The false promise of sovereignty economics

Meta’s recent announcement of a $13 billion CAD, one-gigawatt artificial intelligence data centre in Sturgeon County, Alberta—paired with a dedicated $4.6 billion natural gas plant under the province’s “bring your own power” rules—is one of the largest private-sector investments in Canadian history. The major announcement lands inside a debate that has come to dominate Canadian economic and technology policy. To the Council of Canadian Innovators (CCI) and its chair, Jim Balsillie, the practice of foreign multinationals building Canada’s digital infrastructure is evidence of a “rentier economy,” in which Canada supplies the land, the power, and the data while the returns flow elsewhere. According to the CCI’s Prairie director, such investments in Alberta merely reflect our “extractive relationship” with the United States and Canada’s “branch plant” status. This worldview, which reframes foreign investment as a problem rather than an opportunity for Canada, has moved from advocacy to the core of national strategy. Ottawa’s own AI strategy, released in June, states plainly that “the underlying infrastructure must be operated under Canadian control and Canadian law,” and commits to building “sovereign compute” capacity of its own. Call this doctrine “sovereignty economics.” Its core claim is that Canada’s prosperity and security in the digital age depend on domestic ownership and control of technology—firms, infrastructure, and intellectual property—and that the state’s central economic task is to engineer that ownership through subsidies, preferences, protections, and procurement. Its adherents include some of the country’s most successful entrepreneurs and most energetic policy advocates. Its influence now extends from advocacy campaigns to party platforms to the machinery of government. Its ascendance is one of the more consequential developments in Canadian political economy in recent memory, which is why it deserves to be taken on directly. The purpose of this DeepDive is to describe sovereignty economics and the sources of its rise; to show why its prescriptions would leave Canada poorer and, ultimately, less sovereign; and to set out the alternative—call it “diffusionism”—that better fits the economics of general-purpose technologies and the realities of a small, open economy. The contest between these two worldviews will shape Canadian policy for years to come. So far, only one side has been making its case. The sovereigntists are describing something real. Data moves under foreign law. Intangible assets behave differently from factories. A platform that a foreign government can switch off is a genuine vulnerability. And nothing in what follows argues against Canadian technology champions. Canada can build globally successful firms and should want more of them. The argument is against the proposition that government policy should use brute force to engineer them in sovereignty’s name. That proposition fails on two counts: it overstates the state’s capacity to do the engineering, and it overestimates the benefits of domestic ownership while underestimating the gains from global integration. The first is a claim about what governments can do. The second is a claim about what’s worth doing. Sovereignty economics gets both wrong. Start with capacity. The five largest American hyperscalers (the handful of firms, including Amazon and Microsoft, that build and run the world’s largest cloud and AI computing infrastructure) spent a combined $448 billion USD on AI infrastructure in 2025—more, in Canadian dollars, than the federal government’s entire $595 billion budget for this year. A country that cannot match 12 months of private hyperscaler spending with the totality of its public resources should not organize its economic strategy around pretending otherwise. The second count cuts deeper. The economic history of general-purpose technologies teaches that the returns to ownership flow to shareholders, while the returns to adoption flow to everyone: equity markets reward the first, productivity and living standards reward the second. Canada has work to do on both fronts, but it is the usage gap that sovereignty economics ignores almost entirely. Sovereignty gospel The tech-sovereigntist position is built on a sophisticated premise that the global economy has undergone a structural shift from physical production to intangible assets, and in this landscape, wealth and power belong to those who own the underlying intellectual property, the algorithmic architectures, and the data repositories. Its concerns centre on three arguments: Data sovereignty—Physical location is not the safeguard it appears to be. Under the 2018 U.S. CLOUD Act, American authorities can compel any provider under U.S. jurisdiction to produce data it controls, wherever that data sits. A Canadian company storing information with a U.S.-controlled cloud provider hasn’t fully escaped American legal reach. The tenant dynamics—If domestic innovators rely entirely on foreign compute infrastructure, they become corporate tenants. The facilities may sit in Alberta, but the intellectual property, commercialization, and commercial margins flow back to Silicon Valley. The critical mass problem—Without a homegrown champion backed by state-supported sovereign compute, Canada risks losing its technical elite to foreign ecosystems, sealing a long-term productivity decline. Parts of this are legitimate. Canada’s own Treasury Board has acknowledged that as long as a cloud provider operating in Canada remains subject to a foreign country’s laws, Canada will not have full sovereignty over its data. For instance, the Department of National Defence runs its everyday collaboration tools on Defence 365, a Microsoft 365 platform, and no amount of data-residency policy fully mitigates that exposure. A platform that a foreign government can restrict or withdraw at will is a vulnerability for systems that need to keep running. Ottawa is right, therefore, to treat this as a real problem for critical government and defence functions, and right to screen foreign control over assets touching those functions narrowly, through the Investment Canada Act’s national-security review or its own procurement processes. But that is a targeted response to a specific risk. It is a different thing entirely to conclude that an ordinary commercial data centre needs the same protection. The distinction, in broad terms, is between procurement and state-led development. Buying sovereign capacity for defence workloads is the state acquiring what it needs on conditions it sets. Funding a champion’s commercial buildout is the state deciding which firm should win a market. The first is legitimate and bounded. The second is the doctrine at issue, and the sovereigntist method is to dress the second in the language of the first. Conflating the two is where the argument tips into digital autarky, misreading both Canada’s capacity and its needs. Small open economy, real constraints The foundational flaw in the sovereigntist worldview is a refusal to accept Canada’s economic reality. We are a small, open economy of 41 million people, roughly one-eighth the size of the United States and one-eleventh of the European Union, lacking both a market of hundreds of millions and the fiscal capacity to out-subsidize superpowers. Compare recent government outlays on AI and other technologies. Washington’s Inflation Reduction Act committed hundreds of billions in production credits to reshore clean-tech manufacturing. Brussels answered with a different instrument aimed at the same goal: “Made in EU” procurement and content rules under its Industrial Accelerator Act, which redirects billions in existing public spending toward EU-origin firms. Canada will only impoverish itself trying to copy the protectionist strategies of economic giants. The Sovereign AI Compute Strategy’s original 2024 commitment was $2 billion over five years, less than a sixth of Meta’s single Sturgeon County project. The sophisticated sovereigntist reply is that Canada need not match the hyperscalers. It needs only a sovereign niche for government administration itself and a national champion or two. Ottawa has already run a version of this experiment. The Sovereign AI Compute Strategy’s $240 million agreement with Cohere, a Canadian AI firm, was meant to anchor domestic capacity—and Cohere promptly contracted the U.S. firm CoreWeave to run its Ontario data centre. Balsillie calls the arrangement “an own goal of just the most severe proportions.” The criticism misfires even on sovereigntist accounting, since the firm doing the work, owning the resulting IP, and capturing the commercial upside is Canadian-founded, Canadian-headquartered, and still run by its founders. But the episode is more revealing than either side allows. Cohere contracted American infrastructure because that’s how a competitive AI firm operates: the industry’s supply chains run through global markets for compute, capital, and talent, and any Canadian firm good enough to matter will be embedded in them. The own goal isn’t the CoreWeave contract, in other words, but the belief that public money could conjure a self-contained national stack in the first place. The deeper objection to champion-building is a matter of principle, and it has three parts. First, engineering a champion requires Ottawa to know, years ahead of the industry itself, which layer of the AI economy will capture durable value—foundation models, inference, applications, or the infrastructure beneath them. Nobody knows this. The Cohere agreement is really a bet that foundation models are that layer, made by the institution least equipped to judge and least able to exit when it’s wrong. Second, a firm sustained by policy becomes a firm shaped by policy. Its most important relationship shifts from the customers who need its products to the government that needs it to exist, which dulls the commercial discipline that made it worth backing. The corrosion runs both ways: once Ottawa’s credibility is invested in its champion, it acquires an interest in tilting regulation and procurement toward its ward—including against better tools that Canadian businesses might otherwise adopt. Sovereignty economics ends by taxing Canadian AI use to subsidize Canadian AI ownership. Third, the strategy fails on its own terms. A subscale champion dependent on foreign compute, foreign capital markets, and periodic public replenishment delivers dependence with a maple leaf on it. Real leverage in the AI economy comes from being needed, and a firm the state must keep alive is, by definition, a firm the world doesn’t yet need. None of this is a criticism of Cohere. A Canadian firm raising private capital and competing for global enterprise customers is exactly what should happen. The objection begins where government policy treats its survival as a national interest. A strong company can survive a subsidy. The selection process that produced it can’t survive scrutiny, and it won’t always land on a strong company. Others, by the way, have run the full-scale experiment. Quaero, a 2005 Franco-German project to build a state-backed rival to Google, collapsed shortly after Germany withdrew, having produced nothing resembling a competitor. GAIA-X, launched in 2019 to build sovereign European cloud infrastructure, has, after six years, produced a platform most European organizations still don’t use. Both consumed real public money and delivered little beyond the appearance of action. Diffusion, not ownership If ownership is the wrong organizing principle, what is the right one? The political scientist Jeffrey Ding’s recent book, Technology and the Rise of Great Powers, offers an answer. Examining every industrial revolution from British mechanization to American electrification, Ding finds that for general-purpose technologies, long-run economic and even geopolitical advantage depends less on who invents or owns them than on who diffuses them most effectively across the economy. The countries that pulled ahead were rarely the ones that monopolized frontier innovation. They were the ones who adopted it most broadly. The insight generalizes into a first principle that sovereignty economics gets backwards: countries do not become rich chiefly by owning things. They become rich when their people are productive. A Canadian nurse, farmer, or machinist working with the world’s best AI tools is better off than one confined to a domestically owned but inferior alternative. Forty-one million Canadians equipped with frontier technology will generate more prosperity than a handful of protected national champions ever could. Sovereignty economics is, at bottom, mercantilism for the intangible age. It measures national success by what the state can secure for domestic firms rather than by what its people can do. Mercantilism was wrong to treat gold as the measure of national wealth, and sovereignty economics makes the same mistake with intellectual property. The economics underpin the history. In a landmark study, Yale economist William Nordhaus estimated that innovators capture only about 2.2 percent of the total social value their innovations create. The rest flows to users through lower prices, better products, and higher productivity. On Nordhaus’s arithmetic, sovereignty economics amounts to spending billions competing for the 2 percent while neglecting the 98. The sovereigntists’ reply is that the intangibles economy has changed this arithmetic—that AI’s winner-take-all rents make evidence like Nordhaus’s obsolete. The early returns point the other way: open-weight models now trail the frontier by an average of four months, inference prices for a given level of capability have been falling by an order of magnitude or more each year, and the frontier labs themselves fund their model-building from investor capital rather than operating profit. If durable rents are accruing anywhere in the AI stack, they sit in the compute infrastructure that Canada can host and in the application layer, where value follows adoption. The layer sovereignty economics wants Ottawa to own is the one the market is busiest commoditizing. Canadian research tells a similar story. A comprehensive Statistics Canada analysis covering all industries from 2000 to 2014 found that multinationals operating in Canada were about 23 percent more productive than comparable domestic firms, and the advantage held equally for Canadian-owned and foreign-owned multinationals. Global integration and competition drove the gains, not who held the equity. Nor is foreign ownership capturing the rents sovereigntists fear. Foreign-controlled enterprises owned 16 percent of the assets in Canada’s information and cultural industries in 2024, according to the latest enterprise ownership data. That share is up from 6.5 percent in 2010, yet foreign-controlled enterprises collected just 1.7 percent of the sector’s operating profits. Domestic firms kept the remaining 98.3 percent. Across the whole corporate economy, the foreign-controlled share of assets has been declining for 15 years. To this, add one of the oldest insights in economics: small and medium-sized countries prosper by specializing and trading, not by pursuing self-sufficiency across the technology stack. Integration was central to how Canada became one of the richest countries on earth. That meant selling our strengths—energy, agriculture, mining, advanced-manufacturing niches, financial services, and select technologies—into markets far larger than our own. The sovereigntist ambition to replicate a mini-American innovation stack abandons the formula that helped build Canadian prosperity. Where Canada genuinely lags is adoption. Although business AI use has tripled in two years (from 6.1 percent of firms in 2024 to 19.2 percent in 2026), two in five Canadian businesses still say AI is not relevant to them at all. Adoption in construction, wholesale trade, and agriculture remains in the single digits. Canada’s AI adoption rate lags behind other advanced economies, including the Nordic small open economies most comparable to ours. The pattern long predates AI: Canadian firms have trailed their American peers in technology investment per worker for decades. As University of Toronto economist Kevin Bryan argues, diffusion is not a lever governments pull directly. It is endogenous to competition policy and business culture, areas where Canada is also at a disadvantage. Weak competitive pressure means firms do not have to adopt to survive, and the same weak pressure partly explains why so few Canadian firms grow into the kind of company that could own a category outright. Interprovincial trade barriers and restrictions on foreign investment both shrink the market that would reward either. Canada’s challenge, in short, is less a shortage of domestic champions than a shortage of domestic adopters. The sovereignty camp expends enormous energy on the question of how to build the next Shopify. The more consequential question is how to get more Canadian businesses using the best technologies available, whoever makes them. None of this counsels fatalism about Canadian technology. Canada’s actual champions—Shopify, Constellation Software, OpenText, CGI, and, in its era, Research In Motion—share a common origin: each was built by competing in open global markets for customers, capital, and talent. Not one was engineered by an ownership preference. RIM’s history makes the point most sharply. The most successful technology firm Canada has produced was a triumph of precisely the global integration its co-founder now warns against, and its decline came from losing a product competition rather than from foreign capture. That’s how champions are made and unmade: in markets. The task for policy is to make Canada a place where more firms can run that race, and then to let them run it. Even when the sovereigntists brush against the adoption problem, they answer it with an ownership program. CCI argues that governments should use procurement to hand Canadian scale-ups their first reference customer. The symptom is real—Canadian governments are slow, risk-averse technology adopters—but the remedy transforms an adoption problem into a nationality preference. The defence experience described below shows where that leads: higher costs, slower delivery, and firms that learn to invest in lobbying rather than product. Procurement reform should make government a faster, smarter buyer of the best available technology. Where Canadian firms are genuinely competitive, they will win their share on merit. What Canada should do instead A diffusion-first agenda for Canada has two planks, and neither requires Canadian ownership. The first is physical: be the place where compute wants to live. Canada cannot out-subsidize the hyperscalers, but it has what their capital is searching for—land, cold, abundant energy, and, where governments choose it, regulatory speed. That is a genuine comparative advantage rather than a manufactured one. Alberta’s terms show how to exercise it without a subsidy auction: projects bring their own power and make no claim on ratepayers. Emerging First Nations data-centre partnerships suggest how broadly the benefits could be shared. The opening is widening because data-centre politics are turning hostile in the United States. Maine’s legislature passed a statewide moratorium this spring, stopped only by a governor’s veto; New York’s has passed one of its own; Seattle, Minneapolis, Denver, and Baltimore have enacted municipal pauses; and prominent voices in Congress want a national freeze. Capital that is unwelcome there must go somewhere. And here the sovereigntists’ own anxieties argue for openness. If the fear is that American firms, or the American government, could one day restrict Canadian access to models and compute, then every gigawatt of the continent’s AI infrastructure that physically operates on Canadian soil, under Canadian law and on Canadian grids, is worth more as insurance than any subsidized domestic clone. The real hedge against dependence is making Canada home to so much of that infrastructure that restricting it becomes unthinkable. The second plank is where the larger prize lies: be the economy that uses AI best—through sharper competition policy, an internal market without provincial barriers, predictable regulation, competitive taxes, and a marketplace that rewards risk-taking. Get the environment right, and both diffusion and the next Shopify follow from it. Hosting the world’s compute will not by itself make Canadian firms use it. Hosting and adoption proceed on separate tracks, and both need attention. What they share is a premise: Canada’s opportunity lies in being indispensable to how AI is built and deployed—hosting it, using it, and owning pieces of it where Canadian firms earn that place on merit. From tech to defence: the spread of a dogma The sovereignty narrative’s real danger is how readily it generalizes. Once a government accepts that ownership matters more than efficiency, economic nationalism becomes a universal justification for bad policy, and the pattern long predates the current movement. Canadian defence procurement has for decades been shaped by Industrial and Technological Benefits requirements, effectively a domestic-content mandate obligating winning bidders to undertake business activity in Canada equal to the value of the contract. The federal auditor general’s December 2024 review found that negotiating and managing these obligations delayed equipment deliveries and raised contractors’ costs, and that the responsible department could not demonstrate the policy was meeting its objectives. The result is a procurement system that leaves the military under-equipped while driving up costs for taxpayers, undermining the security that sovereignty rhetoric promises to protect. Similar logic shapes R&D tax policy. CCI has documented that the SR&ED credit, worth around $4 billion a year, flows substantially to Canadian subsidiaries of large foreign multinationals. Ottawa’s recent reforms sharpened the program’s focus on Canadian-controlled firms. The reform is a testament to the tech lobby’s organizational effectiveness, whatever one makes of the premise. But a Canadian subsidiary of a foreign multinational that hires Canadian engineers and pays Canadian salaries is not extracting Canadian wealth. The R&D activity, and the human capital it builds, stays in Canada regardless of who owns the resulting patent. Cohere is itself the proof. Its founders came up through Google’s AI research operation in Toronto, and one of them co-authored the transformer paper that underpins modern AI while still a University of Toronto student. The branch plant seeded the national champion. Shield of the rent-seeker None of this is new. Canada ran the sovereignty experiment half a century ago. Fears that American business was swallowing the country whole produced the Foreign Investment Review Agency, Petro-Canada, and a suite of ownership preferences. Economic nationalists like Mel Watkins and Kari Levitt urged the program, warning of a “branch plant economy” in language that, with artificial intelligence swapped in for manufacturing, is virtually interchangeable with today’s. The economist Harry Johnson called it nationalism “diverting Canada into a narrow and garbage-cluttered cul-de-sac.” The experiment produced distortions, delivered little, and was quietly euthanized in 1985. The industries change. The premise— that Canadians should own more of the economy—never does. It persists because it pays. In Ottawa, “sovereignty” has become a convenient shield for corporate rent-seeking. When an industry cannot justify subsidies on efficiency or commercial grounds, it drapes its business model in the national flag, recasting a standard commercial challenge as an existential crisis of national independence. Sovereignty economics also never states the condition under which it would be judged to have failed. If the public subsidies produce no champion, the answer will be that the fund was too small. If the champion falters, the answer will be insufficient procurement preference. A program that treats every failure as evidence for its own expansion is a doctrine rather than a policy. Diffusionism names its metric in advance—adoption rates, business investment per worker, productivity growth—and can be fired if it misses them. And it carries an opportunity cost the debate rarely acknowledges. Canada faces a genuine productivity emergency: business investment has lagged, and investment per worker has fallen behind international peers. Public and private capital are scarce. Every dollar Ottawa devotes to owning a sliver of the technology frontier—sovereign compute funds, subsidized champions, protected procurement—is a dollar unavailable for the diffusion agenda where the evidence says the returns actually lie. A pro-digital agenda for Canada does not have to be a sovereignty agenda. If it becomes one, the cost will not only be the money wasted. It will be the adoption forgone. Meta did not need to be coerced into Sturgeon County. Alberta built a workable regulatory framework and got out of the way. That is the model. Sovereignty built on excluding foreign capital is isolationism by another name. The diffusionist’s burden It would be comforting to believe that ideas this flawed will collapse under their own weight. They will not, because their advocates are organized. Over two decades, Balsillie has built a formidable institutional ecosystem—the Centre for International Governance Innovation, the Balsillie School of International Affairs, the Centre for Digital Rights, the Council of Canadian Innovators, the Canadian Shield Institute, to name a few—dedicated in whole or in part to advancing this worldview. Its adherents publish, convene, testify, and brief. They have cultivated sympathetic officials and politicians across parties and departments. And they are winning: the language of Ottawa’s AI strategy is their language. Whatever one thinks of the doctrine, the discipline behind it deserves respect and a response. Their political moment, however, is borrowed. These ideas predate Donald Trump. Balsillie has been making this case for years, but it never gained real traction until Trump’s tariffs, taunts, and annexation talk made anxiety about the United States the organizing emotion of Canadian politics. Sovereignty economics has skillfully attached a decades-old policy program to a brand-new fear. The fear is understandable. But policies adopted in a defensive crouch tend to outlive the crisis that justified them. The Foreign Investment Review Agency (FIRA) was born of the Nixon-era shocks and lingered for more than a decade after the panic passed. The fear deserves a better answer than the one it’s getting. If the organizing anxiety is American leverage, the question is what actually generates counter-leverage, and the answer runs through hosting rather than owning. A Canada that operates a meaningful share of the continent’s compute under Canadian law has more standing in Washington than a Canada with a subsidized model shop. The sovereigntists have correctly identified the emotion of the moment and attached the wrong program to it. Diffusionists enter this contest at a structural disadvantage that the economist Mancur Olson would have recognized instantly. Sovereignty economics has concentrated beneficiaries—the firms that receive the subsidies, the preferences, the protected contracts—and concentrated beneficiaries organize. The beneficiaries of diffusion are everyone: millions of workers, consumers, and businesses, each gaining too little individually to lobby for it. That asymmetry, more than the strength of the arguments, explains why one side of this debate has an institutional ecosystem, and the other has scattered op-eds. The case for diffusion, therefore, has to be made deliberately: by analysts willing to say in public what they agree on in private; by provincial governments prepared to demonstrate the model, as Alberta has; by business organizations whose members’ productivity is the prize; and by a federal government willing to measure its AI strategy by adoption rather than ownership. The sovereigntists ask how to ensure Canadians own the future. The better question—the diffusionist question—is how to ensure Canadians benefit from it. Canada ran the ownership experiment once and spent a generation finding its way out of the cul-de-sac. The technologies are new. The choice is not. DeepDives is a bi-weekly essay series exploring key issues related to the economy. The goal is to provide Hub readers with original analysis of the economic trends and ideas that are shaping this high-stakes moment for Canadian productivity, prosperity, and economic well-being. It features the writing of leading academics, area experts, and policy practitioners. The DeepDives series is made possible thanks to the ongoing support of the Centre for Civic Engagement.