
Nvidia Set To Launch China-Specific AI Chips After US Export Ban (Image: X/nvidia)
Nvidia is once again redesigning its artificial intelligence chips to cater specifically to Chinese customers, aimed at complying with the latest US export restrictions.
According to a report by The Information, the chipmaker has informed major clients, including Alibaba, ByteDance, and Tencent, that new, tailored AI chip models could be available as early as June.
Also, read| Project DIGITS: NVIDIA Unveils Personal AI Supercomputer
The move comes just weeks after the US government imposed tighter curbs on Nvidia’s H20 chips, effectively blocking their sale to China. The company previously warned that these restrictions could cost it up to $5.5 billion in lost revenue.
During a mid-April trip to Beijing, Nvidia CEO Jensen Huang personally briefed clients on the company’s strategy to continue supporting Chinese businesses while staying within the boundaries of US regulations.
Nvidia is reportedly not only redesigning existing chips but also developing a China-specific version of its next-generation AI chip, codenamed Blackwell. Samples of the updated chips are expected to begin shipping to Chinese partners in the coming weeks.
Earlier in January, at CES 2025 in Las Vegas, the AI chips and software maker NVIDIA announced the launch of Project DIGITS, a “personal AI supercomputer” in May.
Designed mainly for AI researchers, data scientists, and students, it provides access to the company’s Grace Blackwell platform.
Also, read| Musk’s Grok Gets Its Dedicated App And Introduces Latest Version
The supercomputer will feature the new NVIDIA GB10 Grace Blackwell Superchip, which possesses enough processing power to run sophisticated AI models, at the same time, it is concise enough to fit on a desk and run from a standard electrical outlet.
The chipmaker states that a single Project Digits unit can run models up to 200 billion parameters, large language models. Besides, two Project Digits machines can be linked together to run up to 405-billion-parameter models.