The US Can Slow Down AI and Stay Ahead of China

American Enterprise Institute | Hal Brands

Chinese leader Xi Jinping and United States officials are preparing to address artificial intelligence safety during bilateral talks in Washington amid growing fears of recursive self-improvement in frontier models. While some American executives advocate for a coordinated global slowdown, unilateral US restraint risks ceding a critical technological advantage to Beijing.

To prevent strategic self-harm, Washington is weighing diplomatic engagement against coercive pressure to secure Chinese compliance. One proposal involves negotiating mutual limits through a dialogue led by Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng. However, Beijing views these proposals as a Cold War containment strategy. Alternatively, the US could tighten export-control regimes to restrict semiconductor manufacturing equipment and close cloud-computing loopholes. This unilateral pressure aims to starve Chinese labs of necessary computing power. Such coercion carries severe economic risks. China would likely retaliate with rare-earth export controls, collapsing the fragile bilateral détente and escalating geopolitical tensions.

Comment

The Bureau of Industry and Security's efforts to restrict Chinese access to NVIDIA H100 graphics processing units reveal the limits of physical containment in a software-driven domain. While hardware bottlenecks temporarily slow the training of large language models in Beijing, Chinese laboratories frequently bypass these constraints through algorithmic optimisation and cloud-based compute harvesting. This technological adaptation demonstrates that ASML lithography restrictions and GPU export bans cannot permanently freeze Chinese software development. Consequently, the technological gap between American systems and Chinese alternatives remains highly fluid despite strict enforcement of the Foreign Direct Product Rule.

Strategic Question for Discussion

If Chinese laboratories successfully offset hardware deficits through algorithmic optimisation, does the Foreign Direct Product Rule remain a viable mechanism for maintaining a technological lead, or does it merely accelerate indigenous Chinese innovation?

The pattern suggests that while the Foreign Direct Product Rule imposes immediate friction on Chinese development, it also incentivises domestic workarounds that reduce long-term reliance on Western supply chains. My assessment is that algorithmic efficiency gains will increasingly narrow the performance gap, rendering hardware-focused regulatory regimes less effective over time. Consequently, the technological lead will likely depend more on rapid domestic deployment cycles than on the denial of specific hardware like NVIDIA H100 GPUs.

Share your assessment in the comments below.

To read in details click here