Canada's 40 Per Cent AI Refusal Is an Answer About Products, Not Offices

Statistics Canada asked Canadian businesses two questions in the same survey this spring and got two answers that the country now reads as a single story. Did you use artificial intelligence to produce goods or deliver services? 19.2 per cent said yes — up from 12.2 per cent a year earlier and 6.1 per cent in 2024, a 3.148-fold increase in two years. Is the use of AI relevant to your business? 40.0 per cent said no.

The first number is the one that travels. It is the number in the strategy document, the number in the minister’s release, the number that justifies CAD 2.3 billion of new spending. The second number is the one that sets the size of the market, and almost nobody quotes it — which is odd, because it is printed in the same release, in the same table, under the same heading.

Both are true. But only one of them is a fact about the thing most people think it is a fact about.

The question was about the product, not the office

The Canadian Survey on Business Conditions asks respondents whether AI was used to produce goods or deliver services. That wording is not incidental. It means the 40.0 per cent is a statement about what a business sells, and it is not a statement about how that business runs.

Consider the four businesses that most often answer this way. A machine shop quoting on press-brake work. A landscaping firm scheduling crews across a season. A dental practice handling recalls and insurance pre-authorisations. A trucking company dispatching loads and chasing proof of delivery. Every one of them is answering honestly when it says AI has nothing to do with its product, because it does not. And every one of them runs on quotes, invoices, schedules, dispatch notes and email — the text-in, text-out work that language models happen to be very good at.

There is no Canadian survey that asks the second question. So the most consequential number in this market is measured on the axis that generates the least commercial information. The Canada Enterprise AI Adoption Market report states the consequence plainly rather than burying it in a risk appendix:

If the question had been asked about back-office work, the ceiling would be higher and no one would know by how much.

That is not a quibble about survey design. It is the difference between a market with a 40.8-point addressable pool and a market with a materially larger one, and it is the single assumption on which this forecast would most likely be revised upward.

Two federal documents describe the same refusal, 1.58 times apart

Canada does not have one adoption statistic. It has several, and they do not agree with each other.

The strategy that sets the 60 per cent target says, in its own words, that 78 per cent of non-adopting firms report that they do not see how AI benefits the goods or services they provide, and calls this “a translation problem” rather than resistance. Apply that rate to the 80.8 per cent of businesses not using AI in 2026 and it implies 63.02 per cent of all Canadian businesses have concluded AI does not help what they sell. Apply it to the strategy’s own 2025 baseline — when 87.8 per cent were not adopting — and the implied figure is 68.48 per cent.

Statistics Canada’s print is 40.0 per cent. The two figures describe the same refusal, in the same country, in the same eighteen months, and they are 1.58 times apart.

None of this is a scandal. It is what happens when one country publishes adoption through four instruments and reports all of it under one word. But it has a price, and the price is quantified: the choice between the location count and the enterprise count is worth USD 0.78 billion of the 2026 market — a quarter of it — before any other assumption is touched, and the gap between the 8 per cent SME figure and the 19.9 per cent band print is 2.49 times, which no published table resolves.

The report uses the Statistics Canada series throughout, because it is the one with a published methodology and a three-year time series, and says why on the page. A reader who has seen the 8 per cent figure deserves to know that the two numbers in circulation differ by a factor of two and a half.

The ceiling is the market

Take the published numbers at face value and the market’s shape falls out arithmetically. Of the 80.8 per cent of businesses not using AI, 49.5 per cent have declared themselves a non-market. What remains addressable is 40.8 percentage points of Canadian business — and Canada’s national target is that pool’s exact complement.

Sixty per cent by 2034 requires every business that has not yet refused, converted inside eight years, with no allowance for the firms that will join the not-relevant group as they are founded. One business that moves from undecided to “not relevant” has to be offset by a business that was never in the undecided pool at all.

The rate is not the constraint. Forty point eight points over eight years is 5.1 points a year, against 6.1 points delivered in 2025 and 7.0 in 2026. In businesses, the target needs 558,960 net conversions, or 69,870 a year; Canada delivered roughly 95,900 in the single year to mid-2026 — 37.25 per cent above the average pace required. Momentum is already outrunning the target. The pool is what no policy addresses, and it is why the base case stops at 44 per cent in 2033: 24.8 of the 40.8 points available, 60.8 per cent of the pool, with the last sixteen points sitting behind a group of businesses that has already answered the question.

Most of the money is in deepening, not conversion

Here the headline and the revenue diverge in a way that should change how anyone plans against this market. The forecast runs two cohorts rather than one growth rate, because they move in opposite directions. The 263,040 businesses already using AI in 2026 deepen — their spend compounds as pilots become production systems. The 339,760 that adopt between 2027 and 2033 enter below today’s average, at 60 per cent of it, or USD 7,200 a year, because the marginal adopter is a smaller firm buying a narrower tool.

In 2033 those cohorts are worth USD 6.15 billion and USD 2.45 billion. 71.5 per cent of the market comes from accounts that already exist in 2026 spending more. A vendor building a Canadian plan on the tripling — on conversion — is planning against 28.5 per cent of the money.

The account list is short and it is knowable. 1.4 per cent of Canadian businesses — 19,180 accounts — hold 55.3 per cent of employment and adopt at 27.8 per cent. That is 5,331 buying accounts, of which roughly 1,610 engage an external consultant or vendor. The other 91.2 per cent of businesses hold 24.4 per cent of employment, adopt at about 18.9 per cent, train existing employees at 2.84 times lower rates and hire AI skills at 4.00 times lower rates. The capability gap between large and small Canadian firms is wider than the adoption gap between them — and capability is the one that compounds, because the firm that trained its people in 2026 is the firm that gets value from its next purchase.

What a Canadian business actually buys

The application mix is the clearest evidence that this is a distribution market, not a technology market. Among adopters, data analytics is used by 36.6 per cent, text analytics by 34.5 per cent, virtual agents and chat bots by 28.2 per cent. The categories that require process ownership sit far lower: machine learning at 18.2 per cent, decision-making systems at 13.7 per cent, deep learning at 13.1 per cent, machine or computer vision at 4.6 per cent.

Restated against all businesses rather than adopters — which is the version that matters for market size — data analytics reaches 7.03 per cent of Canadian businesses, one in 14.22, and large language models reach 4.76 per cent, one in 21.01. A technology described as general-purpose, and available free at the consumer tier, is used by roughly one Canadian employer business in twenty-one.

One caution on the year-on-year series, because it is easy to over-read. Three categories moved by more than ten points in a single year and moved in both directions: image and pattern recognition fell from 21.8 to 11.4 per cent and then recovered to 14.0; natural language processing fell 5.8 points and then rose 3.9; marketing automation rose 7.9 and then fell 3.3. Deployed technology does not stop being deployed because the year changed. That oscillation is respondent interpretation of overlapping category names, which is why the three-year direction is the only one worth trading on — and where it matters, it is clear: data analytics up 11.6 points, decision-making systems 2.25-fold, deep learning almost seven-fold in share.

The number that would change this forecast

Average annual AI spend per adopting business is not published anywhere in Canada. Statistics Canada measures whether businesses use AI, not what they pay for it; the TechStat program funded in Budget 2025 does not begin publishing business AI statistics until 2027. The model sets USD 12,000 and states the exposure rather than defending it.

A USD 4,000 revision moves the entire 2026 market by 33.3 per cent — more than the whole of the base case’s growth between 2026 and 2030. So the honest statement is the one the report makes: the direction of this market is measured and its level is estimated.

Which makes the falsifier refreshingly simple, and it is not about spending at all. If any future print puts the not-relevant share below 35 per cent, businesses are changing their answer — something the model currently assumes they do not do — and the ceiling moves. The fuller test arrives in 2027, when TechStat’s first business publications will show whether a question scoped to operations finds materially more use than a question scoped to output. If operational use exceeds product use by more than five points, this forecast is too low, and it is too low for a reason that, today, nobody can test.

What the number is, and what it is not

The market this describes is USD 3.16 billion in 2026 and USD 8.60 billion in 2033 at 15.4 per cent, inside a band of USD 5.99 billion to USD 11.83 billion — 67.9 per cent of the base case wide, which is wider than most readers expect for a market whose adoption curve is surveyed quarterly and cannot be argued with. The width is not in the curve. It is in the average spend, and in the fact that Canada publishes nothing that settles it.

The 40 per cent, meanwhile, will keep being quoted as an obstacle. It is better read as a boundary drawn by a question — and the most commercially useful thing a Canadian vendor can do with it is stop treating it as a refusal and start treating it as a measurement of product relevance the buyer has never been asked to revisit.

Sources: Statistics Canada, Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026 (11-621-M; Canadian Survey on Business Conditions, Table 33-10-1167-01) and business counts, June 2026; Innovation, Science and Economic Development Canada, Canada’s National Artificial Intelligence Strategy: AI for All (4 June 2026) and Key Small Business Statistics; the Prime Minister’s launch release; The Globe and Mail on the CAD 2.3 billion commitment; RSM Canada’s 2026 middle-market AI survey; PwC Canada’s 2026 Trust in AI report. No third-party market-research estimate is used.

Related reading

The full sizing, the three forecast paths, the cohort decomposition, the value chain, the regulatory picture and five falsifiers are in the Canada Enterprise AI Adoption Market Report, Size, Share and Forecast 2026–2033. The market announcement is in the companion press release, Canada Enterprise AI Spend Reaches USD 8.60 Billion by 2033.

• Global Artificial Intelligence Market — the worldwide envelope this Canadian spend sits inside.
• Global Artificial Intelligence (AI) as a Service Market — the delivery model through which most of this spend is bought.
• Global Artificial Intelligence (AI) in Insurance Market — Canada’s second-fastest-adopting sector, where large language model use runs at 38.8 per cent against a national 24.8 per cent among adopters.
• Global Hyper-Personalized Technology Market — the application layer above the recommendation systems 17.9 per cent of Canadian adopters already run.
• Canada EMR and Clinical AI Add-On Market — Canadian AI adoption in health care, where privacy concerns limit 26.4 per cent of businesses.
• India Artificial Intelligence (AI) Market — a peer-market view of AI at national scale.