Where Canada is building AI data centres, who is paying for them, and the power and sovereignty questions that follow.
AI data centres in Canada have moved from a niche infrastructure story to a provincial economic strategy in under two years. Alberta and Saskatchewan are building. Ontario and Quebec already host most of the existing capacity. The money is largely American. This page tracks what has been announced, what it means for Canadian businesses buying AI services, and where the arguments are.
What has been announced in Canada
| Project | Where | Scale |
|---|---|---|
| Microsoft AI infrastructure commitment | Canada wide | $19 billion |
| Olds and Bonnyville data centres | Alberta | $1.7 billion |
| Moose Jaw AI health town | Saskatchewan | $485 million, 150 MW |
| Federal sovereign compute and quantum | Federal | $1 billion |
| Nvidia acquisition of CentML | Toronto | More than $400 million |
Alberta’s own build is heading toward a distributed model rather than a handful of giant sites. We covered what that looks like through 2028 and why the shape of the build matters as much as the total.
Why AI data centres matter to a business that will never own one
Three reasons, and none of them are about real estate.
Latency and region availability. Whether a vendor offers a Canadian region decides how fast your tools respond and, more importantly, whether your data stays in the country. That question now reaches the board, which is the argument in why data sovereignty became a board level AI issue.
Power prices. Data centres compete with everyone else for electricity on the same grid. In Alberta that competition is explicit provincial policy.
Bargaining position. A country that hosts compute has leverage in a way a country that only rents it does not. That is the thread running through Alberta’s push on AI sovereignty.
What to ask a vendor about Canadian hosting
- Which region does our data sit in, and is it named in the contract or only in a support article.
- Does inference run in the same region as storage, or does the prompt leave the country even when the record does not.
- What happens to residency if you move us to a newer model.
- Who else can compel access to that data under the laws of the hosting country.
The arguments worth following
The economic case for hosting compute is not settled. Data centres employ few people once built, they consume power that ratepayers also want, and the tax base they create can be modest relative to the incentives offered. The counter-argument is that compute is the input to everything downstream, and a province without it exports the value.
Alberta has made its choice. We laid out the case in The Alberta AI Advantage must be built with Albertans and the timing argument in the next 36 months.
FAQ
Where are Canada’s AI data centres being built?
Announced projects concentrate in Alberta, at Olds and Bonnyville, and in Saskatchewan at Moose Jaw, alongside existing capacity in Ontario and Quebec and a Canada wide Microsoft commitment.
How much is being invested in AI data centres in Canada?
Publicly announced figures include $19 billion from Microsoft, $1.7 billion in Alberta and $485 million in Saskatchewan, plus $1 billion federally for sovereign compute and quantum.
Does a Canadian data centre mean my data stays in Canada?
Not automatically. Storage region and inference region are separate settings with most vendors, and only the contract settles it.
Do AI data centres create many jobs?
Construction employment is significant and ongoing operations employment is modest. Judge any announcement by the operating headcount, not the build.
Keep reading
- Canada fast-tracks AI patents to keep the IP at home
- Nvidia hit $5 trillion and nobody agrees what it means
- More AI in business coverage
Maintained by the AI Magazine Canada editorial team. Figures are as announced and link to our reporting on each project.