Canada’s AI sovereignty debate is really about the supply chain

Canada’s AI sovereignty debate is shifting away from the tidy idea of building everything at home. Cohere and Microsoft Canada argue that the more useful question is which parts of the international AI supply chain the country cannot afford to lose control over.

Cohere Watch analysis

What happened

At an innovation event in Canada, Microsoft Canada national technology officer John Weigelt described the dependencies behind a single AI response. Helium used to cool chips may be drilled in Alberta and Saskatchewan but refined in Texas. The chips can be made on machinery from the Netherlands using German and French technology, according to Weigelt’s account in Digital Journal.

Michael Pelosi, Cohere’s country manager for Canada, said sovereignty should focus less on whether every component is Canadian-owned and more on control and independence. He pointed to hospitals and national defence as areas where infrastructure needs to be local or otherwise under Canadian control.

Pelosi also argued that waiting for a fully Canadian system, including domestically made chips and models, would leave organisations waiting a very long time. Cohere uses its own large language models alongside models from other companies, according to the report.

Why it matters

AI sovereignty is often presented as a flag-waving contest, with each country invited to build an entire miniature version of Silicon Valley in the shed. The supply chain is less flattering and more useful. Critical systems depend on minerals, chipmaking equipment, specialist manufacturing and computing infrastructure spread across several countries.

That matters for organisations deciding where to deploy AI. A system hosted in Canada may still rely on overseas chips, cloud services, components or technical expertise. The practical issue is not simply where the model sits, but which dependencies could be interrupted, restricted or controlled by someone else.

The report also places the debate in the context of adoption. Weigelt cited Canadian business AI use at 19.2 per cent in the latest Statistics Canada survey wave. Pelosi said regulated sectors such as banking, energy and natural resources have good reason to move carefully because mistakes can be expensive and difficult to unwind.

Our read

This is a useful reframing from Cohere and Microsoft Canada: sovereignty is not an all-or-nothing bid for a sealed national AI box. It is a risk-management question about the specific links a country must be able to protect, replace or govern.

The argument is persuasive as a starting point, but it is not a finished policy. The supplied report does not identify a definitive list of Canadian dependencies, quantify the cost of replacing them or show how organisations should weigh local control against capability and price. Those are the details that turn a good panel question into a serious national strategy.

For buyers and policymakers, the sensible next step is to map the supply chain behind important AI services rather than accepting “sovereign” as a comforting label. A maple leaf on the server rack is not, by itself, a continuity plan.

What to watch

• Whether Canada publishes a concrete list of AI infrastructure it considers strategically critical.
• How Cohere and other providers define control over models, data, chips and cloud operations.
• Whether regulated industries require domestic hosting, guaranteed access or broader supplier diversity.
• Whether AI adoption grows faster once organisations have clearer answers about resilience and accountability.

Discussion spark: Should governments prioritise domestic ownership of AI infrastructure, or concentrate on controlling the few supply-chain dependencies that could genuinely threaten essential services?

not affiliated with or endorsed by Cohere