TORONTO — Canada can capitalize on the global AI data centre boom and make more computing capacity available to domestic firms and researchers if it clears the way for digital infrastructure and offers the right incentives, the federal government has been told. 

Ottawa should help secure power for new AI facilities, speed up their approval and perhaps even subsidize them, several members of the AI Strategy Task Force said in memos submitted late last year. Some also called for the federal government to support the long-term development of homegrown hardware and software to run AI facilities, though they were more divided on what role foreign cloud services and suppliers should play in Canada’s AI compute build-out in the short term.

AI Minister Evan Solomon announced the 28-person task force, composed of founders, investors and researchers, in late September and tasked them with making proposals for the upcoming renewal of the national AI strategy. Among other questions, they were asked to estimate how much “sovereign” compute capacity Canada will need in the future, as well as identify gaps in current infrastructure and the factors preventing domestic firms from filling them. 

Several submissions cited paperwork delays, power constraints and networking issues as challenges. Build Canada chair Daniel Debow and former Vector Institute CEO Garth Gibson both called for special zones where firms could get their data centres permitted quickly. Firms building AI infrastructure “wish to invest billions,” and policymakers should help them do that, Debow wrote. All of the submissions made by the task force can be viewed here.

CoLab CEO Adam Keating, meanwhile, recommended that the federal government seek out up to five “national building megaprojects,” where large data centres would be located alongside major energy or mining sites. Once designed, the new facilities should be up and running within nine months, he said, referencing xAI’s speedy compute build-out in the U.S. Elon Musk’s firm claims to have constructed its Colossus supercomputer in Memphis, Tenn., in 122 days, though it moved into an existing facility and initially used some illegally-generated power, according to the U.S. Environmental Protection Agency. 

Zoning and permitting are typically municipal issues, while electric utilities are often run and regulated by the provinces. Ian Rae, CEO of Toronto-based compute management firm Aptum, called for Ottawa to work with the other levels of government and Indigenous communities to find sites for “pre-approved and shovel-ready development” of data centres. That could help cut the time to get a 100-megawatt data centre online from five years to two, he said. 

Canada has natural advantages that it can use in the AI race, task force members argued. Several cited the cold climate, skilled workforce and availability of clean energy. As countries around the world compete for AI infrastructure investment, Canada should give itself a bargaining chip by becoming a major producer of something that’s indispensable for compute build-out, said Ajay Agrawal, a University of Toronto professor and founder of the Creative Destruction Lab accelerator. He suggested focusing on power, critical minerals or advanced semiconductor packaging. 

Still, Canada doesn’t yet have a limitless supply of electricity it can devote to AI infrastructure. Provincial utilities and regulators in Alberta and British Columbia are restricting grid hookups for new data centres, while Ontario and Quebec try to build their way out of looming energy crunches. 

Ottawa could encourage the generation of more power by making or underwriting long-term purchase agreements, Rae and Cardinal Policy founder Marc Etienne Ouimette said in separate submissions. Inovia Capital partner Patrick Pichette said Canada should aim to add up to 250 terawatt-hours worth of new electricity capacity per year over the next seven years, amounting to about a tenth of the expected global demand from AI compute.

Task force members have called for Ottawa to use its buying power and balance sheet to help grow the AI industry at home, and some made similar recommendations on AI infrastructure. Federal and provincial governments should act as anchor customers for new compute and cloud capacity that’s built “as Canadian as possible,” said Benjamin Bergen, then-president of the Council of Canadian Innovators, a scaleup lobby group. Rae said the federal government should set up a compute infrastructure fund to back Canadian firms building data centres or offering cloud services with long-term, low-cost capital. 

Ottawa could also help by bulk-buying chips. AI data centres are filled with expensive graphics processing units (GPUs), often from Nvidia, that are hard to come by. “A small company wanting to experiment with a small number of the latest GPUs is likely to find that their order is queued for a long time,” Gibson argued. He claimed that Nvidia has signalled it would be willing to redirect some of the leading-edge chips currently reserved for cloud giants to Canada if the federal government asked. Ottawa did sign a non-binding agreement with Nvidia in February 2024 to collaborate to increase the country’s AI capacity, but neither side has disclosed what that work actually entails.

Prime Minister Mark Carney has said the new federal Major Projects Office will work to help develop a “Canadian sovereign cloud,” to ensure the country has “independent control over advanced computing power.” Ottawa has yet to reveal details of the initiative, and is still working to define what “sovereign” means. In the meantime, U.S. tech giants like OpenAI and Microsoft have pitched themselves as partners to Canada’s sovereign AI ambitions, while homegrown players like Aptum, Bell and Cohere have countered that Canadian ownership and control is a prerequisite.

Task force members agreed that Canada can’t currently go it alone to expand its compute capacity, but opinions differed on the role foreign firms should play. “A fully sovereign Canadian cloud is unattainable in the short or medium term,” said Ouimette, a former AI policy executive at Amazon Web Services. Beyond specialized chips—in which Canada has recently lost its leading contenders—the AI “stack” includes hardware such as electrical and networking equipment as well as software to train and manage models and data. Canadian firms currently produce some of those components, but not all of them. 

Ottawa should launch a funding program to help fill the gaps in the AI stack over time, Ouimette said. In the short term, he proposed a new “strategic adoption cluster,” which the federal government could underwrite with power purchase agreements or loan guarantees. Domestic infrastructure firms would power and set up the data centres, and a portion of their compute capacity would be reserved for Canadian companies and researchers. The cluster could try to sign long-term contracts with large foreign model developers like Anthropic and OpenAI as well as homegrown players like Cohere to buy the rest, Ouimette said.

Other memos also suggested leveraging international demand to help expand Canada’s compute capacity. Keating said Ottawa should offer “aggressive incentives” to model-makers like OpenAI and cloud firms to build data centres in Canada. Rae suggested the federal government court “foreign capital and big tech” to join it as the anchor tenants of two or three new cloud computing campuses around the country.

LawZero co-president Sam Ramadori went even bigger, calling for Canada to spearhead a “middle powers sovereign AI coalition,” in which like-minded countries would all commit capital to build compute capacity that they’d share. Strong candidates would include France, Germany, Japan, South Korea and U.K., he said. The coalition would “create a powerful democratic alternative to the U.S.-China AI duopoly.”

Some task force members called for new rules requiring foreign cloud providers to maximize or ensure Canadian control over data and applications that serve Canadian consumers, companies and governments.

The recommendations come as Ottawa rolls out funding from its AI compute strategy, and refines its approach to digital infrastructure. November’s federal budget gave Solomon the mandate to negotiate deals to back data centre projects, and the innovation department is currently soliciting proposals from firms that want to build them. It’s focusing on facilities larger than 100 megawatts, and will favour bids with Indigenous participation, sustainability measures and Canadian suppliers. 

The Liberal government will release its updated AI strategy this quarter, Solomon said last month.