Canadian AI sovereignty: A dose of realism

Introduction: Why AI sovereignty demands clarity now

The intensification of great-power competition, shifting trade relationships, and rapid advances in Artificial Intelligence (AI) have placed the question of AI sovereignty at the forefront of policy debates across advanced democracies. For Canada, this conversation has taken on particular urgency. A more protectionist and unpredictable world impel Canadian policymakers to articulate what sovereignty means in the AI and digital space—and what it might cost.

Yet the term “AI sovereignty” has become so loosely defined that it risks meaning nothing. It is invoked across wildly different contexts—data residency, compute infrastructure, model training, procurement, IP, supply chains—without disciplined scoping. This ambiguity could lead governments to pursue costly initiatives that deliver symbolic sovereignty without meaningful advances. Canada is hardly alone in this challenge: Australia’s Tech Policy Design Centre recently pivoted its framing from “AI sovereignty” to “AI agency,” recognizing the former term had become less useful.¹

This brief from the Future of Canada Centre at Deloitte proposes a pragmatic and timely approach to AI sovereignty for consideration by Canadian policymakers and business leaders. It looks at much-discussed sovereignty questions—such as the reliance on hyperscalers to provide cloud services to Canadians and what thresholds should be considered in making investments that reinforce genuine AI sovereignty. 

We begin with a more useful AI sovereignty definition for Canada, then offer a starting framework for deciding when and how to pursue it. The focus is primarily on government investment decisions; while AI sovereignty raises important regulatory questions, those warrant separate treatment.

Defining AI sovereignty

Before debating where to invest public funds, we must establish what AI sovereignty means to us. In the traditional sense, sovereignty refers to a state’s supreme authority within its territory. Applied to AI, sovereignty articulates a national (or subnational) interest in asserting control particularly over two dimensions. The first is how AI systems are developed and deployed, including what safeguards protect Canadians, what values are encoded in systems that affect public life, and who owns the systems. The second dimension principally concerns geography and jurisdiction: where data resides, where computation occurs, and how it transits between systems.

Crucially, AI sovereignty exists on a spectrum with different thresholds. Some thresholds, say for national security, defence, or critical systems, might involve “sovereign ownership”— requiring that certain AI capabilities be developed and deployed entirely within Canadian borders by Canadian-owned and headquartered entities. This could include the models that power AI systems, the data centres and infrastructure enabling them, all the way through to the application layer. In an extreme example, it might even go so far as to demand Canadian-developed hardware and IP throughout the AI stack, although this is unrealistic today and likely well into the future.

Lower thresholds, for example concerning the health and financial data of Canadians, might invoke “sovereign requirements”—mandating that systems developed by foreign entities meet Canadian standards, that data remain accessible to Canadian authorities alone, and that critical dependencies be diversified rather than concentrated to ensure that Canada’s key systems remain under our control and jurisdiction. In this latter sense, a solution that meets “sovereign requirements” need not have exclusively components sourced from Canadian-owned and headquartered firms.

These definitions are intentionally narrower than some formulations in our discourse. They focus on control and jurisdiction rather than broader concepts like technological self-sufficiency or Canada-first industrial policy. Technological self-sufficiency and prioritizing Canadian firms in government procurement are both important considerations. But conflating them directly with protecting sovereignty obscures rather than clarifies the policy choices at hand.

A clear definition of AI sovereignty also acknowledges that sovereignty considerations may vary for different orders of government including First Nations, Métis, and Inuit governments asserting AI and data sovereignty for their peoples and territories. This policy brief does not delve into those distinctions; it is better left to authentic voices operating in this space, including those of the First Nations Information Governance Centre and Animkii, both working to further Indigenous data sovereignty.

A trade-off test for sovereignty investments

AI sovereignty serves legitimate ends: economic security, cultural preservation, institutional integrity, and strategic autonomy. At the same time, achieving complete AI sovereignty is effectively impossible given the complex interdependence of components and services that constitute the modern AI stack. As Minister of Artificial Intelligence and Digital Innovation Evan Solomon has articulated, Canada’s goal is “sovereignty, not solitude.” The AI stack is inherently global. It is also likely not viable to pursue many of the benefits of advanced AI without regular cooperation with allies and through multinational partnerships. Therefore, policymakers should always ask themselves whether a proposed AI/digital investment helps Canada deepen integration with trusted partners, or does it isolate the country from beneficial opportunities?

Acting to protect Canadian AI sovereignty is a technology investment decision first, and a regulatory consideration second. The reality is that digital regulations around the world are having limited effect. Strengthening the ability to control and steer systems is where the leverage lies.

However, there are also significant risks and trade-offs involved in making technology investment decisions. In the context of AI and digital sovereignty, it could mean buying or investing in technology that is not necessarily the most competitive or inexpensive available on the market but the one that best meets sovereign objectives. Rather than pursuing the chimera of absolute AI sovereignty, Canadian governments—federal and provincial—could apply three questions to any proposed investment:

First, is there a compelling policy rationale or public interest objective? Does the activity in question genuinely implicate sovereignty concerns—either the “how” of AI governance or the “where” of jurisdiction? Not every AI investment is a sovereignty investment. In many cases, buying or investing in a Canadian AI company or Canadian AI solution may be sound economic policy without being a sovereignty imperative. Or, vice versa, a decision could be a sovereignty imperative but not an economic decision.

Second, is the sovereign solution competitive? Can Canadian alternatives match or approach the capabilities of global options? If not, is the capability gap acceptable for the use case? For some applications—national security, critical infrastructure—accepting a degree of reduced performance (or comparable performance at a higher cost) may be warranted. For general productivity and business applications, cost, reliability, and performance should usually be treated as paramount.

Third, is it viable at Canadian scale? Can the proposed solution be built/bought and maintained and keep up with the global pace of evolution at reasonable cost given Canada’s market size, talent pool, energy availability, and fiscal constraints? This includes not only upfront investment but ongoing operations, upgrades, and the full lifecycle of technical debt.

Ultimately, if an investment passes the first test but fails the second and third tests, an investment may be for sovereignty’s sake—justified only in narrow, high-sensitivity circumstances where the stakes are critical for Canada. In these instances, policymakers may consider sovereign requirements short of ownership to address legitimate policy objectives.

Note that some of these considerations are adapted from the Oxford Martin School’s recent paper, A Blueprint for Multinational Advanced AI Development, which realistically considers the prospects of middle power nations pooling resources to develop frontier AI.²

In reference to this work, a fourth and final consideration may be: if a given technology solution would reinforce Canada’s sovereignty, but might be less competitive or simply not financially viable, is it possible to build it through multinational partnership? Could, for example, pooled AI compute with shared costs and economies of scale help achieve the same or similar objectives, without the fiscal burden? These types of partnerships would be quite novel, so it is up to policymakers to expand the aperture of options, and to be bold, creative, and open to pursuing them. 

Where AI sovereignty matters most

Applying this framework, a couple of priority areas for follow-up emerge for Canada:

High sensitivity use cases. National security applications, population-scale health data, and systems affecting democratic and electoral processes warrant heightened sovereignty requirements or even sovereign ownership where capability gaps exist. The policy rationale is likely sufficiently compelling to accept competitiveness trade-offs.

Critical infrastructure resilience. Energy grids, telecommunications networks, satellites,  financial systems and government operations increasingly depend on AI for operations and optimization. Ensuring these systems can function under adversarial conditions—including supply chain disruptions or foreign disputes—are areas of legitimate sovereignty concern and may benefit from additional requirements.

Where sovereignty may be uncompetitive or unviable

Hyperscaler replacement. AWS, Google Cloud, and Microsoft Azure (and other major cloud providers) are too embedded, too capable, and too cost-effective for wholesale replacement. Canadian alternatives in general-purpose cloud computing for most sectors and uses would likely fail the competitiveness and viability tests decisively. In many cases, white-labelling, air gapping arrangements, or sovereignty requirements introduced through procurement may provide meaningful control. This does not mean Canada should avoid building Canadian-owned AI compute or data infrastructure, but that our country and leaders should be very strategic about where to invest.

Productivity applications. Sovereignty measures that create friction or impose significant cost premiums on routine AI adoption risk compounding Canada’s existing productivity challenge. For applications without clear sovereignty implications, the trade-off calculus should strongly favour competitiveness and global integration.

Regulatory capture through walled gardens. Some sovereignty advocacy may be driven by domestic firms seeking protected markets. The test: does this measure serve Canadian economic security and/or national security, or does it primarily create moats for less competitive players? This question should still be asked rigorously of Canadian solutions, while recognizing that the scales in procurement conversations and domestic industrial policy have often been tilted against Canadian companies.

The missing ingredients

Canada can reinforce its AI sovereignty, but it will be up to federal, provincial, and territorial leaders across the political realm and public service to build the capacity, clarity, and ecosystems to succeed. It will also be up to Canadian business leaders and technology experts to expand their horizons in terms of what they are willing to offer up (for example, secondments, tours of duty, and knowledge from their firms) to further the interests of their country. There are three other key missing ingredients:

State capacity. Government needs technical talent to be an informed buyer rather than a passive procurer. This means competitive compensation and internal capability to evaluate sovereignty claims and manage complex technology programs. Jurisdictions that have invested in public-sector technical talent—the UK’s Government Digital Service, Singapore’s GovTech, the U.S. Digital Service—demonstrate materially better outcomes in AI procurement and deployment. Recent Canadian initiatives like the proposed Build Canada Exchange program could improve compensation and recruit technical leaders. They are steps in the right direction although their adequacy and implementation remain to be demonstrated.

AI framework clarity. AI sovereignty must be pursued within Canada’s complex jurisdictional landscape. The federal government could lead on articulating which layers of the technology stack matter for sovereignty purposes, what thresholds apply at each layer, and who is accountable in each domain. Provinces are already moving independently on regulation—British Columbia and Quebec notably—but federal leadership is essential for coherence. This is not the place or the time for new inter-provincial barriers. (Again, this brief focuses primarily on investment decisions; the regulatory dimension, while critical, warrants separate treatment.)

Ecosystem enablement without capture. Beyond internal capacity, government could enable private-sector delivery through demand signals (e.g., procurement commitments that de-risk private investment), anchor tenancy (government as lead customer for emerging Canadian capabilities), and targeted R&D support for strategically important but not immediately viable activities. Canada has historically – and to this day – not invested enough in building and retaining domestic technology champions.

Considering the UK approach

The United Kingdom offers a relevant reference model for some of these objectives. UK AI policy broadly balances sovereignty ambition with pragmatic execution: public compute investment through the AI Research Resource, dedicated sovereign AI units within government, competitive public-sector compensation, and clear procurement signals—all while collaborating with hyperscalers rather than attempting to replace them wholesale. Given similar scale constraints, talent dynamics, and governments, the UK experience merits close Canadian attention. However, it should be noted that some of the recent major UK AI investments in partnership with the US firms are being delayed due to other trade disputes. Given this consideration, it is essential that Canada invests in reliable partners and builds out alliances to ensure it can continue to grow and prosper unencumbered.

Conclusion

AI sovereignty is a means to more strategic autonomy and national capability, not an end in itself. Pursued without a clear frame, it might produce little more than high cost symbolism. Pursued with focus—grounded in clear definitions, subjected to trade-off analysis, and implemented with adequate state capacity—AI and digital sovereignty will strengthen Canada’s position in an increasingly contested technological landscape. The path forward requires both ambition and restraint: ambition in building genuine capabilities where they matter, restraint in avoiding sovereignty theatre where they do not. The time to act is now; clearer definitions will make for better results.

Footnotes

1. Tech Policy Design Institute, “From AI Sovereignty to AI Agency,” November 2025.
https://techpolicy.au/ai-agency

2. Abecassis, Adrien, Jonathan Barry, Ima Bello, Yoshua Bengio, Antonin Bergeaud, Yann Bonnet, Philipp Hacker, et al. “A Blueprint for Multinational Advanced AI Development.” Oxford Martin School, November 24, 2025.
https://aigi.ox.ac.uk/wp-content/uploads/2025/11/blueprint_for_multinational_ai_development.pdf