The illusion of an arms race: why the US and China are developing AI differently

Global artificial intelligence investments this year will exceed $700 billion — nearly twice the figure from a year ago. By comparison, the American lunar program cost less. But behind these numbers lies something more important than sheer scale: what politicians and journalists have grown accustomed to calling an "arms race" between the United States and China increasingly resembles an illusion — a compelling narrative with little connection to how things actually work.

The arms race metaphor did not arise by chance. Back in the 2010s, when machine learning surged to prominence, figures like Stephen Hawking and Elon Musk warned of an inevitable merger between AI and military and economic power. The Cold War provided a ready-made template for making sense of technological competition, and the media eagerly adopted it. Major laboratories, venture investors, and analysts are drawn to simple, measurable metrics — model size, benchmark results, computing power. It is convenient, understandable, and marketable. The problem is, it is not true.

"The United States and China are running down completely different tracks," says Selina Xu, who leads research on China and AI policy within Eric Schmidt's team at the former Google CEO's organization. Washington is betting on scaling language models in pursuit of AGI — artificial general intelligence capable of surpassing humans in any cognitive task. This strategy fits organically into the structure of the American economy: services, finance, media, legal services, and technology companies — these are precisely the sectors where powerful generative models deliver immediate and tangible impact.

Beijing, by contrast, is moving along fundamentally different lines: AI is viewed primarily as a tool for productivity gains in the real sector. "Dark factories" — fully automated manufacturing with no workers, roboticized logistics, AI in medical diagnostics and agribusiness — these are China's priorities. Over decades of rapid growth, the country has accumulated a massive industrial sector, and automating it has become the main driver of national strategy.

The difference in goals exposes a fundamental contradiction in the very concept of a "finish line." If AGI is indeed the end goal of this race, then a paradox emerges that Stanford researcher Graham Webster points to: machine intelligence surpassing humans is by definition beyond the control of its creators. "Even if superintelligence emerges in a particular country, there is no guarantee that country will reap the benefits it expects," the researcher warns. In other words, "victory" in this race could turn out to be the most dangerous possible outcome.

But the real threat posed by the "arms race" narrative is not metaphysical but utterly practical. Carson Elmgren of the Institute for AI Policy and Strategy formulates it with precision: an arms race can become a self-fulfilling prophecy. When companies and governments adopt the logic of "win at any cost," safety concerns and ethical constraints begin to be seen as ballast — extra weight slowing forward momentum. Testing protocols are shortened. Oversight mechanisms are circumvented. Risks of systemic failures grow. And this is no longer a theoretical danger: this exact dynamic was observed in the history of nuclear and biological weapons.

Reality is more complex and interesting than any military analogy. The United States and China are not competitors on a single track — they are building different infrastructures for different economic needs, and there is a logic to this. The problem arises when foreign policy and corporate strategy begin to take shape based on a false map. Decisions made based on illusory symmetry of threats lead to real and asymmetrical consequences.

The true challenge of our time is not to outpace a rival but to establish common rules before the stakes become too high. The history of technological revolutions offers plenty of examples when countries that considered themselves adversaries ultimately turned out to be hostages to the same risks. Artificial intelligence in this sense is no exception — it merely accelerates and sharpens an already familiar dilemma between speed and caution, between national interests and global responsibility.