---
title: "Canada’s AI Strategy Focuses on the Wrong Firms"
summary: |-
  Canada may reach its AI adoption target and still miss its promised productivity gains. Delivering those gains will require deep adoption by large firms, supported by regulatory certainty and stronger tax treatment for AI deployment.
date: "2026-08-19"
issues: ["Artificial Intelligence", "Technology Diffusion", "National Competitiveness", "Enterprise Policy"]
authors: ["Lawrence Zhang"]
content_type: "Blogs"
canonical_url: "https://itif.org/publications/2026/08/19/canadas-ai-strategy-focuses-on-wrong-firms/"
---

# Canada’s AI Strategy Focuses on the Wrong Firms

For nearly a decade, Canadian AI policy focused on research: fund the institutes, keep the talent, and celebrate that Canadian researchers helped invent machine learning. The [2017 Pan-Canadian AI Strategy](https://web.archive.org/web/20170707054646/https:/www.cifar.ca/assets/pan-canadian-artificial-intelligence-strategy-overview/) barely mentioned adoption. Canada’s new AI strategy, [AI for All](https://ised-isde.canada.ca/site/ised/en/canadas-national-artificial-intelligence-strategy-ai-all), corrects that mistake by putting adoption at the centre and promising nearly $200 billion in productivity gains. But it hands the job of producing those gains to small- and medium-sized firms through financing, advisory support, and readiness programs, when a payoff of that size can only come from deep adoption by Canada’s largest firms.

The strategy aims to lift business AI adoption from 12 percent to 60 percent by 2034 and unlock nearly $200 billion in GDP gains through strengthened labour productivity. The two are related, but hitting the adoption target does not deliver the $200 billion on its own. That depends on which firms move and how deeply they rebuild their operations around the technology, yet the strategy's instruments are not built to reach the firms capable of generating gains at that scale.

In practice, its adoption instruments are small-business programs. The Business Development Bank of Canada (BDC) will lend against AI-related purchases through its Lead with Innovation and Focus on Technology financing stream. Regional Development Agencies will distribute $500 million in grants and advisory support to firms in their regions. The Compute Access Fund, which subsidizes access to computing capacity, is being expanded to take smaller applicants. The rest of the adoption agenda consists of readiness assessments, advisory services, and student placements. Taken together, these measures are designed primarily to encourage AI adoption among firms not yet using it. However, none of these measures is specifically designed to help firms already using AI undertake the kind of operational rebuild that produces measurable output gains. These programs may raise the number of businesses using AI, but they are unlikely to produce anything close to $200 billion in additional output.

The payoff from adoption favours firms already equipped to use AI well. Productivity gains depend on complementary capabilities, such as R&D, cloud infrastructure, and data analytics, and a [Statistics Canada study](https://utppublishing.com/doi/full/10.3138/cpp.2025-065) linking the digital-technology survey to tax records found no measurable productivity effect from AI once it controlled for those capabilities. Productivity gains ultimately come from the interaction between AI tools and all the capabilities a firm has already built around them.

Most small firms have not built any of these capabilities. A clinic with 12 employees, a single-location retailer, and an independent plumber generally have no R&D spending, no operational data, and no data analyst on the payroll. The exceptions—small advanced manufacturers, software firms, and other technical firms with digital capacity to build on—matter, but they represent a small share of Canadian businesses.

Firms cannot hire a third of a data scientist, and data infrastructure does not come in half sizes. These costs arrive in full regardless of how much activity a firm has to spread them across. A national retailer spreads these costs across thousands of transactions a day. A small firm spreads them across eight employees and a relatively small revenue base. [BDC's survey of 1,247 firms](https://www.bdc.ca/globalassets/digizuite/51604-artificial-intelligence-imperative-canada-entrepreneurs.pdf) found that 25 percent of micro-businesses (1 to 4 employees) using AI reported reduced costs, compared with 41 percent of larger SMEs (100 or more employees). The survey attributed that gap to budgets and economies of scale. Firms facing that arithmetic tend not to make the investment.

This is what makes the 60 percent target misleading. Getting 60 percent of Canadian businesses to use AI by 2034 means getting small firms to use AI, since SMEs with 1 to 99 employees [account for 98.2 percent](https://ised-isde.canada.ca/site/sme-research-statistics/en/key-small-business-statistics/key-small-business-statistics-2025#t3) of all Canadian employer businesses. Canada could therefore reach 60 percent and still miss the $200 billion because the target measures breadth while productivity depends on depth.

Most of the $200 billion has to come from somewhere else. Large firms are [roughly twice as productive per hour](https://macdonaldlaurier.ca/mli-files/pdf/Nov2021_Big_is_beautiful_Atkinson_PAPER_FWeb.pdf) as small ones, can make large process-oriented capital expenditures, and run complex systems where AI compounds across multiple functions. Banks, railways, mines, and national retailers are where adoption compounds. A major railway that improves demand forecasting also changes how it schedules maintenance, positions equipment, and routes traffic, and each of those feeds back into the others. The same investment touches several parts of the operation at once, a scale of integration a 12-person firm cannot match.

The answer is not to hand large firms the same subsidies. Large firms have ample capital and technical capacity, and financing programs solve problems they don’t really have. Their barriers are regulatory uncertainty and the arithmetic of the investment itself: Deep adoption means spending heavily upfront to rebuild systems that currently work, enduring years of disruption during implementation, and waiting for the payoff to show up somewhere downstream. The strategy largely ignores both.

Canada’s [largest firms operate](https://ised-isde.canada.ca/site/sme-research-statistics/en/key-small-business-statistics/key-small-business-statistics-2025#t3) in heavily regulated sectors such as finance, telecommunications, energy, and health. They can afford to adopt but face uncertainty over which data they may use, how regulators will treat automated decisions, and who bears liability when systems fail. That uncertainty produces pilots instead of operational deployment. Clarifying the rules would likely do more for serious adoption than another financing program.

Firms also undertake hard organizational change when competition forces them to, not when they are paid to. Earlier productivity waves came from firms reorganizing operations to survive. U.S. retailers, for example, rebuilt their logistics under [Walmart-scale cost pressure](https://www.mckinsey.com/~/media/McKinsey/Featured%20Insights/Employment%20and%20Growth/Whatever%20happened%20to%20the%20new%20economy/MGI_Whatever_happened_to_the_new_economy_perspective.pdf#page=28) in the 1990s; they did not do it because government made the equipment cheaper. Yet much of the Canadian economy operates behind foreign-ownership restrictions, interprovincial fragmentation, and sheltered domestic markets that dull that pressure. Paying firms to adopt is a weak substitute for exposing them to conditions that make adoption necessary.

The one instrument in the government’s strategy best suited to large firms is the [Productivity Super-Deduction](https://budget.canada.ca/2025/report-rapport/chap1-en.html#a5), which improves the return on investment rather than firms’ ability to finance it. Making it a centrepiece of the adoption agenda, rather than a single bullet buried in the middle, means extending that treatment explicitly to AI deployment costs: connecting new AI tools to a firm's existing systems, retiring or merging redundant software as processes are rebuilt around those tools, and developing the software layer that sits on top of the hardware. The existing categories were not written with AI deployment in mind. It also means giving firms a multiyear commitment rather than a phase-out schedule, because a carrier or a bank planning a five-year systems replacement needs certainty that the treatment will still apply when the spending lands.

Ottawa may well hit 60 percent. The financing and advisory programs could succeed in getting small firms to start using AI tools, and there are enough small firms in Canada that the number would move. Whether the $200 billion in productivity gains follows depends on whether the largest firms rebuild their operations around the technology, and the parts of the strategy aimed at that consist of one tax measure and a set of unresolved regulatory questions.

---
*Source: Information Technology & Innovation Foundation (ITIF)*
*URL: https://itif.org/publications/2026/08/19/canadas-ai-strategy-focuses-on-wrong-firms/*