OpenAI is weighing Canada as a site for new artificial intelligence data centre capacity, but the two provincial power grids most often marketed to hyperscalers have already run short of the electricity such a project would require.
George Osborne, the former British chancellor who now leads OpenAI’s global Stargate data centre expansion, told the ALL IN conference in Montreal on Sept. 17, 2026 that the company has made no decisions but that Canada “would be an obvious place to look,” according to BetaKit. He cited the country’s energy supply, available land and what he called a “can-do attitude,” and praised Prime Minister Mark Carney’s national AI strategy for its emphasis on adoption. Bloomberg reported separately that OpenAI executives raised the prospect of Canadian data centre partnerships at Carney’s investment summit earlier in the week.
The supply picture is considerably tighter than the pitch suggests.
In Alberta, the province widely described as the centre of Canada’s data centre build-out, the Alberta Electric System Operator has already committed every megawatt available under its interim 1,200-megawatt cap for large-load connections. TransAlta’s Keephills project received 230 megawatts, and a Pembina Pipeline project near Edmonton, reported as a prospective Meta site, took the remaining 970 megawatts. As of July 30, data centre proposals sitting in the AESO queue represented close to 19,600 megawatts of potential new demand, roughly 16 times what the operator was able to hand out. That leaves 37 remaining proposals to either bring their own generation or wait for transmission to catch up. Under Alberta’s data centre regulation, in force since June 2026, projects that pair their demand with dedicated new generation or storage move to the front of the connection line.
Quebec, long sold to global operators on the strength of inexpensive hydroelectricity, has tightened in the same direction. Any new load of five megawatts or more now requires authorization from the province’s minister of economy, innovation and energy, who may refuse an application outright or defer it indefinitely. The province has signalled a competitive call for projects that would cap allocation over a two-year window at 300 megawatts for AI and 100 megawatts for other data centres. Hydro-Quebec has separately asked the provincial energy board to roughly double the rate charged to new data centres drawing more than five megawatts, to 13 cents per kilowatt hour. Quebec’s data centre sector currently draws about 200 megawatts, and the utility projects peak demand of roughly 1,000 megawatts by 2035.
The arithmetic matters because it reframes what Canada is actually offering. A single hyperscale AI campus of the type OpenAI has built elsewhere can draw several hundred megawatts on its own, which is more than the entire two-year AI allocation Quebec has proposed and a substantial share of what Alberta was able to release in a full cycle. Osborne’s own caveat pointed in the same direction: data centres are long-term, capital-intensive projects, and the company needs confidence that a jurisdiction will remain predictable for decades.
For Canadian businesses and ratepayers, the practical question is no longer whether the country can attract the investment. It is who pays to build the generation and transmission that would make it possible, and on what terms. Both provinces have answered by pushing that cost toward the developer, through bring-your-own-power rules in Alberta and higher tariffs plus ministerial discretion in Quebec. Any OpenAI decision on Canada will be made against those rules rather than against the abundant-energy story that has been used to sell the country abroad.
