Image via The New York Times
From a distance, the contest for supremacy in artificial intelligence appears to be a simple two-horse race. In one lane, America, with its mighty private-sector labs and foundational inventions. In the other, China, pouring state resources into a strategic priority. The betting markets would seem to favour the incumbent. But the reality is not so simple. China may not be poised to “win” in the conventional sense. It is, however, playing a different game, where breadth of adoption could eventually matter more than the brilliance of the initial breakthrough.
The data still point to American dominance, though it is starting to strain. American institutions produced 40 of the world’s most notable AI models in 2024, compared with 15 from China. The United States still employs a disproportionate share of the globe’s top AI researchers, however its share has slipped from 59% in 2019 to 42% in 2022. China’s portion grew from 11% to 28% in the same period. In the esoteric but telling metric of citations in top-tier research, America leads but its share is declining. China, by contrast, now accounts for nearly 70% of the world’s AI patents. It is clear where things are heading.
America’s advantage rests on deep foundations: a vibrant venture-capital ecosystem, world-leading semiconductor expertise and a culture that celebrates disruptive experimentation. Yet these are being matched by Chinese strengths of a different order. Where America innovates, China iterates and implements. Chinese researchers, constrained by American export controls on advanced computer chips, have become masters of efficiency. The training of DeepSeek-V3, a powerful open-source model, required just 2.6 million GPU-hours, a fraction of what its American counterparts consume. This forced frugality has bred algorithmic sophistication. Chinese models are learning to do more with less.
This efficiency is accelerating deployment. China has overtaken America in the monthly downloads of AI models. Its existing dominance in digitally native sectors such as fintech, e-commerce and logistics provides a ready-made landscape for applying AI. Local governments and large enterprises are already integrating reasoning models into public administration and supply chains. This rapid, large-scale adoption creates a virtuous cycle: more use generates more data, which in turn leads to faster improvements. The focus is shifting from raw power to practical integration.
A societal divergence reinforces this technical split. Stanford University’s 2025 AI Index found the Chinese public to be the most optimistic in the world about the technology’s potential. This enthusiasm is institutionalised from the top down. The Ministry of Education is integrating AI literacy into school curricula for all ages, ensuring a future workforce fluent in its language. This stands in brutal contrast to the more ambivalent public discourse common in the West. Decades of top-down coordination have created a system unusually effective at pushing large-scale technological adoption with minimal social resistance.
The two nations’ model philosophies are also diverging. America’s leading companies largely favour proprietary, closed models. China has embraced open-weight models, making their code available for others to use and build upon. Alibaba’s Qwen models are now among the most downloaded open-weights globally. Even Sam Altman, the chief executive of OpenAI, has conceded that his firm may have been “on the wrong side of history” on open-source strategy. This is more than a technical debate: it is a contest between two visions for how a transformative technology should proliferate.
Inevitably, there are constraints. China’s greatest weakness remains its dependence on foreign semiconductors. Export restrictions have throttled access to the most powerful chips, creating a grey market for banned components and forcing labs to recycle hardware. Although domestic chip production is expanding, it cannot yet match the cutting edge. Furthermore, the same engineering-state mentality that efficiently builds infrastructure can be brutally applied to the social sphere, potentially stifling the creative dissent that fuels foundational innovation.
The race, therefore, is not a single sprint but a series of parallel marathons. America continues to set the pace for pure, frontier research. China is forging ahead in diffusion and application. The ultimate advantage in a general-purpose technology may lie not with the nation that invents it first, but with the one that harnesses it most thoroughly. ■