Hydro-Québec’s proposed data centre rate may help protect households from the cost of AI’s power demand. Whether it brings Canada closer to sovereign compute is a separate question.

On Hydro-Québec’s own price comparison, Montreal offers a large industrial customer the second cheapest power in Canada: 5.52 cents a kilowatt hour, behind only Winnipeg. Now Hydro-Québec proposes that large data centres pay 13 cents. If the Régie de l’énergie agrees, then on today’s rates Quebec would move from among the cheapest places in Canada for large power users to among the most expensive.

“The period of energy surpluses is now behind us,” Hydro-Québec’s Sarah Trabelsi told the opening day of hearings on October 1 (translated). Its supply plan expects data centre demand at winter peak to grow roughly sixfold, from 180 MW in 2024-25 to 1,102 MW by 2034-35.

Cheap power has drawn data centres to Quebec, but not necessarily Canadian ones. According to a research map, none of the 644 MW planned or under construction is Canadian-owned. That suggests a trade-off facing the country as a whole, between cheap power, AI capacity under Canadian control, and fair rates for households. Pushing hard on any two may come at the cost of the third.

The price of the next kilowatt hour

Large data centres pay 6.82 cents on average today, less than households pay for their first block of power (7.065 cents), on Hydro-Québec’s 2026 rate grid. The proposed CD rate would apply to existing and new sites with 5 MW or more of authorized capacity and phase in to 13 cents by 2030. The logic is marginal cost pricing: new load should pay what new supply costs, rather than draw on the heritage pool of old, cheap hydro reserved for Quebec customers. Google wants existing sites left on their current rate and argues that new generation “will function as system resources providing benefits to all” customers.

Smelters and hydrogen producers also need new power, so why give data centres a rate of their own? Part of the answer is pace: Hydro-Québec expects their demand at peak to grow roughly sixfold in a decade. Large new loads of every kind already face scrutiny, since Quebec requires government authorization for any new or added load of 5 MW or more. Whether the same pricing logic should eventually extend to them is a fair question.

There may be a middle path, in which new load bears a large share of the cost but not necessarily all of it. The sums are significant: serving about 9 TWh of data centre demand at marginal cost would add roughly $1.3 billion a year in supply costs (2026 dollars). Whatever data centres do not cover would fall to other customers.

Price is unlikely to decide where data centres locate; power availability, speed of connection and regulatory certainty tend to matter more. But at the margin, price counts. Large sites can bring investment, jobs and a broader provincial tax base, though the gain is easy to oversell: they employ relatively few people for their size, and power sold at cost adds nothing to the utility’s bottom line.

Rates are not the only tool

Governments can also ask operators to contribute directly. Alberta’s data centre levy charges up to 2% of computing hardware value, and Alberta’s 2026 budget ties it to grid power actually drawn. A non-refundable credit against Alberta corporate income tax means profitable operators pay nothing extra, so it works as a minimum contribution. A levy would likely work best alongside a cost-based rate rather than in place of one.

Making the rate flexible

Google objects to charges on unused power, including a premium when a site’s peak demand stays below 60% of its planned ramp for a full year. Its own counterproposal would let a customer reduce its reservation, by up to the lesser of 10% or 50 MW a year, if Hydro-Québec has not yet invested to serve it. Going further, a workable valve would require 24 to 36 months’ notice, return released capacity to the authorization queue, and charge the operator the supply cost committed on its behalf minus what Hydro-Québec recovers by reselling it. The aim is flexibility, not a penalty.

Quebec’s tightest moment is the winter peak, so the rate could also reward sites that agree to reduce consumption at peak, where that power can serve others.

The 5 MW threshold may also need regular review. Option consommateurs, which supports the rate, has asked for a status report in the next rate case, including on projects split to stay under the line.

The competitiveness risk

A higher rate still carries a competitiveness risk, though the comparison is rough: the table uses base rates at an 85% load factor (how steadily a site draws power), the CD rate assumes 95%, and Ontario’s programs for large users are left out.

There is no public sign of a company threatening to leave; interveners argue fairness and cost. Nor are alternatives wide open. British Columbia makes AI and data centre projects compete for up to 400 MW in the first two years, and Ontario has legislated a ministerial approval gate for connections, with rules still being finalized. The more likely risk is a quiet shift in where the next site goes.

The sovereignty contradiction

Ottawa’s AI for All strategy wants large-scale sovereign compute, meaning Canadian-controlled AI capacity; partners have proposed 850 MW by 2030. Governments want homegrown AI, yet are rationing the power that data centres need. The tension is over price and priority, not refusal: BC lists data sovereignty among its criteria for allocating power, and Ontario proposes to.

The map by Collectif NRS, with COlab and Université Laval, finds 572 of 1,488 MW of installed capacity is Canadian-owned, though that total includes crypto mining; the share that is AI capacity, the kind Ottawa means by sovereign compute, is likely smaller. Ownership is half the battle. Physical control is leverage: provinces decide who gets power, and the Investment Canada Act allows national security review of any foreign investment, including new facilities, on grounds such as sensitive technology and critical infrastructure. But location is not enough. A federal white paper warns that data stored in Canada “may be subject to the laws of other countries.” Law and contracts matter as much as concrete. The ideal is compute that is Canadian end to end.

A data gap, and Ottawa’s limits

There is no official public inventory of Canada’s data centres: where they are, how much power they draw, and who owns them. The first public map of Quebec’s sites came not from a government but from a research collective. Without that data, it is hard for any government to track its own sovereignty goals.

Ottawa cannot set the price. Its data centre principles include protecting ratepayers, but they are voluntary, and rates are provincial. Ottawa’s main tools are incentives, most plausibly funding. One option worth exploring with the provinces: ask projects that receive federal compute funding, such as the planned $700 million expansion of the Compute Access Fund, to report publicly on their power use and how its cost is shared.

The Régie resumes hearings in December, and other provinces will be watching. A cost-based rate can do one job well: keep households from paying for the AI boom. It cannot do the other two. Cheap power has so far drawn capacity that Canadians do not own, and a higher price will not change that. Sovereign compute will come, if it comes, from choices about who gets power, on what terms, and under whose law.