Beijing approves Nvidia's H200 chip sales as the company builds a China-ready version of its Groq inference chip

Nvidia has received long-awaited approval from Beijing to sell its second-most-powerful AI chip, the H200, to Chinese customers, Reuters reports. The company had halted production of the chip last year due to regulatory hurdles on both sides of the Pacific.

CEO Jensen Huang confirmed at a press conference during the GTC developer conference in San Jose that the company's supply chain is ramping back up. Nvidia has secured both US export licenses and orders from numerous Chinese customers, he said.

Even though the H200 is a previous-generation chip, it outperforms every alternative currently available in China. Back in January, Beijing had already granted preliminary import approvals to Bytedance, Tencent, Alibaba, and AI startup DeepSeek. China once accounted for 13 percent of Nvidia's total revenue.

Groq chip targets China's growing inference market

At the same time, Nvidia is preparing a version of its Groq AI chip specifically for the Chinese market, according to Reuters. Nvidia licensed the Groq technology late last year in a $17 billion deal. The chips are designed for inference - running AI systems that answer questions, write code, or complete tasks.

According to Reuters, these aren't stripped-down versions but chips that can be adapted to different systems. Availability is expected in May. In Western markets, Nvidia plans to pair the Groq chips with its upcoming Vera Rubin processors, which are barred from export to China.

Nvidia faces significantly more competition in the inference market than in training. Chinese companies like Baidu are already producing their own inference chips. At GTC, Huang predicted that Nvidia could generate more than one trillion dollars in revenue from Blackwell and Rubin systems alone by 2027.