AI funding in Quebec in 2026: who pays for what, and how to build an application

Funding an artificial intelligence project in Quebec in 2026 is still possible, and often for a fraction of the sticker price. But the landscape has shifted. The Canada Digital Adoption Program (CDAP) is over. If a salesperson is still offering it to you, their information dates from 2023.

Six levers remain active in 2026 to fund an AI project in Quebec, and they form a coherent whole: a direct provincial grant (ESSOR stream 1, administered by Investissement Québec), two federal programs built around R&D (NRC IRAP and SR&ED), and two Quebec tax credits from the 2025-2026 budget: the newly created CRIC, and the reworked C3I. On top of that sits a lever almost everyone forgets: training, for which Services Québec can fund up to 75% of eligible costs.

One warning before going further. None of this support is guaranteed. The programs are discretionary, the budgets are limited and the criteria change from one fiscal year to the next. The reference points in this article were re-verified against the official pages as of August 2026, but every figure remains subject to confirmation of eligibility.

Here are the six main levers, from the most direct to the most specialized. The table below compares each program’s organisation, rate, ceiling, eligibility and deadline, as of the August 2026 figures: it answers most of the questions we get during project scoping.

ESSOR stream 1 is the natural front door for an AI project at a manufacturing or services SME. Two conditions shape eligibility: 250 employees or fewer, and revenue of at least $2.5M. The support is non-repayable: it is a grant, not a loan. Applications can be filed until March 31, 2027.

The stream breaks into three sub-streams, each with its own ceiling.

NRC IRAP (the National Research Council’s Industrial Research Assistance Program) funds 60% to 80% of direct R&D labour costs for SMEs with fewer than 500 employees. Its distinctive feature: everything starts with a conversation with an ITA, an industrial technology advisor. The ITA assesses the project, supports the company and recommends funding, or does not. Get in touch before you write the first line of your application.

A custom AI project qualifies more often than people think. Adapting an open language model to a company’s data, building reliable bill-of-materials extraction from technical drawings, hitting a target accuracy rate on documents written in Quebec French: all of these involve genuine technological uncertainty, which is what opens the door to federal funding.

SR&ED (scientific research and experimental development) rewards that same uncertainty, but after the fact, as a tax credit. For a Canadian-controlled private corporation the rate is 35%, refundable, and the reform enacted in March 2026 (Bill C-15) raised the ceiling on eligible expenditures from $3M to $6M. Large corporations are entitled to 15%, non-refundable. The crux is documentation: timesheets, hypotheses tested, experiments that failed. No contemporaneous record, no credit.

Quebec’s 2025-2026 budget created the CRIC, the tax credit for research, innovation and commercialization. It is refundable for companies of every size: 30% on the first $1M of eligible expenditures above the exclusion threshold, then 20% beyond that. It applies to tax years beginning after March 25, 2025.

The C3I, the tax credit for investment and innovation, targets assets instead: computer hardware and management software packages, for example a server meant to host your AI models on site. The rate depends on the region: 15% in Montreal and Quebec City, 20% or 25% in the regions depending on the economic vitality of the territory. The assets have to be acquired before January 1, 2030. Careful: the C3I and the CRIC cannot be combined on the same asset. The right choice depends on the nature of the expense and on your tax position.

Then there is training. Reimbursement from Services Québec generally runs up to 50% of eligible costs, boosted to 75% notably for training tied to the introduction of new technologies, and reaches 100% for literacy training. It is the most under-used lever, even though an AI project fails more often for lack of adoption than for technical reasons. Training employees on the tool, but also on the internal rules of use (our AI acceptable use policy template gives you the structure), is among the expenses that fund well.

Take a concrete case, drawn from a real engagement at a steel structure fabricator in Granby. The project: five sovereign AI assistants (meeting summaries, searchable procedures, bill-of-materials extraction, schedule conflict detection, drafting support), including data preparation. Invoice before support: between $45,000 and $85,000 depending on how wide the scope is drawn.

The funding package combines two programs. ESSOR stream 1C reimburses up to 50% of eligible implementation expenditures, to a maximum of $50,000. NRC IRAP covers 60% to 80% of the R&D labour on the part of the project that involves technological uncertainty, such as automatic bill-of-materials extraction from DXF drawings.

A good application tells the story of a transformation, not a purchase. The applications that stand out start from a precise business problem (quotes that take three weeks, a welder’s know-how heading into retirement), describe a phased approach with verifiable milestones, and present a realistic funding plan where the support requested complements a serious contribution of your own.

Compliance is an underused asset in an application. A project that plans a privacy impact assessment before deployment and documents its Law 25 compliance work shows the analyst a company that manages its risks.

Here are the mistakes we see most often.

Three moves are enough to get going. First check your eligibility for ESSOR stream 1: count your employees (250 or fewer) and your revenue ($2.5M or more). Then find the NRC IRAP office for your region and ask for a first conversation with an ITA. Finally, frame your project in two phases: a diagnostic fundable under stream 1B, then an implementation built on that diagnostic under stream 1C.

At Cogio we build this net-invoice calculation into the scoping stage: every proposal shows the gross cost, the plausible support and the estimated net, with the assumptions written out in black and white. Not because grants should decide the project, but because an executive who knows the real cost decides better. And if your project touches employee or client personal information, plan the compliance work alongside the funding: our 12-step Law 25 checklist is a good starting point, because the privacy impact assessment is prepared before deployment, not after.