"Whoever wins AI wins." This formula, repeated this week at the highest level of American politics, is not an argument: it is a metaphor. And like any established metaphor, it decides what is thinkable before we have even begun to think. It assumes three things, and none of them is obvious.
On Thursday we reported on the public explosion of the debate over pace. Here is the image that structures that debate on both sides.
Assumption 1: there is a finish line
A race assumes a point at which it ends. Yet no one can say what "winning" at AI would mean.
The first model to cross a threshold? We showed in our article on AGI that no shared definition exists. The largest market share? It changes every quarter. Technical superiority? We put a figure on a three-point gap between the best American and Chinese models, and that gap fluctuates.
A competition with no finish line is not a race, it is a permanent state. And you do not behave the same way in a sprint as in a situation that lasts.
Assumption 2: a lead is retained
This is the weakest point, and it is easy to verify.
In a race, whoever is ahead keeps their lead as long as they run just as fast. In AI, a lead leaks. We documented this in our article on model extraction: a system you can query is a system from which part of the know-how can be extracted.
Add to that deliberate diffusion. A public laboratory released for free an agent that beats a cutting-edge commercial model in its speciality. In a race, you do not hand out your shoes to your competitors.
What is retained, then, is not the technical lead but the infrastructure: the chips, the electricity, the data centres. Those are assets, not a position in a ranking.
Assumption 3: the winner takes all
The formula suggests that the runner-up gets nothing. That is false in just about every comparable case.
Electricity, computing, the motor industry all produced leaders and entire ecosystems around them. No one "won" electricity. The countries that benefited most from it are not necessarily those that invented it, but those that deployed it best.
This is the question we raised in our article on whether you have to be the best to count: carrying weight and not being dependent are different objectives from dominating, and they are achievable.
This is the most important effect, and it is political. In a race, slowing down is always a mistake. The metaphor disqualifies by construction any cautious position, regardless of its arguments. It also makes cooperation unthinkable: you do not coordinate with a competitor. Yet a good share of the risks in question — offensive capabilities, incidents, testing standards — are exactly the kind of problem that can only be solved collectively. The metaphor does not rule out an argument: it rules out an entire category of solutions.
What to replace it with
No image is neutral, and the point is not to impose a better one. Two others are in circulation, and they shed a different light.
Building infrastructure. You do not win an electricity grid, you build it, you maintain it, you decide who has access to it. This image puts duration, maintenance and distribution at the centre — exactly what we document week after week.
Industrial safety. In aviation or nuclear power, competition exists and it coexists with common standards, cross-inspections and shared reporting, because an accident at one company costs everyone. This morning we reported on the embedding of independent evaluators within the labs: a move that makes no sense in a race and a great deal in an industry.
What to take away
The race is not a bad metaphor because it would be false: there is indeed a competition, it is intense, and the stakes are real.
It is bad because it is too poor. It reduces a situation to a single dimension — ahead or behind — whereas what we describe every day involves at least four: technical capability, physical infrastructure, the legal framework and public trust.
And when a metaphor becomes familiar enough, we stop seeing it as an image. We think we are describing the world when we are repeating a figure of speech. It is probably time to look at it.