Canada’s AI Build Needs a Power Deal Canadians Can See

A new data centre can be a valuable electricity customer. It can also leave a utility planning around demand that arrives late or never materializes. The agreement signed before the servers switch on determines how much of that uncertainty reaches everyone else’s bill.

The Saskatchewan Rate Review Panel’s August 28 report raised a specific concern about Bell’s planned data centre near Regina: SaskPower’s improved financial forecast relied partly on additional electricity sales, but the timing of those revenues remained uncertain. The panel wanted updated information before recommending whether the proposed February 2027 rate increase should be confirmed or revised.

In the opening article of Canada’s AI: From Power to Payoff, we examined the wider system behind Canadian AI. This electricity chapter follows the agreement between a very large new customer and the system serving it, from the connection bill to the risk of delayed demand.

A connection has several bills attached

A project needs enough generating capacity to serve it, transmission that can deliver that power and equipment connecting the site. It also needs an agreement about how much electricity it may draw, when service begins and when the utility can ask it to reduce consumption. A price per kilowatt-hour answers only part of that question.

Capacity and energy are different quantities. Megawatts describe the rate at which a facility draws power; kilowatt-hours measure energy used over time. For scale, 50 MW running continuously through a non-leap year amounts to 438 million kWh. Change the energy price by one cent per kWh and that assumed annual bill moves by $4.38 million. Actual consumption, demand charges, taxes and other adjustments would still have to be accounted for.

Canada’s Responsible Data Centre Development Principles put project-driven costs on proponents, including directly attributable generation, transmission, substations and grid upgrades, proportionate to the project’s scale and impact. That reaches beyond the wire at the property line.

The federal framework complements provincial, municipal and Indigenous processes. Signing on establishes a public commitment; it does not let a reader inspect the provisions of a particular electricity contract. Those provisions are where payment obligations, delay protection and consequences for non-performance need to be tested.

Saskatchewan has two power arrangements to explain

The province’s September 14 expansion announcement retained 300 MW from the provincial grid and proposed up to 900 MW of additional, self-supplied capacity. Together they provide a path to a 1.2 GW hub. They should remain separate entries in any accounting of supply, cost and approvals.

Saskatchewan’s August framework requires new proponents to supply their own power. It also identifies Canadian ownership and headquarters, local jobs and partnerships among its assessment principles. That changes the setting from the one in which the original grid-served Bell project was announced.

For that original project, the panel’s technical consultant records a combination of firm and interruptible service, with existing rate code E24 intended for the firm portion. It also records SaskPower’s statement that Bell will pay all associated connection infrastructure costs, including two new 230-kilovolt services. Interruptible service permits consumption to be curtailed under its terms; it should not be described as identical to an uninterrupted supply commitment.

That payment assurance deserves weight. The reviewed records do not establish that households are paying for Bell’s hook-up. Protection against a slower ramp-up, a departing tenant or unused reserved capacity is a separate financial question.

The panel treats the existing service arrangement as an interim approach and recommends a policy ensuring data-centre customers bear the full costs and risks of their loads. Its concern is broader than construction spending. Anticipated sales can help a utility; depending too confidently on their arrival creates a different exposure.

For the expansion, Bell says natural-gas generation at a separate Saskatchewan site would be funded by Bell and a third-party operator partner. It says the expansion would not draw load from the provincial grid, and construction will wait for approvals and a tenant agreement. Those are the company’s stated plans, not an operating power station or an independently verified finding of zero ratepayer impact.

Self-generation may relieve pressure on the electricity grid. It also makes fuel supply, generation financing, environmental approvals and any backup arrangements part of the project’s viability. Private generation still needs a credible account of fuel use and emissions. The public materials reviewed for this article do not establish the complete allocation of those risks.

B.C. asks applicants to pay more and offer flexibility

B.C. is rationing access through a competitive process. Its 2026 allocation targets total up to 400 MW over the first two years. BC Hydro’s current update reports 15 applications requesting nearly 800 MW, with final notifications delayed. Requested power is demand from applicants, not connected capacity.

The application rules require willingness to pay at least a 10% premium on both demand and energy charges. Applicants must also offer curtailment of at least 10% of requested load on 24 hours’ notice, under specified event and availability requirements. Evaluation considers utility costs and revenues alongside community, environmental, data-sovereignty and First Nations benefits.

The same document anticipates electricity agreements or tariff supplements that enforce applicants’ commitments and may include performance security. Selection gives a project eligibility to enter or retain a place in the connection queue, subject to the utility’s rules. It does not switch on the power.

This approach puts a value on flexibility as well as revenue. For a developer, the practical question is whether its service can tolerate the promised reductions. For the utility, it is whether that flexibility will be available when needed. A willingness written into a bid has to become an operating capability.

Québec’s headline price hides an important mechanism

Hydro-Québec’s February proposal described a new large-data-centre rate averaging 13 cents per kWh. That is a proposed average, not a universal energy-only price or a tariff this article can treat as approved.

The August 4 revised tariff text combines a monthly demand charge of $32.704 per kW with an energy charge of 8.710 cents per kWh. It also contains ramp-up provisions, including a charge when a customer’s highest demand over the preceding 12 consumption periods falls below 60% of the projected demand in its agreement.

The proposed mechanism would charge a customer whose consumption falls substantially short of its agreed ramp-up. Alongside payment for energy used, the utility would have a way to recover money when the expected load fails to arrive. The Régie’s docket remains open and contains October hearing records; the proposal must remain labelled as such.

A developer comparing provinces therefore needs more than a league table of cents per kWh. Demand charges, interruption terms, construction obligations and the treatment of an unfinished ramp-up can change the economics. Québec’s proposal, B.C.’s allocation process and Saskatchewan’s project arrangements are different instruments at different stages.

The host community needs a return it can measure

Federal principles also call for lasting local benefits and independently verifiable reporting on material commitments. They identify opportunities such as training, local and Indigenous procurement, research partnerships and access to computing for local businesses and institutions. None is established merely by announcing a large electrical load.

Bell says it will report annually to Saskatchewan on community investment and economic benefits. It also says studies are underway on reusing surplus heat. Annual reporting offers a useful place to assess delivery; a heat-recovery study should remain a study until there is an actual customer, connection and usable supply.

A workable public account would separate construction work from continuing employment and report Canadian supplier spending alongside access to the computing service. Where a project claims that host communities or Indigenous partners will gain particular benefits, the measures should reflect their priorities and distinguish training, contracts and ownership.

Commercial confidentiality can protect a tenant’s business and a facility’s security without making every public consequence unknowable. At minimum, readers should be able to understand the categories of costs the developer covers, how delayed demand is handled and how commitments are checked. That would give both residents and investors a clearer basis for judging the deal.

Make the deal hold up when the forecast slips

An AI project that pays for the capacity it needs and delivers useful computing can be worth welcoming. Before each new phase, its public record should identify the power available, show who funds the infrastructure and explain who carries the risk if demand falls short. Reporting should then follow the Canadian benefits actually delivered.

Canada can welcome ambitious customers while being a demanding seller of electricity. The servers should earn their place on the system—and the people hosting them should be able to see how.

Editor’s note: This is the second article in Canada’s AI: From Power to Payoff. The opening article, Canada’s AI Ambition Needs a Working Stack, examines the wider chain from infrastructure and research to practical Canadian benefits.