There Is No A.I. "Race": Diffusion, Not First-Mover Advantage, Determines Technological Dominance
Source: Chang Che. "There Is No A.I. “Race” | The New Yorker." September 24, 2026. www.newyorker.com
The Gist
The author argues that the panic over the U.S.-China 'A.I. race' is misplaced because history shows that whoever invents a technology first rarely ends up being the country that benefits most from it—what matters is how well a country integrates the technology into everyday life, like the U.S. did with electricity even though Britain invented key parts of it first. Rather than racing recklessly to build the most powerful A.I. out of fear of China, America should focus on safely and broadly adopting A.I. across its economy, since doomsday scenarios about Chinese A.I. dominance ignore how hard and slow it actually is to turn scientific breakthroughs into real-world catastrophic capabilities.
Conclusion
The prevailing narrative of an existential U.S.-China 'A.I. race,' where being first to develop advanced capabilities determines who 'wins' and losing risks catastrophe, is fundamentally misguided; historical precedent shows that diffusion—the broad integration of technology into economic and social life—not initial invention, determines which nation ultimately benefits from a technology, and America should therefore focus on safe, widespread A.I. adoption rather than racing to the frontier out of fear of China.
Premises
- Historical cases (electricity, steel, chemicals, automobiles) show that the nations or firms that invented key technologies were rarely the ones who ultimately dominated the resulting industries—America pioneered fewer than a third of relevant innovations in these four industries yet became the world's leading economy.
- Britain invented pivotal electrical technology (Parsons' turbine) but fell far behind the U.S. in electricity adoption within decades, because diffusion into infrastructure and society mattered more than initial invention.
- Catastrophic scenarios often invoked to justify racing ahead (drone armies, ethnic bioweapons, knockout cyberattacks) require extensive real-world diffusion and tacit knowledge that historically takes far longer and is far harder to achieve than simply possessing the underlying technology or recipe, as shown by the Soviet anthrax program and the nine-plus years needed to integrate Project Maven into military use.
- Prioritizing safety and public trust (‘pacing the frontier’) is historically compatible with, and even necessary for, successful long-term diffusion, as shown by how the Three Mile Island disaster stalled nuclear adoption in the U.S. for decades.
- China's own A.I. policy is focused primarily on diffusion (e.g., the 'AI+' initiative) rather than racing toward speculative superintelligence, undermining the premise that China is single-mindedly pursuing a decisive strategic advantage.
- The 'But China' rhetoric is also strategically used by U.S. tech companies to justify deregulation and unchecked growth (e.g., Zuckerberg, OpenAI), suggesting the race narrative often serves self-interested purposes rather than reflecting genuine strategic necessity.
- The distinction between a Chinese A.I.-enabled disruptive attack (an act of war) and mere continued espionage (the current status quo) is being blurred by race rhetoric, when in fact escalation to actual catastrophe would require far more than a lead in A.I. models—it would require a broader breakdown in diplomacy.
Assumptions
- Past patterns of technological diffusion (steam, electricity, nuclear, industrial technologies) are sufficiently analogous to A.I. to predict its trajectory.
- Institutional and bureaucratic constraints on diffusion apply similarly to authoritarian states like China as they do to the U.S. and other democracies.
- Public trust and safety-oriented rollout will not, in this instance, be strategically exploited or ignored by less safety-conscious actors gaining disproportionate advantage.
- The speed of A.I. diffusion will follow a similarly gradual timeline as historical technologies, despite A.I.'s potentially faster iteration and self-improving capabilities being categorically different from steam engines or nuclear reactors.
- Diplomatic breakdown severe enough to enable use of catastrophic A.I. capabilities is unlikely in the near term, or at least decoupled from the AI capability race itself.