Who argues what in Canada's sovereign AI debate?

Canada has been arguing about sovereign AI for two years, and the argument got louder on October 29, 2026, when Ottawa committed up to $1.6 billion for a stake in Cohere. BetaKit called it likely the largest federal equity investment ever in a single private tech company, and reported reactions "from across the spectrum".

The positions below are not two camps. Someone can favour Canadian compute and oppose the Cohere stake, or support the stake and doubt it delivers sovereignty. It helps to take them one at a time, in each side's own terms. Our separate article explains what the Cohere deal contains.

The government's case: capability in a less friendly world

Prime Minister Mark Carney places AI in a security argument. "Sovereignty now requires secure access to semiconductors, critical minerals, payment systems, clean energy technologies, vaccines, space-based communications, and artificial intelligence (AI)," he said in Montréal, on the day Ottawa released a new national security strategy. His approach has three steps: build at home where possible, partner with reliable allies, and gain scale through them.

AI Minister Evan Solomon defends the stake in practical terms. He told CTV it was done on favourable terms and "will give us a seat at the table", and that sensitive data will stay in Canada under Canadian law. Sean Mullin, a senior fellow at the Munk School, offers the strategic version: "Do we want to have critical parts of our future digital infrastructure run by AI models that are entirely outside of our control?" He does not claim Cohere will replace American giants, only that Canada and Europe need options.

The builders' case: we can supply it

Companies that stand to benefit make the commercial argument. Cohere CEO Aidan Gomez calls his firm the largest AI company outside the United States and China. Bell CEO Mirko Bibic told The Logic that too much of Canada's AI computing is "being served from other places, particularly the U.S.", and described his alliance's stack as "fully sovereign". The Logic notes the limit in the same report: Canadian firms control the facilities, network, software and models, but the chips come from Nvidia.

Telus makes a demand argument. CEO Darren Entwistle points to its Rimouski site, which sold out, as proof that Canadian innovators want this infrastructure. Both companies sell what they are describing, which is worth remembering, and so is the fact that they are putting up large sums of their own.

Researchers: welcome, with an urgent to-do list

Mila, the Quebec AI institute, welcomed the AI for All strategy. Its CEO Valérie Pisano stressed scientific excellence and "cultural and linguistic relevance", and said the vision must be deployed "with an acute sense of urgency". Mila also has a partnership with Cohere on Quebec French. Its interest is plain too: the strategy funds research chairs and programs it runs or relies on.

Other academics are far less warm, as the next sections show. Research voices in Canada do not speak with one mind.

Critics of the deal: is this government's job?

Policy analyst Charles Lammam makes the fiscal case in The Hub. By his math, $1.6 billion at a valuation near $36 billion buys about 4.5%, and "Sovereignty is a big word for a stake that small". He also notes the money is borrowed, and that private investors including Schwarz Group, Nvidia and an arm of Saudi Aramco were already in the round. The government has not disclosed its actual percentage, and Solomon says more stakes are coming. BetaKit reports that venture investors still want to know the milestones and what control the deal gives Ottawa.

A reply to this line is that Canada's strategy argues the opposite: that France, Japan and the United States already take such stakes, so standing aside has costs of its own. That is the disagreement in one sentence: industrial policy versus getting out of the way.

Critics of the definition: location is not control

The sharpest technical objection comes from Joshua van Es, in a BetaKit opinion. He argues Canadian policy rewards footprint, meaning buildings, hiring and local spending, rather than authority: who holds encryption keys and administrative access, and whose laws bind the operator. His example is Ottawa's earlier $240 million for Cohere, which supports a Cambridge, Ontario facility that the government's own statement says is "built and operated by CoreWeave", an American company.

Others make similar points. Guillaume Beaumier and Hubert Cadieux argue in Policy Options that data centres run by American firms stay under American jurisdiction through the Cloud Act, a law letting U.S. authorities compel those companies to hand over data they store. They want more money directed to public infrastructure for non-commercial uses. Keith Jansa, of the Digital Governance Council, says the real test is whether Canadian organizations can change providers "without becoming locked into systems they can't control".

Note that the government has moved on this point. AI for All says infrastructure must run under Canadian control and law, a stricter test than data location. Whether projects are held to it is the live question.

Critics of the model: why this kind of AI at all?

A third group questions the destination. In The Walrus, AI governance researcher Maroussia Lévesque says the sovereign compute plan "copies the American model" and proposes smaller models on local data centres. Fenwick McKelvey sees American hardware running American software with Canada supplying the energy. Danica Pawlick-Potts points out that much of this infrastructure would be built on treaty land, with Indigenous sovereignty missing from the sovereignty talk. University of Ottawa's Michael Geist warns of "digital self-sabotage" from policies that raise barriers to the adoption the strategy seeks, and his colleague Teresa Scassa says the promised "light, tight and right" regulation has so far reached only "light".

There is one more voice, from outside. RBC reports that in March 2026 the U.S. Trade Representative named Canada's sovereign computing initiative among trade barriers. So the policy has a diplomatic cost that its supporters and critics both need to count.

Where the sides agree

How can you tell whether an AI service is sovereign in the way you need?

Ask five questions in order. The first two are about place, and the last three about power.

No line of this table makes a service good or bad. It tells you what you are buying. A clinic bound by privacy law may care most about the operator. A startup may care most about the price and speed, and rightly choose a foreign cloud. Decide which risk you are guarding against first: foreign legal orders, supplier cut-offs, or simply losing your data when a vendor changes hands. Then ask the vendor that exact question, in writing.