On Thursday morning in Toronto, Prime Minister Mark Carney and Evan Solomon, Canada’s Minister of Artificial Intelligence and Digital Innovation, launched AI for All, the federal government’s new national AI strategy. The headline ambitions are not shy: $200 billion in GDP gains, up to 250,000 new AI-related jobs by 2031, and an increase in business AI adoption from roughly 12 percent today to 60 percent by 2034. Carney framed the stakes as a question of distribution — whether AI “will improve the lives of all Canadians or benefit only a few.”
But the most revealing number in the strategy document isn’t any of those. It’s 44th. That’s where Canada ranked out of 47 countries on AI training and literacy in the KPMG–University of Melbourne global trust study the strategy itself cites — alongside a 42nd-of-47 ranking on trust in AI systems. The country that helped invent modern AI, home to Hinton, Bengio, and Sutton, has a population in which fewer than a quarter of people report ever having received any AI training, and which is split nearly evenly on whether AI is good for society (34 percent) or harmful (36 percent). Half of Canadians regard AI as a threat to humanity.
That juxtaposition — frontier research capability sitting on top of bottom-decile public trust — is the actual problem AI for All exists to solve. The strategy’s own causal chain is stated plainly: for Canadians to benefit from AI they must use it, to use it they must trust it, and trust requires safeguards and sovereign control. Everything else in the document — six pillars, five priority sectors, roughly three billion dollars in identifiable new commitments — hangs off that single psychological wager. For anyone leading an organization in Canada, the useful exercise is not to absorb the strategy’s optimism or its critics’ dismissals, but to separate three things the document deliberately blends: what is genuinely new money, what is a re-announcement of existing programs, and what is a target with no mechanism attached.
The Adoption Math the Strategy Doesn’t Show
Start with the number that anchors the whole plan: 60 percent business adoption by 2034. The strategy reports, via Statistics Canada, that only 12 percent of Canadian businesses used AI to produce goods or services between mid-2024 and mid-2025, rising to 14.5 percent planning to do so by mid-2026. Among small and medium-sized enterprises — which the document notes represent 99 percent of Canadian businesses and employ 14.3 million workers — adoption sits around 8 percent.
Now look at the comparison set the strategy itself supplies. The Nordic leaders it cites are at 29 to 42 percent SME adoption. Germany is at 26 percent. France is at 18 percent. In other words, the strategy’s target doesn’t just close Canada’s gap with the leaders — it requires Canada to leapfrog every comparator country named in the document by a wide margin, quintupling adoption in eight years. The document offers no methodology for how 60 percent was derived, no intermediate milestones between the 14.5 percent planned for mid-2026 and the 2034 endpoint, and no published model of which interventions contribute which share of the gain. It may be achievable; diffusion curves for general-purpose technologies do steepen. But as written, 60 percent is an aspiration with a date attached, not a forecast.
The GDP claim deserves the same scrutiny, and here the strategy quietly undercuts itself. The headline commits to “unlocking a 3% increase in GDP, representing nearly $200 billion in GDP gains” from labour productivity. Two paragraphs into its own economic overview, though, the document notes that broader AI adoption is already forecast to generate a 0.3 to 1.1 percent annual labour productivity increase “even before the Strategy’s impacts,” and cites estimates that generative AI alone could add $187 billion annually to the Canadian economy by 2030. Read those numbers together and an uncomfortable question surfaces: how much of the $200 billion target is incremental to what existing forecasts already project would happen without the strategy? The document never says. For a plan whose central premise is that government action is needed to convert research leadership into economic outcomes, the absence of a counterfactual baseline is a real gap — the target may be largely a relabeling of growth that was coming anyway.
There is one finding buried in Pillar 3 that does more analytical work than any of the targets, and it’s worth quoting because it reframes the entire adoption problem. According to Statistics Canada, 78 percent of non-adopting firms report that they simply do not see how AI benefits the goods or services they provide. The strategy’s gloss on this is sharp and, I think, correct: “That is not resistance; it is a translation problem.” Canadian businesses aren’t refusing AI on principle — nearly half of SME owners have experimented with generative AI tools — they’re waiting for sector-specific applications with proven value and a tolerable adoption path. The gap between experimentation and deployment is where the Canadian problem actually lives, and it’s a more tractable problem than “Canadians don’t trust AI” — but also one that literacy courses and awareness campaigns, which absorb a meaningful share of the new spending, address only obliquely.
Where the Money Actually Goes
Tally the dollar figures the strategy document itself attaches to named initiatives and you get a picture rather different from the announcement-day framing. The identifiable new and expanded commitments:
| Initiative | Commitment | What it funds |
|---|---|---|
| Compute Access Fund expansion | $700M | Affordable sovereign compute for SMEs |
| Canadian Tech Growth Fund | $500M | Scale-up capital, including federal equity stakes in AI firms |
| Regional Artificial Intelligence Initiative expansion | $500M | AI adoption and commercialization via Regional Development Agencies |
| AI Missions Program (first mission: health) | $200M | Targeted public-good AI deployments, starting with health outcomes |
| National AI Institute commercialization programs | $130M | Founders-in-Residence and research-to-company pipelines |
| Health Sector Data Space (with CIHI) | $100M | Linked, standardized health datasets for trials and research |
| VITAL expansion | $100M | Pan-Canadian clinical data platform, five additional provinces |
| Canadian AI Safety Institute expansion | $50M | Risk tracking, technical research, transparent model evaluations |
| Creative Technology Program | $50M | Canadian creators using AI on their own terms |
| CanCode | $30M | K-12 digital-skills training via not-for-profits |
| Total | ~$2.86B |
Alongside that sits a second, larger category: money the strategy leverages but did not create. The $500 million LIFT program is Business Development Bank of Canada capital — BDC is a Crown corporation that lends from its own balance sheet, so this is not a fiscal appropriation in the same sense. The strategy also folds in over $2 billion in existing AI compute investments including the AI Compute Challenge, $1.75 billion in Budget 2025 venture-capital commitments, the $1.7 billion Budget 2025 talent attraction strategy, $159 million in Budget 2025 IP programs, and $50 million in Budget 2025 Job Bank modernization. None of this is illegitimate — coherently pointing existing instruments at a single objective has value — but a reader skimming announcement coverage could easily come away believing the strategy represents five-plus billion in new spending when the genuinely new envelope is closer to half that, and some of it (the partnership-dependent compute build-out, discussed below) is contingent on private capital that hasn’t closed.
Two structural moves matter more than any single dollar figure. The first is the equity-stakes provision. The Tech Growth Fund will “enable the federal government, at times, to take equity stakes in the most promising Canadian AI firms,” explicitly modeled on what the strategy describes as the approach of “France, Japan, and now the United States.” Wherever one lands on the wisdom of state shareholding in technology companies, this is a genuine departure for Canadian industrial policy — the federal government as an investor in, not merely a funder of, its AI champions, with the recently announced Sovereign Wealth Fund available “where appropriate” to go further. The stated logic is anchoring: nearly 70 percent of Canadian-led startups end up headquartered outside the country, and the strategy treats ownership as the lever that keeps the next Cohere from scaling under another flag.
The second is procurement. The strategy commits to establishing the federal government as a “strategic anchor customer” under the Buy Canadian policy, accelerating AI procurement through the Office of Digital Transformation, and recruiting technical talent into government through a new Prime Minister’s Innovation Fellows Program. For Canadian AI vendors, this is arguably the most commercially significant line in the document — government demand is the one adoption lever Ottawa controls directly, and it requires no behavioural change from skeptical SMEs. It is also the hardest to evaluate from the outside, because the strategy attaches no procurement volume targets, no timelines, and no dollar figure to the anchor-customer commitment.
The Sovereignty Gap, in Gigawatts
The sovereignty pillar is where the strategy is most candid about Canada’s position and where the arithmetic is most sobering. The document states the dependencies plainly: Canadian researchers train models on foreign cloud platforms, Canadian companies store sensitive data in foreign jurisdictions, government operations run on infrastructure Canada doesn’t own, and GPU fabrication “sits almost entirely offshore.” Its answer is a “build-partner-buy” doctrine — build domestically where possible, partner with trusted allies or buy from the market where not.
Here is the math the strategy provides, assembled in one place. Canada will require an estimated 5.5 gigawatts of AI compute for its commercial players by 2030 — the document attributes this to unnamed analysis and concedes estimates vary. Against that requirement, the strategy’s sovereign build-out consists of partnerships, described as “being finalized,” that have proposed 850 megawatts of capacity by 2030, with scaling potential to 2.3 gigawatts and “corresponding investments in the tens of billions.” Even the optimistic upper bound covers well under half the projected national requirement, and the near-term committed figure covers about 15 percent of it. The strategy acknowledges this directly: much of the 5.5 GW “will be delivered through large-scale operations by hyperscalers,” whose foreign investment Canada “will continue to welcome.” Sovereign compute, in other words, is planned as a strategic minority share — an alternative that exists, under Canadian law and control, for the workloads that need it — not a wholesale replacement of foreign infrastructure. That is probably the realistic posture for a country of Canada’s size. It is worth being clear-eyed that this is what “sovereign AI foundation” means in practice, because announcement language like “world-leading public supercomputer” (committed by 2031, with no named site, vendor, or budget line in the strategy) can suggest something more total.
The physical advantages underwriting the build are real and specific. More than 83 percent of Canada’s grid runs on renewable and low-emission sources, the cold climate cuts data-centre cooling costs that competitors “are spending billions to replicate artificially,” and the companion National Electricity Strategy contemplates doubling electricity infrastructure by 2050. The binding constraint, which the strategy admits, is that available power for large-scale AI build-out is limited today — the clean-grid advantage is a 2030s asset being marketed against a 2026 compute crunch.
On chips, the honest summary is that the strategy contains one concrete action — building on the spin-off of the National Research Council’s Photonics Fabrication Centre — against a dependency it describes as nearly total. Photonics is a genuine Canadian niche (the document elsewhere names Ranovus among the hardware value chain), but no reader should mistake a photonics fab spin-off for a GPU sovereignty plan. There isn’t one, because for a market of Canada’s size there realistically can’t be; the strategy’s implicit answer is the alliance structure instead.
That alliance structure is the strategy’s most original contribution. The Sovereign Technology Alliance, launched with Germany in February 2026, is framed as a coalition of “middle powers” pooling research, talent, compute, and procurement to offer “a credible alternative to the dominant market actors” — paired with a commitment to lead multi-stakeholder investment in open-source AI as a hedge against proprietary lock-in. The strategy counts 11 of Canada’s 20 recent economic and defence partnerships as explicitly advancing AI cooperation; the Prime Minister’s announcement release counts 12 and lists twelve countries, a small discrepancy between the two official documents that mostly illustrates how fast the partnership ledger is moving. Either way, the strategic read is the same: Canada is positioning itself as the convening node for countries that want AI capability without full dependence on American or Chinese platforms. Whether procurement pooling among middle powers can actually generate hyperscaler-competitive economics is the open question the document doesn’t engage — alliance procurement has a long history in defence, where it works, slowly, and at famous cost premiums.
Legislation Promised, Legislation Pending
The trust pillar — the foundation of the strategy’s own causal chain — is also its least specified. The commitments are framed in future tense throughout: Canada will modernize consumer privacy legislation to enshrine a fundamental right to privacy, will introduce online safety laws, will provide legal tools against deepfakes, will protect elections from AI-enabled interference, will create a Trusted AI Certification program. No bill numbers, no parliamentary timelines, no draft frameworks accompany any of these. The only enacted-or-introduced legislation the announcement materials point to is the Protecting Victims Act, introduced in December 2025, targeting non-consensual sexual deepfakes and intimate-image distribution.
The history here warrants the skepticism. Canada’s previous attempt at AI legislation — the Artificial Intelligence and Data Act, bundled into Bill C-27 along with the privacy reform this strategy now re-promises — spent two and a half years in committee and died on the order paper when Parliament was prorogued in January 2025. The new strategy conspicuously does not revive AIDA or propose a successor horizontal AI law; instead it disaggregates the agenda into privacy modernization, online safety, and sectoral measures, while routing AI-specific oversight through the Canadian AI Safety Institute’s expanded evaluation mandate and a voluntary certification mark. That’s a defensible regulatory philosophy — several jurisdictions are retreating from omnibus AI acts — but it means the “rules and safeguards that build and maintain trust” remain, as of launch day, almost entirely prospective. The strategy frames its flexibility as a virtue, committing to “dynamism” because AI “moves faster than any single strategy can anticipate.” The same dynamism clause that lets the government adapt also lets every unfunded, unlegislated commitment quietly slide.
What the trust pillar does fund concretely is the $50 million Canadian AI Safety Institute expansion — risk tracking, technical research, and transparent model evaluations — plus renewed funding for the Standards Council of Canada’s AI program and work on content watermarking. Modest, but real, and the evaluation capacity in particular compounds: a government that can independently test frontier models negotiates with their vendors from a different position than one that can’t.
Reading It as an Operator
Strip away the vision language and the strategy resolves into a set of signals that are actionable for anyone making technology decisions in a Canadian organization.
The money is flowing toward five named sectors — health and life sciences, energy and natural resources, transportation, agriculture, and manufacturing and robotics — with health unmistakably first among equals: the $200 million inaugural mission, both $100 million data-infrastructure investments, and the strategy’s most developed case studies (the CHARTWatch early-warning system at St. Michael’s Hospital, which a Canadian Medical Association Journal study associated with a 26 percent drop in unexpected ward deaths, and the VITAL platform already running AI applications across multiple provinces with more than 80 companies building on health data). If your organization touches Canadian health data, the Health Sector Data Space and VITAL expansion are the items to watch, because they determine whether the decades of siloed clinical data the strategy calls “a strategic national asset” become genuinely accessible — and on what governance terms.
For SMEs, the practical instruments are the Compute Access Fund’s additional $700 million in subsidized sovereign compute, BDC’s LIFT financing for AI tooling, the Regional AI Initiative’s $500 million through the development agencies, and a promised AI readiness self-assessment tool. The honest caveat is that program details, eligibility thresholds, and application mechanics for most of these don’t yet exist publicly; the strategy is the announcement, not the program guide. The same applies to the workforce commitments — up to 90,000 placements is a figure assembled from existing program envelopes (Student Work Placement, Canada Summer Jobs, Skills for Success, Mitacs), and “up to” is doing measurable work in that sentence.
For AI vendors and scale-ups, the two lines worth building a thesis around are anchor procurement and the equity-capable Growth Fund. A federal government that has committed, in writing, to buying Canadian AI at scale and co-investing in its champions is a different counterparty than the one Canadian founders have spent two decades routing around on their way to Delaware incorporations. Whether it follows through is unknowable today — but the commitment is now public, specific in direction if not in volume, and attributable, which gives industry something to hold Ottawa to.
The strategy’s deepest bet remains the one in its title. AI for All wagers that the binding constraint on Canadian AI prosperity is not capital, not talent, not even compute — it is the trust of a population ranked 42nd of 47 countries on exactly that measure, and the literacy of a workforce ranked 44th. Most national AI strategies are industrial policy wearing a public-interest preamble. This one inverts the proportions: the industrial machinery is real but mostly re-aimed rather than new, while the genuinely novel commitments — universal literacy training, AI agents for every post-secondary student, library-based community programs, pro-worker adoption framing negotiated with unions — target the psychology of adoption rather than its inputs. Nobody has demonstrated that a government can manufacture public trust in a technology on an eight-year schedule. By 2034, Canada will have run the experiment.