Give China a musket, and can you stop it from making machine guns? Gavin Baker's counterargument on chip controls

In a post on X, Gavin Baker takes aim at China hawks who do not understand AI and hardware. His judgment is stark: their policies are putting U.S. national security at risk. Original tweet

First, some background. The United States restricts exports of NVIDIA's advanced GPUs to China and has also tried to block compute obtained through Southeast Asian cloud services. Supporters of controls believe this can slow China's AI progress. Gavin Baker reaches the opposite conclusion. He says that under the export strategy backed by Jensen Huang, China's chances of catching up with the United States in AI are zero. Chinese labs will continue training models on weaker American GPUs. Under scaling laws, American labs will remain ahead. In the event of a conflict, the United States can also shut off Southeast Asian cloud services.

He compares this to a generational gap in weapons: the United States has machine guns, and gives China muskets, with the aim of preventing China from building its own machine-gun variant. The analogy is striking and lays out the policy logic plainly. Export-control advocates want to maintain dependence and keep China inside the American technology stack.

But Gavin Baker argues that hawkish policies are close to delivering what their proponents want, at the cost of possibly forcing China onto another path: a less energy-efficient but self-reliant AI hardware route. According to him, if weaker GPUs remained available through Chinese and Southeast Asian clouds, China would be very unlikely to take this path. Now that controls have tightened, China has a chance to catch up. More troubling, if China catches up or takes the lead in AI and shares the technology with Russia, Iran, and North Korea, the world will become more unstable.

The replies show no consensus. Some agree, saying the whole thing is ugly and stupid. Some blame Nvidia for China's catch-up, saying it gave chips to China. Some say Chinese labs are still far behind and politicians are putting on a show. Others push back against Gavin Baker: China cannot depend forever on American chip technology; if the goal is to surpass the United States, why would China not pursue both paths at once—buying American GPUs while developing its own alternatives? Some commenters bring up past examples: the United States long sold inferior equipment, yet China still developed an equipment industry at 28nm and above, so Gavin's argument is flawed. Someone asks directly how stricter chip controls actually help China.

The replies also include discussion of White House personnel, Lutnick's position, and whether Jensen Huang will stay or go. Some say Lutnick, fearing escalation, is unwilling to restart an extreme chip ban, and now he is leaving. Others discuss Jensen Huang's future, hoping he is not resigning under pressure. Some speculate that someone inside the government leaked material to keep Jensen Huang from joining a trip. These claims lack independent sourcing and should be treated only as opinions in the replies.

At the heart of this debate is the policy goal: maintain China's dependence, or accept the cost of forcing China to develop substitutes. Gavin Baker stresses the risk of the latter. The replies remind him that substitution may already be happening. Both sides need evidence; slogans cannot substitute for it.