At the White House, Chinese President Xi Jinping recently told President Donald Trump that China and America, as “leading nations in artificial intelligence,” have “both the capability and responsibility” to develop it. On Capitol Hill, meanwhile, a bill would impose “an immediate federal moratorium on AI data centers.”
Whether AI turns out to be the most dangerous technology ever built or the most beneficial, the consequence is the same: The country that builds it will decide how it is used. Optimists and pessimists disagree about almost everything else. Not that.
Two facts drive the case. First, the two camps disagree about whether this technology will be good or bad, not about how powerful it will be. Second, China is building it regardless, so an American pause would only change who leads it. Put those together and leading becomes what economists call a dominant strategy: A course that comes out ahead, however things turn out. If AI is dangerous, whoever holds it holds the danger. If it is beneficial, whoever builds it benefits first.
Eliezer Yudkowsky, among the most prominent voices warning about AI, wrote in 2023 that superhuman AI would most likely mean “literally everyone on Earth will die.” Dario Amodei, Anthropic’s chief executive, wrote in 2024 that AI could compress the medical progress of “the next 50-100 years into five to 10 years.” One expects the end of humanity, the other a century of medicine in a decade. Neither expects anything small.
The power is already visible. This spring, an OpenAI model settled a mathematics problem posed in 1946 that had defeated humans for 80 years. University of Illinois researchers found that, handed a software vulnerability’s public description, an AI agent could exploit 87% of them.
These are the early versions. The leading labs’ stated goal is artificial general intelligence, or AGI: A system that can do whatever a capable person can do at a computer. OpenAI’s charter defines it as “highly autonomous systems that outperform humans at most economically valuable work”; Amodei calls it “a country of geniuses in a datacenter.” Today’s models help a scientist or a hacker; that one would do the job without them. Something that capable will be governed by someone. The question is by whom.
Suppose the pessimists are right and AI is as dangerous as they fear. Even then, slowing down at home changes only who builds it.
The State Council’s 2017 plan set the goal of making China “the world’s primary AI innovation center” by 2030. Last year, Xi told the Politburo to “firmly maintain the initiative in AI development.” China has 32 gigawatts of data-center capacity and is adding eight more this year.
The brakes on building are mostly ours. In the second quarter alone, local opposition blocked or delayed 45 data-center projects worth $68 billion, regulatory scrutiny spread to 49 states, and the moratorium bill would halt such construction nationwide. China’s constraint is the chips it cannot buy; ours is the permits we deny ourselves.
A pause in Washington is not a pause. It is a handoff.
If AI becomes a weapon of unprecedented reach, unilateral restraint does not remove the weapon. It chooses the owner, and what separates the two candidates is not their engineers but their institutions. In the United States, a company can refuse a government request and go to court, a newspaper can report what a model is used for, and voters can replace whoever sets the rules. In China, the National Intelligence Law provides that “all organizations and citizens shall support, assist, and cooperate with national intelligence efforts.”
In one system, the most powerful technology ever built answers to courts, a free press, and voters; in the other, to the state, by law, and to no one else. A technology inherits the system it is built in, and the more powerful it becomes, the more that matters.
And if the optimists are right, leadership matters more, not less. China has already decided what its AI will carry: Its rules for generative AI, in force since August 2023, require that output “uphold the Core Socialist Values” and bar anything “harming the nation’s image.” A century of medicine in a decade is worth having on terms a free society would recognize.
SENATE DEMOCRATS JUST FILIBUSTERED YOUR POWER BILL TO SPITE REPUBLICANS
To be sure, not every rule slows development; transparency, liability, and standards for hospitals and courts slow no lab down. The rules that do are the moratorium, the licensing regime, the compute cap, the county-by-county veto, and each has the same effect: Progress slows here and nowhere else. Europe, which regulated first, produced about two notable models last year by the AI Index’s count, against 50 from the U.S. and 30 from China.
None of this settles whether the pessimists or the optimists are right. It need not. A technology this powerful will be built by someone, and every restriction on building it here has the same result: Not less AI, but AI built elsewhere, under other laws and values. That makes this a question of security as much as progress. Fear the technology or welcome it, the country that builds it decides what it becomes.
Vitor Melo is a Regulatory Policy Fellow at the Archbridge Institute and host of The Cause & Effect on YouTube. He holds a PhD in economics from Clemson University and completed postdoctoral work at the University of Chicago.