When AI gets more powerful, do countries always end up racing each other — or can other forces drive what happens next?

This explores whether each jump in AI capability turns into a contest between nations, or whether other forces do as much to shape what happens next. The collection has only one note that looks squarely at nation-state rivalry. That note shows a strong pull toward competition. Several other notes suggest the national race is only one of the contests going on, and possibly not the most important one.

The clearest evidence of the geopolitical pull comes from Amodei's proposal to slow AI development down in a coordinated way. According to Can AI safety pacing work without government cooperation?, both Trump and Xi rejected it within days. Romero argues that the obstacle was not technical. Once governments see AI as a source of national power, safety arguments lose to strategic ones. The Future of Life Institute reaches a related conclusion from the other direction. In Can companies alone manage the risks of AI systems?, it argues that companies cannot police themselves and calls for binding government limits, with hardware verification to check that the limits are kept. That puts governments at the centre of AI safety, and governments are the actors most likely to treat AI as a strategic race. Put these two notes together and you get a trap: the only bodies strong enough to enforce restraint are the same ones with the strongest reasons to compete.

The competition is not only between countries, though. In Will self-sovereign AI agents inevitably emerge despite policy efforts?, Ball argues that AI agents acting with growing independence will appear because of economic incentives, whatever any government decides. On his account, banning them would push the legitimate ones toward crime rather than stop them. If he is right, AI development has a momentum that state rivalry neither causes nor controls. What bottlenecks define the path from AGI to superintelligence? makes a similar point. It describes four separate routes from human-level AI to superintelligence, each with its own bottlenecks. That makes a single 'race to the finish' an oversimplification.

Here is the part you might not have expected: the race may be over the wrong thing. Is the AI capability gap really an interface problem? argues that the next big jumps in usefulness will come from better interfaces, not better models. Do automated benchmarks hide what frontier AI systems can really do? shows that standard benchmarks both overstate and understate what models can really do. A race scored on model leaderboards could be measuring the wrong thing. The risk picture is also unexpected. In Where do frontier AI models actually pose the greatest risk today?, recent frontier models crossed warning thresholds for persuasion and manipulation, but not for cyberattacks, self-replication or doing AI research on their own. So the capabilities most relevant to geopolitics today look more like tools for influence campaigns than like the superweapons people usually imagine.

Finally, Does incremental AI replacement erode human influence over society? points to a danger the race frame misses completely. Institutions stay in line with what people want partly because they depend on human workers who care about the results. As AI replaces those workers, that check weakens in every country, whoever is 'winning'. So the collection's answer is not that competition is inevitable. Capability does get pulled into geopolitics quickly when states are paying attention. But the economic, institutional and social changes may matter just as much, and they do not wait for any race to be decided.