Canada has named the AI dependency. Businesses still need a plan.

The national strategy says the quiet part plainly

The Government of Canada describes a structural exposure: Canadian researchers train on foreign cloud platforms, Canadian companies store sensitive data and intellectual property in foreign jurisdictions, and government operations rely on infrastructure Canada does not own. It warns that products built and governed elsewhere can be shaped by decisions, rules, and legal regimes beyond Canada’s direct control.

That diagnosis is broader than data-centre geography. The strategy’s sovereign-infrastructure pillar calls for domestic capacity operated under Canadian control and Canadian law, not a platform that a foreign government can unilaterally restrict or withdraw.

Four pillars, one infrastructure foundation

What “sovereign” means in the federal programs

The AI Sovereign Compute Infrastructure Program guide gives Canadian buyers a useful test. It defines sovereign infrastructure as Canadian-located and Canadian-governed, with data residency, operational control, and decision-making authority remaining in Canada. It expects safeguards against a foreign party unilaterally restricting use or access.

The major compute programs

Commercial sovereign capacity

The AI Compute Challenge supports net-new Canadian AI data-centre and compute capacity, with objectives that include domestic data processing, affordable capacity for Canadian users, Canadian champions, and sustainable infrastructure. The strategy describes up to C$700 million for this commercial-capacity stream. The published call is closed.

Public supercomputing

Canada’s public-infrastructure program seeks a Canadian-owned and located AI supercomputing system for researchers and industry. The 2026 application window closed June 1. This is national shared infrastructure—not a private on-premises system for an ordinary firm’s internal documents.

Compute access for SMEs and innovators

The strategy describes affordable compute support, including an additional C$700 million in sovereign compute for Canadian SMEs under AI for All, alongside earlier Compute Access Fund commitments. Published funding envelopes and implementation announcements are not the same as an open application window. Verify the current ISED program page before planning around support.

Large sovereign data centres

The government also solicited proposals for Canadian projects above 100 MW. The national strategy says proposed partnerships could deliver 850 MW by 2030, with potential scale to 2.3 GW, while estimating that Canada’s commercial AI players may require 5.5 GW by 2030. These are planning figures and proposals, not capacity a business can assume is available today.

Open weights are explicitly part of the answer

AI for All supports open-source AI as a way to reduce cost, increase flexibility, and enable on-premises deployment where privacy, security, or sensitive-data concerns are paramount. In procurement language, open weights add two practical powers: the model can execute inside a chosen boundary, and the organization can evaluate a replacement without rewriting its entire capability around one closed endpoint.

Not every downloadable model is open source. Some leading checkpoints carry custom licences with hosted-service or revenue conditions. Canada’s strategy creates the policy opening; organizations still need model, licence, security, and operations diligence.

The gap between national infrastructure and Monday morning

A supercomputer, a data-centre incentive, and a compute subsidy all matter. None decides whether a law firm can safely query client files next month, whether a manufacturer should index its maintenance corpus locally, or whether a software company can survive an API disruption. That is the organizational layer:

• Which workloads are business-critical?
• Which data can cross which boundary?
• Which model passes the actual task and licence?
• Should the destination be Canadian-hosted, on premises, or hybrid?
• Who administers it, what is retained, and how does it recover?
• Can the business change model or operator without starting again?

What Canadian organizations should do now

Frequently asked questions

Is Canada’s sovereign AI strategy only about training models?

No. The strategy covers compute, cloud, connectivity, data, research, commercialization, adoption, skills, safety, and infrastructure. Inference capacity and the ability of Canadian businesses to deploy and use AI are part of the stated problem.

Are current funding programs open?

The major SCIP and large-data-centre calls described here are closed, and earlier Compute Access Fund windows have also closed. New funding announcements or delivery mechanisms may follow. Verify directly with ISED before acting; this page is not a funding notice.

Does a Canadian cloud region satisfy the federal sovereignty definition?

Not by location alone. The federal program definition also emphasizes Canadian governance, operational control, decision authority, and safeguards against unilateral foreign restriction.

Does Canada need a Canadian foundation model?

Canadian model development matters, and Cohere provides a credible domestic enterprise option. Operational sovereignty can also come from running a commercially usable foreign-origin open-weight model under Canadian control, while documenting the upstream dependencies that remain.

What can a business do before national capacity comes online?

Build an organizational exit plan, qualify an open-weight fallback, move one sensitive or repeatable workload to an on-premises system, or scope Canadian-hosted capacity with explicit control and exit terms.

Official sources reviewed August 24, 2026: Canada’s National Artificial Intelligence Strategy: AI for All; SCIP program guide and sovereignty requirements; Canadian Sovereign AI Compute Strategy; AI Compute Challenge.