Canada Built the AI. Now Build the Market.

From: Harvey Naglie 

To: Innovation Observers 

Date: September 8, 2026 

Re: Canada Built the AI. Now Build the Market. 

Canada's artificial-intelligence problem is coming into focus. We know how to produce research, talent and promising companies. We have not learned how to give those companies a domestic market large and coherent enough to scale in. 

I have argued in this space that fragmented regulatory and procurement systems risk turning a small domestic AI market into a collection of even smaller ones, and that the test for any support instrument is whether Canada will still have something when the funding ends. A new C.D. Howe Institute working group communiqué brings those arguments together and complicates both. 

Health is a useful test case. Canada has world-class AI researchers, extensive health data, sophisticated institutions and clinicians willing to adopt useful technology. However, fragmentation across 13 provincial and territorial procurement and regulatory systems is the principal structural barrier to commercialization, according to the communiqué. What looks like one large national market behaves like several smaller ones. That is not simply a regulatory inconvenience, it is a barrier to scale.  

There is evidence that a different model works. 

Canada Health Infoway's AI Scribe Program set common clinical, business, privacy and security requirements through a national advisory group, pre-qualified vendors after testing and live demonstrations, and offered clinicians a funded 12-month licence rather than leaving every institution to assess technology on its own. Demand outran the design: More than 6,000 providers registered within two days and more than 10,000 within two weeks, exhausting the notional allocation and prompting several jurisdictions to close registration. The program has since supported more than 11.5 million patient encounters. 

It also measured results rather than counting licences. Roughly 90 percent of clinicians rated the tool valuable to their practice in an independent evaluation of more than 12,000 clinicians; about 80 percent reported reduced cognitive load and close to 70 percent reduced administrative burden. Time savings were widely reported but largely recycled: Of the three-quarters who reported them, about half redirected the time to other administrative or clinical work and only about a fifth to seeing more patients.  

Coordination is no guarantee of quality, as subsequent testing by Ontario Health and ministry staff found that all 20 evaluated products exhibited at least one issue: More than half produced incorrect medical information, 45 percent produced fabricated content and 30 percent omitted material. Those findings describe tools available through the summer of 2025 in a fast-moving field, and human review remains the necessary safeguard. The underlying point is that buying together is an advantage only if what is bought together has been properly qualified. 

AI adoption is not the same thing as AI commercialization, and neither necessarily produces Canadian value capture. 

A nationally coordinated procurement system could spread AI through Canadian hospitals and clinics while most of the resulting intellectual property, enterprise value and supplier capability accrues elsewhere. Canadian healthcare would be better for it. Canada would still have failed at commercialization. 

Procurement policy therefore needs two tests rather than one. An adoption test: Does the technology improve productivity, service quality or other measurable outcomes? And a capability test: Does the way Canada buys and deploys it help competitive Canadian firms develop the customers, expertise, intellectual property, capital and market access they need to scale? The capability test asks a different question: Not who is favoured, but what remains. 

Market architecture is itself productive infrastructure. 

Compute, talent, research institutions and common technical standards are infrastructure. So are interoperable regulatory and procurement requirements that let a successful Canadian company move from one province to the next without effectively starting over. Interoperability rather than uniformity is the objective, and the Committee on Internal Trade already provides a federal-provincial-territorial mechanism through which that work could be advanced. 

Harmonization requires 13 governments to accept constraints on purchasing discretion each currently exercises for its own reasons. That is a reason to start narrow, where the payoff is demonstrable.  

Health AI is that narrow place. Governments could develop a common qualification framework for defined classes of health-AI applications, with shared evidence requirements for safety, privacy, cybersecurity, performance and interoperability – the requirements the Auditor General found missing the first time. A product meeting them would not be automatically purchased everywhere, but it would not need to establish the same credentials from scratch 13 times. 

The scoreboard would then need to extend beyond spending and adoption counts: the time and cost for a qualified Canadian company to enter additional provincial markets, and how many Canadian firms convert pilots and funded licences into repeat paying customers.  

The uptake of AI scribes shows Canadians will use it when it solves a real problem. The Auditor General's findings show that enthusiasm is not the same as diligence. The remaining challenge is to build a common, large domestic market in which Canadian technologies meeting a demanding standard can actually scale. 

Harvey Naglie is a financial services policy analyst and a former senior policy advisor with the Ontario Ministry of Finance.   

To send a comment or leave feedback, email us at blog@cdhowe.org.    

The views expressed here are those of the author. The C.D. Howe Institute does not take corporate positions on policy matters.