�鶹������ý

Most of the debate about the United States versus China AI competition fixates on capabilities and milestones: who has more compute, better models, faster deployment. 

The more consequential divergence runs deeper. American AI policy follows the logic of the frontier: restless, expansive, compelled to press forward. Chinese AI policy follows the logic of rejuvenation: controlled, consolidating, oriented toward closing vulnerabilities. These are not just competing strategies, but competing national mythologies. And when they collide in high-stakes crisis situations, it’s likely that neither side’s AI will behave the way its designers presumed.A growing body of recent analysis this divergence: while Washington frames competition as a sprint for frontier model performance (the biggest systems, the most compute, the closest path to AGI), China AI as infrastructure, deploying it at scale across manufacturing, logistics, and healthcare. It is building toward semiconductor self-sufficiency, its models globally dominant through open-source accessibility, and AI into physical systems in ways that play to its manufacturing and engineering strengths. 

The American Frontier

The vocabulary we use to talk about AI today is revealing, too. 

Consider the word ““; it does a lot of work for Americana. Rooted in the logic of Manifest Destiny—the 19th-century belief that American expansion across the continent was not just desirable but inevitable—the word “frontier” treats limits as temporary, and stagnation as failure of will. Limits are not accepted, but overcome. Falling behind is not merely a setback, but a betrayal of national character.

This, of course, is a mythology, which means it may be as image-generating as reflective. And yet, it endures. Rather than dying with the closing of the continental frontier, it , into Silicon Valley’s founding stories, and into the AI boom. When American technologists and policymakers speak of frontier AI, they are not simply describing a technology. They are invoking a cultural script: that the leading edge is where Americans belong, that pressing forward is both inevitable and necessary, and that the appropriate response to any limit is to find a way past it.

In practice, this produces specific institutional behaviors. The —titled, without apparent irony, “Winning the Race”—clarifies the logic: AI companies must be “free to innovate without cumbersome regulation.” The official doctrine seems to be: deploy first, refine later. It frames technological leadership as a prerequisite for global strategic leadership, equating falling behind with a sign of civilizational regress, rather than a simple capability gap.

Most critically, it accelerates cognitive offloading to automated systems. Project Maven, now carrying a $1.3 billion contract through 2029, was explicitly designed so operators “wouldn’t have to stare at a screen.” This is the governance problem the “frontier” frame cannot solve from within itself: AI makes institutions faster without making them wiser, and in high-stakes domains, that gap between velocity and judgment is precisely where accountability disappears.

China’s Counter-Narrative

China has a distinct relationship to AI, one of rather than expansion. The”—the period from the mid-19th to mid-20th century during which China suffered military defeat, colonial exploitation, and internal collapse—remains a live reference point in Chinese political culture. This politico-cultural backdrop frames AI not as a frontier to conquer but what was lost: status, security, rightful place in the international order. China’s 2017 New Generation AI Development Plan that AI development supports “the great rejuvenation of the Chinese nation.”

Unlike the U.S. accelerationist approach, Beijing’s calculus around AI emphasizes closing strategic vulnerabilities before they can be exploited. The result is centralized oversight, subordination to party-state priorities, and the formal maintenance of humans in the chain of command. 

Interestingly, the party-state insists on human authority in principle while systematically building the infrastructure that makes it optional in practice. The demands of national rejuvenation undermine the human authority point as well. China’s 2023 Generative AI Interim Measures AI systems to be “secure and controllable,” with human oversight written into law. The formal architecture is detailed, and in some respects more elaborate than its American counterpart. But it governs civilian and commercial systems. The infrastructure operating at the scale of the rejuvenation frame actually demands—an estimated 600 million AI-enabled surveillance cameras, a national credit platform holding over 80 billion records, PLA decision-support systems explicitly designed to accelerate military decision-making—exists in a governance space these regulations don’t reach. Rejuvenation at scale requires automation at scale, and automation at scale is precisely what makes meaningful human oversight increasingly nearly impossible to sustain.

When Mythologies Collide

The interaction of these two mythologies may be more dangerous than either alone, because they narrow the interpretive range each side applies to the other’s moves. This is distinct from a standard great power competition logic, which  can explain why two states accelerate in response to each other. It cannot explain why de-escalatory signals fail to register. That failure is where mythology does its work. A Chinese deployment move that a neutral observer reads as infrastructure catch-up looks like an existential threat through the American frontier frame. An American model release that looks like commercial competition, reads like deliberate window-closing through the Chinese rejuvenation frame (buying time against permanent disadvantage). There’s no move either side can make that the other is built to read as genuine restraint. The spiral doesn’t just sustain itself: it forecloses the exits.

Both sides have deployed AI to watch the other’s AI. U.S. targeting and decision-support AI reads Chinese early-warning activity as preparation for first strike; Chinese early-warning AI reads U.S. targeting activity as an imminent attack. Routine operation on both sides produces threat signatures on the other—indistinguishable, to the systems involved, from actual attack preparation. The misperception isn’t the result of bad intelligence or bad faith. It is baked into the architecture.

The standard reassurance in the face of these concerns is that humans remain in the loop. Both sides formally preserve human authority over their most consequential AI systems. The U.S. requires senior review before deploying autonomous weapons—but includes a waiver provision allowing that requirement to be suspended in cases of “urgent military need.” China maintains human authority in doctrine, in law, and in its international governance positions—but at the scale its systems actually operate, meaningful oversight was always an abstraction. No human reviews a surveillance network of 600 million cameras or a credit platform holding 80 billion records. Under normal conditions, these safeguards function. Under crisis conditions, when they matter most, both are likely to fail simultaneously—one by urgency, one by scale.

There is a scenario that exploits all of this at once. Recent PLA doctrine suggests that future conflict may be defined less by autonomous systems acting on clean information than by rival efforts to corrupt and destabilize each other’s AI—targeting data, algorithms, and computing power simultaneously. PLA forces already train to deceive AI-assisted targeting and to override faulty machine recommendations under adversarial pressure. The deception scenario activates the interpretive filters, manufactures the threat signatures, and generates the urgency that suspends oversight—all at once. You don’t need to fire a shot. You just need to make the other side’s systems see one coming.

Why Naming the Mythology Matters

If the problem were purely strategic, existing tools would be adequate. Export controls can constrain capabilities. Testing frameworks can evaluate performance. Safety protocols can govern deployment. These instruments matter, but they operate at the capability layer, not the narrative layer, and it is the narrative layer that determines how each side interprets the other’s moves, calibrates its own risk tolerance, and decides when a technological development crosses a threshold demanding a response. You can negotiate capabilities. You cannot easily negotiate mythology.

Conventional analysis tends to reach for historical analogy to fill this gap. But there is an irony in that instinct: the frontier myth and the century of humiliation are themselves forms of historical analogy—stories about the past that shape decisions in the present. Reaching for more historical analogies to analyze mythologies built on historical analogies is unlikely to break the cycle. The escalation pathways generated by the interaction of these two logics are not legible to any single actor; they emerge from the system, under pressure, in ways no one has designed and no one fully perceives.

This is precisely where computational modeling, simulation, and analytically structured wargaming become essential—not as academic exercises, but as policy instruments. Simulation allows the narrative dynamics themselves to be encoded as decision rules: U.S. speed and experimentation versus Chinese vigilance and control can be modeled as competing logics, stress-tested under crisis pressure, and examined for the failure modes they generate together. Where does cognitive offloading tip into catastrophic deference? Where do comparable AI-driven threat signatures produce simultaneous escalatory responses? Where are the off-ramps, and how much time exists to reach them? These are not questions that intuition or historical analogy can answer reliably. They require tools capable of running the interaction forward—before it runs itself.

The U.S. is not going to stop striving for the “frontier.” But strategists and policymakers who understand the mythology embedded in that word—and who can map its Chinese equivalent—will be better equipped to anticipate where misperception is most likely to emerge, where decision timelines are most likely to compress, and where off-ramps need to be deliberately constructed before a crisis makes them impossible to reach. Naming the mythology is where the work begins. Simulation is how you find out what it costs.