Canada has announced two tools that could shape the country’s artificial intelligence (AI) future.
The Canada Strong Fund is an arm’s-length Crown corporation that will invest alongside private capital in strategic projects, while the Canadian sovereign AI compute strategy offers federal support to AI computing capacity for researchers, firms and innovators.
But tools are not doctrine. Without a clear mandate, AI sovereignty risks becoming either a slogan attached to a general investment fund or an unaffordable attempt to build everything at home.
Canada needs a narrower, more credible objective — trusted hybrid sovereignty. That means working with allied hyperscalers and private operators where they provide scale and capability, while ensuring Canada has legal control, audit rights, workload portability and emergency fallback capacity for designated critical public sector uses.
In other words, an exit plan that is workload-specific.
A Canadian hospital should be able to leave any single AI vendor without losing patient care continuity. A tax agency should be able to switch fraud detection systems if a vendor’s terms change.
This is not a strategic exit from allied AI co-operation, which Canada cannot afford and does not need. It is the ordinary discipline of not letting any single critical public function depend on a vendor it cannot leave or lose.
Services upon which Canadians depend should keep functioning when access, price, law or geopolitics turns against them.
Not a digital services tax by another name
Canada withdrew its digital services tax in 2025 to advance broader trade negotiations with the United States after criticism from tech giants and U.S. President Donald Trump. Any new mechanism that appears to target U.S. technology firms through the back door would likely face the same issues.
The right model is procurement resilience, not discriminatory taxation. Canada should not tax foreign cloud providers for being foreign. It should set operational requirements for designated critical public sector workloads.
What matters is what a provider can do, not where it is based. Any provider that meets Canada’s legal, security, audit and portability requirements should be able to win these contracts.
If the best available tool cannot meet them, an agency can still use it when public safety requires, but the contract should also include a way out. None of this is a tax on foreign firms. It is a condition on the contracts the government already signs.
That exit path should be the core of the policy.
A minimum sovereignty test and a portability reserve
For any AI workload designated as critical — in health systems, tax administration, benefits delivery, border operations, courts, cyber defence or emergency management — Ottawa should apply a minimum sovereignty test.
Can the workload be run or transferred under Canadian legal control? Can sensitive data be audited? Can the system survive a price shock, cyber incident, export delay or vendor withdrawal? Does the contract expand access for Canadian firms, researchers and public agencies? Does it reduce lock-in or deepen it?
Where a closed, non-portable AI system materially outperforms available alternatives, government should not pretend otherwise. A hospital triage system, a fraud detector or cyber defence tool should be judged on performance as well as continuity. But if public money buys non-portable dependence, public money should also buy a way out.
That mechanism should be called a portability reserve — a condition inside designated public procurements, not a tax on technology firms. A defined share of the contract value, calibrated through Treasury Board procurement review, should be set aside for migration tools, fallback capacity, multi-vendor alternatives, data export standards and emergency transition planning.
This is an insurance premium inside public procurement. The rule is simple: Use the best tool where necessary, but do not let public procurement create permanent strategic dependence.
The binding constraint is electricity
However, a right to exit is worth little if there is nowhere to go. This portability reserve assumes Canada can build capacity in case it needs to switch. Whether it can depends on power, on how government buys and on whether anyone is made to deliver.
Sovereign AI compute will not be decided by slogans or by how many AI chips a country can stockpile. It will be decided by power, cooling, transmission, interconnection queues and provincial co-operation.
Canada’s electricity system is federal in aspiration but provincial in operation. The federal government cannot announce sovereign data centres and expect power to appear. It needs federal-provincial compute-energy compacts.
Ontario, Quebec and Alberta each offer a different pilot logic — Ontario for demand growth and planning scale, Quebec for low-carbon power under tighter pricing discipline and Alberta for market-driven connection experiments.
The point is not to crown a winner. It is to recognize that sovereign AI capacity will be negotiated through provincial power systems, not announced from Ottawa. Each compact should be straightforward. Federal capital and long-term public sector demand in exchange for provincial power commitments, grid planning, local workforce benefits and guaranteed public-interest compute access.
Buy service-level agreements
Canada needs a trigger. Within 18 months, Innovation, Science and Economic Development Canada and the Treasury Board Secretariat should publish criteria for two things: whether Canada has secured substantive compute-energy compacts with the provinces and whether those compacts deliver real capacity for critical workloads.
The auditor general should review implementation and compliance after the fact.
If no credible compact exists by then, Canada should activate a minimum viable sovereignty plan — enough trusted capacity to keep essential services operating during a foreign price shock, cyber incident, export restriction or geopolitical disruption.
The deadline matters less than the discipline. Someone must be required to say whether Canada has secured real capacity or only announced ambition.
A sovereign cluster that is expensive, underused and disconnected from public services or Canadian firms is not sovereignty. It is a technology showcase.
Success is whether a hospital, tax agency, cyberdefence unit or emergency management office can keep operating under stress; whether Canadian small and medium enterprises and researchers can access compute without surrendering all bargaining power; whether sensitive public data can be processed under Canadian law; and whether dependency is measured, limited and made reversible.
Canada should not try to build the entire AI stack alone. But it should build the legal, financial and operational capacity to leave, switch, audit and endure.
AI sovereignty is not about building everything. It is about not being helpless.
Editor’s Note: The author’s submission on Canadian sovereign AI resilience was published by the Standing Senate Committee on Banking, Commerce and the Economy in May 2026.
