Market · North America
Canada. Cheap firm power, if you are allowed to have it.
Hydro at industrial rates, a cold climate, US latency and a government actively funding domestic compute. The catch is that the best power in the country is allocated rather than sold, and the process rewards whoever started earliest.
Where the capacity actually is
- —Québec
The strongest power story in North America — abundant firm hydro at industrial rates, with a climate that supports free cooling for a large part of the year and makes heat reuse genuinely viable. Large loads are assessed and allocated rather than simply connected, so the sequencing of the request matters as much as the site. Montréal also carries excellent latency to the US northeast.
- —Alberta
The fastest route to large-load energisation in Canada, and often in North America. A deregulated market, a system operator that has engaged constructively with data center load, and abundant gas make behind-the-meter and co-located generation a standard structure rather than an exotic one. The trade is merchant price exposure, which is manageable and hedgeable but must be priced deliberately.
- —Ontario
The deepest enterprise market and the best connectivity, with Toronto serving both Canadian users and the US midwest and northeast at low latency. Power is available but the process is slower than Alberta and the pricing is higher than Québec, which puts Ontario in the middle on economics and at the top on convenience.
- —British Columbia
Clean hydro and a favourable climate, with real transmission constraint around the Lower Mainland. Best suited to moderate-scale deployments that need Pacific latency rather than to hyperscale training campuses.
- —Manitoba and the Prairies
Underused hydro and low power cost, with limited existing data center infrastructure. Genuinely interesting for greenfield at scale, and almost always absent from a buyer’s initial shortlist.
What is different about transacting here
Canada is not simply a cheaper version of the US market. The differences are structural, and each one changes a term in the contract.
- Power is allocated, not merely purchased. In the provinces with the best economics, a large load is a decision made by the utility and, implicitly, by policy. That means the case you present — jobs, heat reuse, grid flexibility, domestic capability — is part of the commercial negotiation, whether or not anyone calls it that.
- Currency. Revenue in USD against costs in CAD is a real economic position, and whoever carries it should be carrying it on purpose. This belongs in the contract explicitly rather than as an assumption in someone’s model.
- Data residency is provincial as well as federal. Public sector and health workloads in particular carry obligations that vary by province, and “in Canada” is not always sufficient.
- Heat reuse is a live revenue line, not a slide. The climate and the district heating context in parts of Québec and Ontario make recovered heat something you can actually contract around, which changes the economics of high-density liquid-cooled deployments.
- Cross-border equipment movement. Duty treatment, tax and the logistics of getting GPUs to a Canadian site are all solvable and all routinely underestimated in schedules.
Where the requirement is large enough to need new build, the development mandate covers the delivery terms; where power itself is the gate, start with power procurement.
What we source here
Asked in this market.
01Why does Québec come up first for Canada?
Because it combines some of the lowest industrial power rates in North America with firm hydro generation and a climate that reduces cooling load for much of the year. The constraint is not price, it is allocation — large loads are assessed and granted rather than simply connected, and that process has to be run properly and early.
02Is Alberta really faster than the US for interconnection?
For large loads it is frequently the fastest path in North America. The market is deregulated, the system operator has been comparatively constructive with data center load, and behind-the-meter generation is a normal structure rather than an exception. You take on merchant power price exposure in exchange.
03Does Canadian data residency actually matter for AI workloads?
It matters enormously for public sector, health and financial workloads, and increasingly for anything touching Canadian personal data. It matters far less for foundation model training on non-personal corpora. Establishing which of those you are is the first question, not the last.
04What about latency back to US users?
Toronto and Montréal are within single-digit to low-double-digit milliseconds of the major northeastern US metros, which is well inside tolerance for most inference and all training. Vancouver serves the Pacific Northwest similarly.
05Is there sovereign compute funding worth aligning with?
There is an active federal agenda around domestic AI compute capacity, and it changes who your counterparty can be and what a build can be structured against. Whether it is relevant depends on your entity and your workload, and it is worth establishing before you shortlist operators.
06Can you source GPU clusters in Canada, not just space?
Yes. Bare metal cluster builds and GPUaaS agreements are both sourceable in Canadian sites, and the platform availability is broadly the same as the US — the differences are in power cost, term and who carries the currency risk.
Power capacity procurement
In Canada the power conversation precedes the site conversation. This is usually where a mandate starts.
Open the mandate →CapacityData center capacity sourcing
Colocation and wholesale, sourced across provinces and benchmarked against US comparables.
Open the mandate →MarketUnited States
The cross-border comparison, run properly — including the markets that beat Canada on date and lose on cost.
Open the market →Cheap power is only cheap if you can get it allocated.
Tell us the load and the date. We will tell you which province can grant it, what the utility will want in return, and where the currency risk should sit.