while preserving its foothold in China.

The new graphics processing unit (GPU), based on Nvidia’s latest Blackwell architecture, is expected to be priced between $6,500 and $8,000—well below the $10,000 to $12,000 price tag of the H20—and could enter mass production as early as June, two of the sources said.

The lower cost reflects scaled-back specifications and simpler manufacturing requirements. Unlike Nvidia’s high-end AI chips, the new GPU will use conventional GDDR7 memory instead of advanced high-bandwidth memory (HBM) and will not require Taiwan Semiconductor Manufacturing Co.’s cutting-edge Chip-on-Wafer-on-Substrate (CoWoS) packaging technology, the sources said.

Nvidia declined to confirm specifics but acknowledged the challenges posed by U.S. restrictions. “We are still evaluating limited options,” a company spokesperson said. “Until we settle on a new product design and receive U.S. government approval, we are effectively locked out of China’s $50 billion data center market.”

A Shrinking Lead in a Critical Market

China accounted for 13% of Nvidia’s sales last fiscal year, but U.S. export controls have eroded its dominance. CEO Jensen Huang said last week that Nvidia’s market share in China has plunged from 95% before 2022 to about 50% today, with domestic rival Huawei gaining ground with its Ascend 910B chip.

“Domestic Chinese technologies like Huawei are expected to catch up with downgraded Nvidia chips within one to two years,” said Nori Chiou, an investment director at White Oak Capital Partners. “Nvidia’s remaining edge lies in its CUDA platform, which is deeply embedded in AI development.”

However, analysts caution that the new chip’s reliance on conventional memory may limit its performance for high-end AI training. Nicolas Gaudois, head of Asia technology research at UBS, said the design could fall short for some advanced applications.

Another Chip in the Works

Nvidia is also developing a second Blackwell-based chip for China, set for production as early as September, two sources said. Details remain unclear, but the company had initially explored a modified H20 before abandoning the plan due to regulatory constraints.

The April ban on the H20 forced Nvidia to write off $5.5 billion in inventory and forgo an estimated $15 billion in sales, Huang revealed in a recent interview.

The latest U.S. restrictions impose a memory bandwidth cap of 1.7-1.8 terabytes per second, far below the H20’s 4 TB/s capability. The new GPU is expected to hover near the 1.7 TB/s threshold, according to Chinese brokerage GF Securities, which speculated it may be branded the 6000D or B40.

As Nvidia races to comply with export rules, the stakes are high: losing China could cede the market to homegrown rivals—unless Washington and Beijing find a new equilibrium in the tech cold war.