Nathan Fast is Director of Branding and Marketing at Vancouver-headquartered HIVE Digital Technologies Ltd.

Canada has never had trouble inventing the future. Holding onto it once it’s real is the harder part.

Insulin came out of a lab in Toronto. So did the pacemaker. So did much of the math behind the AI systems running across the world right now. Geoffrey Hinton and Yoshua Bengio did the foundational work on deep learning in Canadian universities, on Canadian grants. Then the industry their work created largely grew up somewhere else.

That pattern shows up twice in how Canada handles AI, not once.

The first version is about building infrastructure and keeping it Canadian. That starts with training the researchers, publishing the papers and spinning up the startups. Then the compute, the data centres and the actual physical plumbing of the AI economy tends to get built wherever the capital finds it easiest, along with a lot of the talent that goes to run it.

Where that infrastructure sits, and who owns it, matters more than it used to. AI has changed what a company’s most sensitive information looks like. It used to be a database behind a firewall. Now it is the prompt itself.

Every time a hospital, a bank or a law firm runs an AI system, it is feeding patient records, financial histories or client files directly into the model doing the work. Checking the box for the Canadian data centre on a foreign-owned platform does not change who can be compelled to hand that data over.

The second version is adoption, and it might be the more stubborn one. Statistics Canada put business AI adoption in Canada at around 12 percent as of last year. Small and mid-sized firms, the backbone of the economy, sit even lower, well behind Nordic countries, Germany and France on the same measure.

Canada built some of the best AI research infrastructure in the world, and Canadian companies are still among the slowest in the developed world to actually use what came out of it.

The two gaps feed each other. If the infrastructure running Canadian AI sits outside the country and outside Canadian legal control, and the customers buying AI products default to whichever big foreign vendor is easiest to buy from, a Canadian AI company has a hard time finding a reason to stay Canadian. Ask any founder who has had to look abroad for a distribution partner because the local meeting never got booked.

The federal government’s AI for All strategy, launched this June, names both gaps directly. One pillar is building a sovereign AI foundation. Another is driving adoption, including a “Buy Canadian” commitment to make government an actual customer of Canadian AI firms instead of a bystander.

That is a useful signal. It means this is not just industry talking its own book. The strategy also sets a real target: business adoption climbing from that 12 percent baseline toward 60 percent by 2034. Worth watching whether the follow-through matches the ambition. 

Closing the gaps

A strategy on paper is a start, not a finish. What actually closes these gaps is slower and less glamorous.

On infrastructure, it means building AI compute capacity in Canada, on Canadian power, under Canadian law, which takes years of permitting, utility coordination and construction done properly. That means real community input, not a rushed announcement, and no shortcuts around the towns and regions actually hosting the build.

Canada has a real advantage in this, if it’s built with patience instead of urgency. Quebec, Manitoba and British Columbia run largely on hydro that is cheaper and cleaner than what is available in the U.S. markets currently straining under their own data centre demand.

BUZZ High Performance Computing (BUZZ HPC), among others, is working to build a coast-to-coast sovereign AI compute backbone for Canada. [HIVE Digital Technologies is the parent company of BUZZ HPC]. That work is still early, and it’s one project among many still working to earn its place, not something anyone should treat as a foregone conclusion.

On adoption, it means Canadian companies and institutions actually buying from the AI companies training their models in Canada, not defaulting to whichever foreign platform has the biggest ad budget. It means procurement cycles that move faster than nine to 12 months, because the technology does not wait that long for anyone. And it means treating a Canadian AI vendor with a real product as a safer bet than an unproven one, not a riskier one.

Canada has never been short on ideas. The patience to keep building past the idea stage has been in shorter supply, and the follow-through to buy what gets built once it exists shorter still.

That responsibility sits across the board, not just with Ottawa. Procurement leads can put a Canadian AI vendor on the shortlist next time instead of defaulting to habit. Enterprise buyers can ask where the model they are about to run actually sits, and under whose law.

Policymakers can build on the thinking already behind AI for All, one permit, one power allocation, one contract at a time. The researchers who started this field, and the power to run what they built, are both still in the country. The follow-through is the only thing left to build.

R$