US President Donald Trump finally agreed to Nvidia CEO Jensen Huang’s plea that the company be allowed to sell H200s to China, estimated to be a US$10 billion opportunity. Although Trump claimed that Chinese President Xi Jinping responded “positively”, early signs indicate that China may want to limit H200s imports to reduce reliance on the US and develop domestic capability.
The White House’s artificial intelligence (AI) tsar, David Sacks, said that China is rejecting H200s, and the Financial Times reported that China is set to limit access. Meanwhile, China has added domestic AI chips to its official procurement list for the first time, and it is considering the largest-ever state-backed semiconductor incentives programme conceived in history.
Notwithstanding reports that Alibaba Group Holding, which owns the Post, and ByteDance plan to buy H200s, Beijing’s disposition is unmistakable: for self-reliance, buy local AI chips and systems.
Advertisement
While the procurement list addition indicates confidence in Chinese AI systems, the government and relevant stakeholders know such systems are not there yet. So while it is unlikely that there will be a total ban on H200s, the window for importing Nvidia’s chips may be closing.
Constraints unleashed by US export controls have helped drive significant leaps in Chinese AI systems. Former US national security adviser Jake Sullivan’s doctrine held that the US should maintain “as large of a lead as possible” in key technologies. Export controls were a key strategy aimed at delaying Chinese AI progress.
Advertisement
These restrictions, in cooperation with the Netherlands on lithography equipment and Japan on chipmaking tools, proved effective initially. YMTC, China’s largest flash memory chipmaker, was hit hard and reportedly laid off 10 per cent of its workforce. The country’s leading foundry, SMIC, warned of mass production postponements. Nvidia’s graphics card price soared.