Canada Can’t Subsidize Its Way to AI Adoption

This summer, Prime Minister Mark Carney launched AI for All, Canada’s new national AI strategy, and set a hard target to lift the share of Canadian businesses using AI from about 12 percent today to 60 percent by 2034. The goal is right, but the plan to reach it focuses on the wrong problem.

AI for All treats adoption as a cost problem. Its main instruments lower the price of going digital, including LIFT, a $500 million financing program through the Business Development Bank of Canada; a separate $500 million Regional AI Initiative run through the development agencies; and public funding for computing infrastructure and scale-up capital for AI companies. But cost is not what holds most firms back from adopting AI. By Statistics Canada’s own count, more than three-quarters of businesses that do not plan to adopt AI say it is simply not relevant to what they make or do. The problem is not that they are waiting for a cheaper tool; it is that many do not understand what AI could do for them.

Canada has misdiagnosed the adoption problem: The primary barrier is not cost but firms’ difficulty identifying valuable use cases. Policy should focus less on subsidizing purchases and more on reducing information and implementation barriers.

The countries that have adopted AI fastest have worked directly on the demand side. Singapore is the clearest case. It pairs a grant that covers up to half the cost of pre-approved AI tools with something even more important: The government publishes tailored digital roadmaps for 22 industries, outlining which tools firms should adopt at each stage of growth, and offers free advice through SME centres, where business owners can consult with advisers who match their firms with specific, vetted products. The state did the unglamorous work of turning “AI exists” into “here is the tool for your sector, checked and half-paid.” Among SMEs, AI adoption tripled in just a year, from 4.2 percent to 14.5 percent.

By comparison, the United Kingdom offered almost the same subsidy through Help to Grow: Digital—half off approved software—but skipped industry-specific roadmaps, practical demonstrations, and outreach. Firms got a flat catalogue and were left to work out the rest themselves. The scheme spent barely 7 percent of its nearly £300 million budget before it was scrapped.

South Korea shows what correcting course looks like. After years of pouring money into frontier capacity, it launched an AI voucher program in 2020 that matches small firms with vetted suppliers and helps fund project implementation, an effort to reach the businesses its earlier spending had passed over. Canada is making similar bets on sovereign compute today, and it should not wait as long to build the bridge to ordinary firms.

If the barrier is that firms cannot envision an AI use case, the most direct fix is to show it to them. Germany funds one of the most developed networks of demonstration centres in the world. At these centres, a company can watch AI run on real equipment and test it on its own data at no cost, with engineers on hand. The EU has built its Digital Innovation Hubs around the same “test before invest” idea. It is no cure-all: Despite Germany’s network, its 2025 AI adoption rate among enterprises was still 26 percent, well below the Nordic leaders. This serves as a reminder that demonstration is necessary but slow to produce results and worth little without the skills, data, and regulatory environment firms need to act on it.

The highest-performing countries in the EU illustrate another point: Part of what lifts them cannot be quickly replicated. Denmark, the EU’s 2025 AI adoption leader at 42 percent, more than double the EU average, benefits from two decades of shared public digital infrastructure, a near-universal national digital ID, and high trust in government. Together, these factors help make adoption a default rather than a decision. Finland, at 37.8 percent adoption, has emphasized literacy. Its free Elements of AI course reached 1 percent of the population and is now available across the EU in more than 20 languages.

Canada could emulate Finland’s educational approach relatively quickly, and its standing on AI literacy—44th out of 47 countries according to one global study—suggests it should. However, Canada cannot create a national digital ID or fix its low level of trust in AI systems (it ranks 42nd out of 47 in the same study) overnight. The lesson is not that Canada should spend more on AI, but rather that it should focus on policies that promote adoption.

Three changes would do most of the work.

1. Don’t repeat the $4 billion Canada Digital Adoption Program (CDAP)

Canada has already tried subsidizing digital adoption by paying consultants to produce standardized digital plans. But writing a plan is not the same as implementing one. Of the more than 15,000 businesses that received planning grants, only about 5,000 went on to obtain the follow-on loans needed to carry out their projects. This is the public-sector version of token maxxing: measuring AI commitment by the volume of plans and paperwork instead of real-world deployment and productivity gains. The government abruptly cancelled key elements of CDAP two years early, after spending less than one-fifth of its budget. Yet once again, LIFT and the Regional AI Initiative place considerable weight on lowering financial barriers to drive adoption.

If firms still struggle to identify worthwhile applications and implement them effectively, additional funding will just make an unfamiliar technology cheaper. The test for both programs should be how many firms actually put AI to use, not how much money moves.

2. Make the planned AI readiness tool more than a self-assessment

AI for All promises an online AI Literacy and Adoption Assessment tool to help small- and medium-sized businesses gauge their AI readiness, identify practical use cases, and connect with government programs. But a questionnaire that tells a firm it is “not ready” changes nothing. Instead, every assessment should end with a tailored set of recommended AI tools for that firm’s industry and size, along with information about vetted vendors and available financial support. The government should also open the program to the smallest firms. That is the key lesson from Singapore: Do not just diagnose the problem; connect firms directly to solutions.

3. Shift funding toward demonstration and literacy

Two of the strategy’s most essential priorities, demonstration and literacy, remain underfunded. Businesses are far more likely to adopt AI after seeing it work in a company like their own, so use funding from the Regional AI Initiative to help regional development agencies and colleges create vendor-neutral demonstration sites. Likewise, the government should invest in AI literacy by scaling proven training programs rather than building a new one. And because Canada’s innovation system is fragmented across jurisdictions, federal-provincial coordination should be part of the strategy itself, not something policymakers simply hope will happen.

AI for All gets the diagnosis half right: Canada is behind on AI adoption, not invention. But adoption is not something governments buy into existence. They build it by helping firms identify valuable use cases, reducing implementation uncertainty, and spreading practical knowledge across the economy. Until policy shifts from subsidizing purchases to solving those problems, Canada is unlikely to reach its 60 percent target. Ottawa should make that shift now by using the readiness tools to direct firms to appropriate tools and vendors, funding vendor-neutral demonstration sites, and scaling proven AI literacy training.