Countries need to stop viewing AI as a race in which one side must beat the other.

There is no shortage of this type of zero-sum game or competitive framework when it comes to the global AI space. Indeed, even if Anthropic has underlines AI Security at Home, the company’s co-founder and CEO Dario Amodei, has fueled the narrative of an overseas arms race, arguing that export controls are key to slowing China’s development and making the United States win the AI ​​race. In the same way, The chip was author Chris Miller argues that US controls on chip exports, such as prohibition on the sale to China of the most advanced GPUs such as the NVIDIA H100, have “succeeded… [by] considerably slow[ing] growth in China’s chip manufacturing capacity.” Indeed, Trump himself declared America started the AI ​​race in July, and it will win.

Such arguments suggest that the two great powers are engaged in a two-way race – one will win and the other will lose – and that the winner will gain significant advantages at the expense of the loser. Yet from a rational choice perspective, the “AI race” is a misnomer. A two-person race typically involves an environment characterized by a rival resource (which both parties cannot benefit from) that is non-exclusive (neither player can easily prevent the other from using it), and players compete to see who will be the first to gain access to that resource.

In the 1955 film, Rebel without a causeJim Stark (James Dean) runs towards a cliff against his nemesis Buzz (Corey Allen). If both teenagers drive straight, they both die. Whoever swerves first loses. If one driver swerves and the other continues to run to the edge of the cliff, neither can improve their position by changing strategy: we call this a Nash equilibrium. This result is not cooperative: if one swerves, the other must run; but if one runs, the other should swerve.

The geopolitical AI ecosystem is not like that. The use of AI models is excluded: in fact, last year, Sam Altman decided to exclude Chinese users of OpenAI’s GPT, but this use is not strictly rivalry (DeepSeek’s models are released under open source licenses and can be run locally by anyone). Implementations of a model are arguably rival, to the extent that the marginal user imposes an energy/data cost, but that was not the concern that drove Altman’s decision: he excluded Chinese users because he thought the United States should not cooperate with China.

The argument may be that selling chips to China would embolden Beijing and make the United States worse off. Yet this doesn’t take into account the benefits enjoyed by ordinary, middle-class American households through greater access to cutting-edge electronics products at lower prices, or the amount of leverage offered by global reliance on the American technology landscape.

Some economists refer to a situation characterized by non-rival but excludable resources, instead of rival but non-excludable resources, as “deer hunting”, drawing inspiration from a parable by the philosopher Jean-Jacques Rousseau. A speech on inequalities. Consider a group of hunters who can choose to hunt a large prey (deer) together or a small prey (rabbit) alone. The thing is, they can only catch the deer if they cooperate, while each can hunt a rabbit alone. This game has two Nash equilibria: either we work together to hunt the deer, or we each work alone to catch a single rabbit. Yet one of these balances is better than the other: we should work together to hunt deer.

The global AI competition is more like a deer hunt than a race. Whether in politics, governance or trade, cooperation between countries can bring greater benefits than working alone. On the other hand, a breakdown in communication breeds distrust, which could give rise to harmful mistakes, such as a spiral of escalation resulting from an overestimation of the threat posed by the other side, or reckless deployment of AI in conflicts. The “stag” in the US-China AI game therefore lies partly in the mutual prevention of such errors and the gains resulting from mutually beneficial commercial development of AI for the benefit of the general public.

There are many common challenges facing China, the United States and the world, from AI manipulation, deception and coercion to labor displacement caused by the implementation of AI in the labor market. Such mutually beneficial cooperation requires trust, transparency and cooperation, as opposed to erratic politicization: this is how we move from rabbit hunting to deer hunting.

To achieve this, policymakers should seek to develop effective multilateral AI governance institutions, including establishing and monitoring dispute resolution mechanisms. Trading capital also arises from unconventional alignments of mid-sized powers, each with their distinct niches.

For example, Energy-rich Saudi Arabia is striving to become the third largest AI market in the world, while key players in France and Israel commit to becoming leaders in specialized AI applications. With its huge population and growing emphasis on education, India is poised to become a leading provider of engineering and IT talent.

The international order is becoming increasingly multipolar, and the world of AI is no exception. Instead of trying to “win the AI ​​race” at all costs against its rival, the United States and China should build bridges and seek common ground with friends and rivals.

This essay is adapted from the authors’ upcoming book, Geopolitics of artificial intelligencewhich will be published in 2026 by Cambridge University Press as part of its Elements series.

The opinions expressed in comments on Fortune.com are solely the opinions of the authors and do not necessarily reflect the opinions and beliefs of Fortune.