US-China AI Relations: The Pax Machina Plan

For most of the last two thousand years, China was the richest and most technically advanced society on earth. The brief span in which it was not, roughly from the Opium Wars through the late twentieth century, is the exception that a whole generation of Americans mistook for the rule. Podcast host Nathan Labenz spent two weeks traveling through China this year, and he returns with a blunt correction: the country's return to the technological frontier is not a plot twist. It is a reversion to the mean.

This is the third installment of Labenz's China series, and it is the argumentative one. The first two episodes described what he saw on the ground and the safety culture inside Chinese AI labs. This one asks a question he says almost never gets asked out loud in Washington: what, exactly, is the goal of the US-China relationship? Not slogans about winning, but an actual end state. His answer is what he calls a Pax Machina, or Pax Robotica: a world in which both countries, and everyone else, get to enjoy cheap energy, cured diseases, and robot labor without also inheriting a new nuclear-style sword hanging over humanity (00:27, 10:51).

Labenz is careful to note that this puts him at odds with Anthropic, the company that sponsors his podcast. He admires Dario Amodei's essay Machines of Loving Grace, and he separately admires Amodei's own 2017 warning that races raise the odds of safety catastrophes.

"Technology races can raise the risk of safety catastrophes." — Nathan Labenz, quoting Dario Amodei (14:11)

Labenz's complaint is that Anthropic's China policy contradicts Anthropic's own founder. If racing is dangerous, then a strategy built on outracing China toward decisive advantage is a strategy built on the very dynamic Amodei once warned against.

The Mandate of Heaven, Not the End of History

Much of the episode is an exercise in what Labenz calls cognitive empathy: trying to see the relationship the way Beijing sees it. The starting point is the "century of humiliation," the period from Western and Japanese incursions through the chaos of the Cultural Revolution, which ended only in 1976 (67:31). People who are seventy years old today lived through the tail of it. The lesson Chinese leadership drew, in Labenz's telling, is simple: falling behind on technology is an existential risk, and dependence on foreign powers for a critical technology is a standing invitation for those powers to dictate terms again.

That history reframes things Americans tend to read as pure aggression, like the chip export ban and the pressure campaign against Huawei. Chinese officials, Labenz says, read Snowden-era revelations about American surveillance of allies, including the German chancellor's phone, and conclude that accusations of Huawei spying are hypocrisy dressed up as security policy (76:22). He does not fully endorse that reading, but he thinks Washington badly underestimates how corrosive it is to trust.

He also thinks Americans misjudge the legitimacy of China's government. Borrowing the old Confucian idea of the Mandate of Heaven, the notion that a government earns its right to rule by delivering for its people, Labenz argues the Chinese Communist Party currently has it in spades, because two generations of rising living standards have made this, for ordinary Chinese citizens, an unusually good time to be alive.

"In China, I think they are living their eternal 1991 right now. Times have never been better in any living memory than they are right now." — Nathan Labenz (39:02)

The reference is to 1991, the fall of the Soviet Union, when Americans of Labenz's generation briefly believed history had ended in their favor. He thinks Beijing is having that moment now, which is one reason he doubts any American appeal to democracy or dissent would land the way policymakers imagine.

The economic evidence backs some of this up. Headline dollar GDP puts the US ahead, at $30 trillion against China's $20 trillion (28:37). But converted into purchasing power parity, a method that measures what money actually buys inside a country rather than what it converts to at the exchange rate, the picture flips: China's economy measures at $45 trillion, about 50 percent larger than the American figure (29:05). Labenz points to cheap dinners, $12 pedicures, and 200-mile-per-hour trains that run on schedule as everyday confirmation that the PPP story is the more honest one (25:57, 30:39).

China's AI buildup follows the same long-arc pattern. Its first five-year plan to mention artificial intelligence was adopted in 2016, the same year Google declared itself an AI-first company (46:28). By Labenz's count, roughly 11 American companies and 12 Chinese companies are currently shipping frontier or near-frontier large language models (47:25), a rough parity that undercuts the idea that China is merely fast-following a purely American invention.

Trading Chips for Trust

None of this makes Labenz an uncritical China booster. He flags real vulnerabilities: an aging population projected to shrink from 1.4 billion to 800 million (106:03), no visible succession plan behind the 73-year-old President Xi, and a military that, like Russia's before Ukraine, has not been tested in decades and may not know its own strength.

His policy proposals, though, are aimed almost entirely at the American side of the ledger. He is skeptical of export controls, not because they fail to slow China, but because the rationale keeps shrinking, from denying military applications, to denying frontier models, to now limiting how many AI agents Chinese firms can run (127:06). He also argues the controls have a chilling effect nobody talks about: American AI safety researchers, worried about violating rules they do not fully understand, are quietly avoiding collaborations with Chinese counterparts who would otherwise welcome them (132:45).

His alternative is a series of trust-building trades. Sell chips in exchange for Chinese expertise in batteries, solar, and robotics, with manufacturing located in the US. Let Huawei back into Western markets once formal verification methods, mathematical techniques for proving what software will and will not do, can certify its equipment does not phone home. Most strikingly, use confidential computing, a technique that lets one party run a program over another party's data without ever seeing the data itself, so each government's AI systems could review the other's sensitive communications and simply certify that no attack is being planned, without either side reading the other's transcripts (177:05).

The unifying worry behind all of this is what Labenz calls legibility. Right now, American and Chinese AI run on roughly the same paradigm, are trained on similar hardware assumptions, and are published, in China's case, in English. That mutual visibility, he argues, is what prevents catastrophic miscalculation. If the chip ban pushes China entirely onto its own silicon and its own closed research culture, both countries lose the ability to know what the other has, and that unknown is far more dangerous than any near-term capability gap.

The essay closes on a case for optimism that Labenz calls hyperstition, the idea that a boldly stated belief about the future can help bring that future into being.

"If you believe it's possible, then it might become possible. But if you don't believe it's possible, then by definition, you've already closed off the possibility." — Nathan Labenz (151:35)

He wants the upcoming Trump-Xi summit treated as a genuine opening rather than a photo op, and he wants both governments to agree, at minimum, never to train general-purpose AI systems to pursue military objectives "by any means necessary," the same instinct that makes commercial AI systems unpredictable when let loose in competitive markets. Whether any of this survives contact with actual diplomacy is an open question. But Labenz's real argument is smaller and harder to dismiss: the two countries are not choosing whether China becomes a technological power. That happened. The only choice left is whether the two powers stay legible to each other, or drift into separate, unreadable futures.