What is in AI for All, Canada's national AI strategy?

The federal government released its national AI strategy, AI for All, on June 4, 2026. Its diagnosis is blunt: Canada has strong researchers and more than 3,500 firms working on AI, but it lags in adoption and relies on foreign suppliers at several key layers. The strategy's central claim is that trust, opportunity and sovereignty together will get Canadians using AI on their own terms.

The document runs long, and much of the detail depends on later budgets and laws. This is a guide to what it says, what it commits to and where the first reactions landed.

The six pillars

Sovereignty: "build, partner, buy"

The strategy's most distinctive part describes AI as a value chain: energy, chips, compute, models and applications. It says Canada is present at each layer but has real dependencies at several. Chip fabrication for graphics processors sits almost entirely abroad. Sovereign compute is "nascent", and most of the country's data centre and cloud supply belongs to foreign owners.

The answer is not to build everything. Canada will build key capabilities where it can, partner with trusted allies and buy where that makes more sense. On compute, the numbers give a sense of scale:

• Analysis cited in the strategy suggests Canada's commercial players will need 5.5 gigawatts of AI compute by 2030. A gigawatt is a thousand megawatts.
• The strategy expects much of this to come from foreign hyperscalers, the giant cloud companies, and says those investments remain welcome where they benefit Canadians.
• Alongside them, partnerships still being finalized have proposed 850 megawatts of sovereign capacity by 2030, able to scale up to 2.3 gigawatts.
• An extra $700 million would make Canadian compute more affordable for small and medium-sized businesses.

Worked through, the proposed 850 megawatts is about 15% of that 5.5-gigawatt estimate, and 2.3 gigawatts about 42%. So even at full scale, the sovereign share would be a minority of expected commercial demand, which fits the strategy's own logic of mixing sovereign and foreign supply.

It also says this infrastructure must run under Canadian control and law, not on platforms a foreign government could restrict or withdraw. That sentence sets a higher bar than simple data location, and it will be a useful yardstick for future projects.

Taking stakes in companies

Pillar 5 marks a turn in federal policy. The strategy cites France, Japan and the United States as models, and says Ottawa should become a shareholder in its strongest technology companies. The $500 million Tech Growth Fund would provide growth capital and, at times, let Ottawa become a shareholder in promising AI firms. Government would also act as a "strategic anchor customer", meaning a large, dependable buyer whose contracts help a young company raise more money.

What people are saying

Reaction arrived within hours. Mila, the Quebec AI institute, "enthusiastically" welcomed the plan. Its president and CEO, Valérie Pisano, said it reflects values the institute has long championed, including attention to cultural and linguistic relevance, and urged rapid delivery.

The Walrus asked researchers, entrepreneurs and policy analysts to respond, and many were harsher:

• Blair Attard-Frost, a University of Alberta professor, says the regulatory framework is largely voluntary. The response cites an Ipsos finding that 67% of Canadians feel nervous about AI.
• Maroussia Lévesque, an AI governance researcher at Harvard Law School, praises the talent and alliance plans but says the compute vision "copies the American model". She suggests smaller, purpose-specific models on local data centres instead.
• Vass Bednar, of the Canadian Shield Institute, says a digital sovereignty approach would set rules on data, procurement and compute from the start, so that adoption builds Canadian capacity.
• Fenwick McKelvey sees a contradiction in sovereign compute centres filled with American hardware running American software.
• Danica Pawlick-Potts notes that much of the infrastructure will sit on treaty land, while Indigenous rights and sovereignty are absent from the sovereignty discussion.
• Sheldon Fernandez, the former CEO of DarwinAI, credits the diagnosis but finds the document light on specifics.

These first responses are one day old, and The Walrus chose its panel for critical reading, so they do not measure wider opinion. They do show where the arguments will go next: whether voluntary rules are enough, what "sovereign" requires and who benefits.

What does AI for All change for you right now?

Very little changes today. A strategy sets direction, and nothing in it changes which tools you can use or adds a new legal duty for your workplace. Several commitments depend on future budgets and legislation, and the compute partnerships are not final.

If you run a small business, watch for the expanded Compute Access Fund and adoption programs, since those are the parts aimed directly at you. If you work in health care, expect more activity, because the strategy names health as an early priority for adoption and data platforms. If you work in French, note the promise that AI used inside the federal government will work just as well in French as in English, a concrete test you will be able to check for yourself.

Most of all, the strategy gives you a vocabulary for asking questions. When a vendor or a politician says "sovereign", you can ask which layer of the chain they mean: the energy, the chips, the compute, the model or the application.