A copy of ‘The European Union Artificial Intelligence Act’ during the AI & Big Data Expo 2025 at the Olympia in London, England in February, 2025.Isabel Infantes/Reuters
Jaxson Khan is a senior fellow at the Munk School of Global Affairs & Public Policy at the University of Toronto, where he is co-director of the AI Competitiveness Project. He is the co-author of the recent report Sovereign by Design. Mr. Khan also serves as chief executive of Aperture AI.
As a middle power, Canada must make strategic bets. We can’t do everything. Given how quickly artificial intelligence is developing and reshaping the global economy, it is particularly important that we determine our strategy and path forward. But that doesn’t mean we should fall into the binary trap.
On the technology stack, it is rapidly becoming clear that U.S. models (ChatGPT, Claude, Gemini) and Chinese models (DeepSeek, Qwen, Kimi) are dominant on frontier capabilities and market share. These advances are heavily supported by state-driven subsidies, government and defence contracts, and national security strategies.
On AI governance, there is political pressure for Canada to move toward a more U.S.-style, light-touch approach or a European-Union-style, highly regulated market. Neither the U.S.–China tech binary nor choosing between U.S.–EU regulatory alignment will work well for Canada.
Instead, we need a third path that strengthens Canada’s competitiveness while protecting our economic and digital sovereignty.
Other middle powers – including Britain, Australia, Japan and South Korea – are grappling with the same challenges. There is no perfect solution, but some common patterns are emerging.
Telus plans AI data centre expansion in B.C., including two new centres in Vancouver
The first is that no country, particularly a middle power, can own the entire AI stack. The current Canadian discourse on AI sovereignty conflates five very different things: compute, model, data, cultural and regulatory sovereignty. Treating them as one bundle could invite expensive nationalist projects where the economics do not hold and regulatory posturing where rules don’t make a meaningful difference.
A key chokepoint is cloud infrastructure: We depend heavily on U.S. hyperscalers. However, building all our own infrastructure is neither feasible nor desirable given the cost premium. In other areas, we have real strengths to build on, including in enterprise AI models, where Cohere recently reached a $20-billion valuation through a merger with German Aleph Alpha to focus on sovereign AI, and AI applications, where there are many emerging Canadian players.
A decision-making framework for AI infrastructure (and other major investments) could borrow from Canada’s new Defence Industrial Strategy, with one adjustment: buy, partner, build.
Buy and deploy the best AI fast. Partner with allied countries, particularly for frontier model development. And then build where Canada can own a meaningful slice of the stack, including in the AI application layer, and emergent areas like quantum, where Canadian companies like Photonic and Xanadu demonstrate we still have a window to compete.
Underlying all of this is data. We could tier our policy approach by data sensitivity. Classified workloads, sensitive personal data held by governments and regulated sectors like banking and health care, and routine commercial data each may demand very different treatments of data sovereignty. We also could modernize our federal security framework to stop overclassifying data.
On governance, neither Brussels nor Washington should be our model. The EU has a comprehensive AI Act but has slow AI adoption and innovation. The EU has now passed a Digital Omnibus legislative proposal to simplify its own regulations. The U.S. has world-leading frontier capabilities but a shifting rulebook that may not provide the stability needed to build trust and drive adoption. Canadian AI policy has the opportunity be interoperable with allies, proportionate to actual risk and explicitly designed to accelerate adoption.
Rulemaking on AI without underlying capability in AI is not leadership. Instead, it will keep us on a path of technological dependency.
Canada’s productivity gap is also a real sovereignty problem. Health care, natural resources, defence and financial services are where AI could make or break economic competitiveness this decade. If our hospitals, miners, banks and shipyards do not run on competitive technology, no amount of government infrastructure projects can drive our economy.
A serious AI sovereignty agenda has at least three key pillars. Own the infrastructure that is non-negotiable for national security. Reduce vulnerabilities where the economics hold, for example sovereign cloud at the highest tiers of data sensitivity. And drive the broad adoption of the best AI across our strategic sectors.
Ultimately, countries with deep AI capability will shape the rules for AI. Countries that only regulate will be regulated by others. AI competitiveness is a prerequisite for credible rulemaking, not an alternative to it.
The choice for Canada is not choosing between Washington, Brussels or Beijing. It is whether we design an AI economy of our own, or lose competitiveness and leverage, while watching others write the rules of our economy.