Canada's 'AI for All' national strategy

The strategy's most quotable line is Carney's promise to 'build the foundations of sovereign Canadian AI with compute, cloud, connectivity, data, and talent so Canadians can build and adopt AI on Canadian terms' ^[5]. The math behind that promise is harder. Three U.S. firms — Amazon, Microsoft and Google — hold 85% of Canada's public cloud market today ^[4], and the plan responds with a CA$700M Compute Access Fund expansion, up to CA$1B for public supercomputing, and a target of 850 megawatts of domestic compute capacity by 2030 ^[4]. That is a serious build, but it is a build on top of a hardware stack Canada does not make. As The Register notes, even with a domestic AI software push, 'Canada remains dependent on US chipmakers' ^[5]. The official strategy document itself frames sovereignty across compute, cloud, connectivity, data, and talent ^[2], conspicuously omitting silicon. Sovereignty over models and data residency is achievable on this budget. Sovereignty over the silicon, accelerators, and frontier-scale training runs that define the actual frontier is not — and the strategy quietly accepts that, recasting the goal as 'Canadian terms' rather than Canadian stack.

Buried in the headline numbers is a structural shift in Canadian industrial policy: the new CA$500M Canadian Tech Growth Fund is explicitly designed to 'provide flexible capital and investment support and allow Ottawa to take equity stakes in the most promising Canadian AI firms' ^[3]. This is a different posture than the grant-and-tax-credit machinery that funded the 2017 and 2022 strategies. It puts the federal government on the cap table of national champions like Cohere — directly responding to a long-standing complaint, voiced by Cohere in comments to BetaKit, that Canada has 'the talent, energy resources and research foundation' but that 'execution, access to capital, commercialization and stronger domestic demand will be critical' ^[6]. The equity mechanism also flips a known failure mode: AXL CEO Daniel Wigdor's frustration that the rest of the document treats Canada as a customer rather than an inventor — 'It drives me crazy when people talk about adoption in a country where we invented the god damned stuff' ^[6]— points at exactly the pattern the Tech Growth Fund is meant to break. Whether a state-as-shareholder approach can move faster than the IP-and-talent leakage it is trying to plug is the open question.

The plan opens by acknowledging a documented trust deficit among Canadians toward AI and promises 'new legislation, regulations and standards to protect your data, your privacy, your children' ^[4], plus a CA$50M expansion of the Canadian AI Safety Institute and anti-deepfake and surveillance-pricing rules. But the substance of those guardrails is deferred to future bills, which is why CTV characterizes the package as one that 'pledges thousands of jobs, lacks safety details' ^[12]. McGill researcher Helen Hayes captures the conceptual gap: 'Canada's national AI strategy begins by recognizing that Canadians don't trust AI... Yet despite identifying this deficit, the government has failed to actually ask how AI systems should be governed, designed, or constrained' ^[6]. Vass Bednar at the Canadian Shield Institute is blunter: 'This document seems to be mostly interested in adoption, and then worrying about the tough stuff later' ^[6]. The bet aligns Canada with a broader Western turn — the EU launched its own Technological Sovereignty Package the same week ^[9]— but inheriting the EU's posture also means inheriting its open question of who polices the platforms in the meantime.

The strategy commits to lifting business AI use from 12% to 60% by 2034 and to creating up to 250,000 adoption-driven jobs by 2031 ^[1]. What it does not do, according to organized labour, is legislate for the people on the receiving end of that adoption. Canadian Labour Congress president Bea Bruske: 'This feels like a strategy that businesses wanted to see. We are particularly concerned about who is regulating how employers make decisions about AI and workers' ^[8]. The environmental flank is similarly thin — the National Observer reports that the 49-page document 'dedicates a single box to environmental impact with no new actions on AI data centre water/energy use' ^[11], even as the 850 MW domestic-compute target implies a significant new data-centre footprint. Engadget pushes further upstream, questioning the productivity premise itself: the strategy 'largely ignores evidence that adopting AI technologies doesn't necessarily increase productivity' ^[10]. The sovereignty pact with Germany — explicit about wanting to 'reduce strategic technology dependencies' ^[7]— handles the geopolitics but does not answer the domestic compact. Stack those gaps together — no job-loss legislation, no new environmental rules, a contested productivity thesis, and an outward-facing alliance posture — and the 60%-by-2034 mandate looks less like a target and more like a bet that consent will catch up to the spend.