[DigitalToday intern reporter Hong Kyung-min (홍경민)] Nvidia has decided to stop producing artificial intelligence chips for China and concentrate its manufacturing capabilities on its next-generation platform, Vera Rubin, taking into account ongoing uncertainty over export restrictions between the United States and China.

Cryptopolitan, a blockchain media outlet, reported on Thursday that Nvidia shifted production slots at its foundry partner Taiwan's TSMC from the H200, a chip for export to China, to Vera Rubin, its next-generation architecture. With foundry capacity limited, it judged it would be difficult to secure meaningful H200 sales volumes in China in the short term, the report said.

Nvidia has faced uncertainty between U.S. semiconductor export restrictions on China and policy variables from the Chinese government. The U.S. government is also known to be considering a plan to cap the number of H200 chips Chinese companies can buy at up to 75,000 per customer. The same purchase limit may also be applied to AMD's MI325 accelerator with similar performance, the report said.

If such limits are implemented, it would be difficult for Nvidia to meet demand from major Chinese big tech companies. Chinese technology companies such as Alibaba and ByteDance have been reported to have requested volumes from Nvidia that far exceed the cap. The industry says that while shipments of up to 1,000,000 units are theoretically possible, if per-customer purchase limits are applied, actual supply could fall to the hundreds of thousands.

The H200 that Chinese companies are trying to secure is rated as the most powerful among Nvidia's previous-generation AI chips. It is widely used to train and run generative AI models and is known to deliver about six times the computing performance of products that had previously been allowed for export to China. It is also assessed to have higher performance than AI chips Chinese companies can develop on their own.

Political variables surrounding exports of advanced semiconductors to China remain. The U.S. government is considering allowing H200 exports only to Chinese companies classified as non-military users, but enforcement is expected to be difficult due to concerns that advanced semiconductors could be diverted to other uses.

Amid this uncertainty, Nvidia appears to have chosen a strategy to reduce dependence on the China market and respond to demand for next-generation AI infrastructure. The Vera Rubin architecture unveiled this year is a platform optimized for large-scale AI cluster environments based on faster computing performance and higher memory bandwidth. Rapidly growing demand from U.S. technology companies such as OpenAI and Google is also seen as having influenced the production shift decision.

The industry says this shift in production strategy could also affect the global AI semiconductor supply structure. As the U.S.-China technology rivalry continues, Nvidia is reshaping its production strategy around its next-generation platform, and competition for leadership in the AI infrastructure market is expected to intensify further.