Most people would probably not consider Donald Trump a technologist. This is a man who used to refer to the internet as “the cyber” and claims that he doesn’t even like to use computers. But quite often, his instincts are correct. In a storm of Truth Social posts this week, Trump rightly declared: whoever wins AI, wins.

The post was in response to Anthropic CEO Dario Amodei’s essay entitled “We Must Pace the Frontier“, which called for a slowdown on AI development, to place independent evaluators inside labs, and capability checkpoints before any new model ships. The essay ends with Amodei saying he believes “we owe it to humanity to try”. Yet if the goal is to “win” AI development, this is not the way to go about it.

There are two ways to win the AI race. One is to own the ecosystem everyone else builds on — to create the publicly open-weight model, meaning an AI model that can be downloaded and modified freely by anyone. The other is to reach superintelligence first. The problem is that Amodei’s plan threatens both.

Take open-weight models. Amodei calls for someone to always be in charge of new models: inspectors who can look inside the lab that built them, tests to pass before release, and an owner who can pull them back if something goes wrong. Open-weight models do none of that. Their entire purpose is to allow anyone to freely develop a base model. But the slowdown plan is worded in such a way as to keep that genie firmly inside the bottle.

For the US, such measures would be a strategic error. Trump is only half-right that the country is winning the AI war: in terms of open-weight models, China is already in the lead. Qwen, an open-source Chinese model, was downloaded three billion times in six months, far outpacing its rivals. The developing world is being quickly onboarded to the Chinese AI stack, giving it huge leverage. By cutting off the idea of US open-source models at the knees, the slowdown plan would willingly concede that power.

That leaves superintelligence. Amodei’s plan is, by design, a plan to slow the rate at which AI capabilities advance, and the end of that road is superintelligence. He accepts that pacing only works within America’s existing lead over China. But that requires knowing how large the lead is and how quickly China can close it. This is something that the West does not know, and has repeatedly misjudged.

In this sense, slowing AI development would be a mistake, but some are even calling to stop it altogether. Yesterday, Bernie Sanders, speaking alongside Steve Bannon, called for Trump and Xi Jinping to agree to a total ban on superintelligence. Bannon, meanwhile, labelled AI leaders’ calls for pacing “phoney”, and demanded executive action to stop a “cartel” forming among them.

It is clear that the West is losing its appetite for the race. Recent reports have traced bot farms pushing anti-data centre messaging to China, and Congress has asked the FBI to brief it on foreign campaigns against American AI. While concerns around data centres deserve answers, it is China that benefits when the West begins to turn on AI.

The same AI doomerism is rarer in China. As is pointed out in the book Machine Decision is Not Final, the Chinese term for AI means something closer to “human-made capability” than to “intelligence”, with none of the connotations that make Anglophone minds jump straight to superintelligence. It’s harder to build a doom cult around a word that means human capability.

Slowing down in this crucial race would be an act of painful self-sabotage. The answer is to win over the public, not pause the race. As Trump said, whoever wins AI wins altogether, and for the US, that winner cannot be China. So take any frontier lab calling for the ladder to be pulled up with a pinch of salt, and any anti-development movement with a pinch of Chinese spice.