In a major policy shift on December 8, U.S. President Donald Trump announced that Nvidia (the world’s leading designer of AI accelerators) is allowed to export advanced H200 chips to “approved customers” in China. However, more intriguing is the fact that China – despite lagging in the area of high-end chip design for at least two generations – has responded by limiting domestic buyers’ access to these advanced AI processors.

To understand China’s refusal, one must look beyond the silicon to the software. Nvidia’s dominance is secured not just by chip performance but by CUDA (Compute Unified Device Architecture), a proprietary software platform that has become the global standard for AI development. If Chinese AI developers were allowed to purchase the H200 chips, they would inevitably build their next generation of models on CUDA, creating a self-reinforcing cycle in which Chinese developers optimize for American standards.

In the United States, two arguments had been articulated in support or opposed of the decision to export Nvidia’s products to China. The first argument holds that the more Chinese firms buy Nvidia chips and become dependent on the CUDA ecosystem, the more difficult it will be for leading Chinese chipmakers like Huawei to develop its own alternative standards. The counter-argument says that the U.S. should outright restrict any sales of advanced AI accelerators (whether by Nvidia, AMD, or Intel) to China, thereby hurting its AI development. 

U.S. strategy has been dominated by the second argument for the past........

© The Diplomat