The Un-Glamorous Revolution
Why true sovereignty is built in data centres, not treaties
A colleague of mine was out on his bike this spring when he stopped at the end of a driveway to compliment a stranger on a restored car. Five minutes of small talk later he had learned that the man builds hyperscale data centres in Canada for one of the American cloud companies, and that the two of them shared half a phone book of contacts. So my colleague asked the obvious question. Why aren't you building more? The answer came without a pause. We can build as many as you want. We don't have the power.
Not permits. Not capital. Not talent. Electricity.
I told that story in Rome more than once, because it is the shortest way out of the conversation those rooms wanted to have. Declarations are what summits produce, and I have sat on the drafting side of them. Sovereignty is produced somewhere else, on a different clock.
The clock mismatch is not a metaphor. It is arithmetic. My working rule is that the technology turns over about every seven months. From the day a government says "let us do AI" to the day a contract is signed is about nine. The procurement finishes after the thing it procured is obsolete. Beneath that sits a slower clock still: an electoral cycle of four or five years, asked to make twenty-year decisions about a seven-month technology.
So sovereignty gets decided in the least glamorous rooms available. This autumn, the most consequential sovereignty debate in my own province is not a summit. It is a rate case. Since 2023, any new load of five megawatts or more in Quebec has needed a minister's authorization. Hydro-Québec has asked the energy regulator to roughly double what large data centres pay, to about thirteen cents a kilowatt-hour. In late October a Superior Court judge struck down the two government decrees that backed the proposal. Google is among the companies on the other side, and the hearings resume in December. Nobody will write a communiqué about any of it.
Now the scale. A frontier training campus in the United States is measured in gigawatts. The gigawatt-scale campus Canada talks about most is in Alberta, planned off the grid on natural gas. In Quebec, which is almost entirely hydro and has some of the cleanest power on the continent, every data centre in the province together draws about two hundred megawatts at peak. China added more than five hundred gigawatts of generating capacity last year, more than three times Canada's entire fleet.
Fifteen years ago Montreal was one of the handful of places where modern AI was invented. The ideas stayed. The compute went to Seattle. We invested in the idea and not in the value chain attached to it, and we are making the same decision again. On readiness, on a scale of one to ten, I would put us at one.
Now the objection, and it deserves better than impatience. The people saying no are not stupid. A government that protects cheap power for households is doing something defensible; I have said so myself. Communities that do not want the substation have a point. A utility whose mandate was written for another century is following its mandate. Even the rate case is a serious argument about who should pay for scarcity. These are the actual politics of sovereignty, and they are more consequential than any communiqué signed this year.
But they have to be had honestly, which means saying two uncomfortable things in public. The first is that the bill may have to go up, the way we pay for roads, hospitals and schools, because firm power for a sovereign capability is a public good. The second is that a public utility should be recruiting this load on purpose, choosing the workloads that matter to the country, rather than rationing megawatts and calling the queue a strategy.
One of the European Parliament's rapporteurs for the AI Act said it on a panel in Rome better than I can: "regulation is the beginning of the work. The next challenge is implementation, investment and capacity." The encyclical is blunter: "Nothing in the world of AI is immaterial or magical. Every seemingly immediate and flawless response is the result of a long chain of mediation, involving vast networks of natural resources, energy infrastructure and, above all, people."
Every country that announces a sovereign AI strategy without a power strategy has announced an intention, not a capability. The gap between the two is measured in gigawatts.
Previously: The Automation Trap. Next Tuesday: Democracy's New Crisis.
Follow the season
One episode every Tuesday until 15 December. Get each one the morning it publishes.