Stacking Quebec and Canadian support for an AI project: three standard packages in 2026

A well-built AI funding application looks like a cutting plan: every expense in the project is assigned to one program, and only one. That is the cardinal rule every body checks: the same hour of work, the same server, the same licence cannot be funded twice. Stacking is allowed; overlapping, never.

We set out each program in our overviews of the Quebec programs and the federal programs. This article tackles the next question: how to assemble them. Three packages cover the large majority of the projects we scope at Quebec SMEs.

Before assembling anything, three stacking rules to know by heart.

This is the standard package for an SME deploying its first AI agents with no research component: an internal assistant answering from company documents, quote automation, meeting summaries. The backbone is provincial, the top-up federal.

The sequence starts with ESSOR stream 1B (Investissement Québec): up to 50% of the digital diagnostic and implementation plan, capped at $20,000. That diagnostic is not a formality, it is what makes stream 1C credible: up to 50% of the implementation, capped at $50,000, for SMEs with 250 employees or fewer and revenue of at least $2.5M. ESSOR applications can be filed until March 31, 2027.

The federal top-up takes one of two forms. CED’s RAII offers an interest-free repayable contribution of up to 50% of the costs of an AI adoption project, on expenses distinct from the ESSOR ones. BDC’s LIFT loan ($25,000 to $5M, principal deferrable up to two years, preferential rate for Canadian solutions) funds the portion no grant covers. And while the project is being rolled out, Services Québec can fund up to 75% of employee training: the line item everyone forgets, and the one that determines whether the tool actually gets used.

As soon as a project involves genuine technological uncertainty (reliable data extraction from drawings, a target accuracy on documents written in Quebec French, adapting an open model to a trade vocabulary), a second floor opens up: research.

During the project, NRC IRAP funds part of the R&D labour for SMEs with 500 employees or fewer. Everything runs through the relationship with an industrial technology advisor: talk to them before you write the application, and before you commit the expenses. A Mitacs Accelerate internship can run alongside: $7,500 paid by the company for a $15,000 award, ideal for handing the measurement of the system’s accuracy to a graduate student.

After the project, the tax credits take over, and this is where the federal-provincial split matters. Federally, the SR&ED reformed in 2026 offers 35% refundable up to $6M of eligible expenditures for a CCPC. Provincially, the CRIC (which has replaced the R&D wage credit for tax years beginning after March 25, 2025) gives 30% on the first $1M above the exclusion threshold, then 20%. Both are claimed on the same work, each under its own calculation rules: your tax adviser coordinates the two filings, and your timesheets serve as evidence on both sides.

Documentary discipline is not negotiable: dated hypotheses, recorded trials, failures included. Without contemporaneous records, the claim melts away at the first audit.

Hosting AI on your own premises starts with a GPU server, and that server can be funded. Two routes exist, mutually exclusive for the same asset.

The simple route: the C3I, Quebec’s tax credit for investment and innovation, covers computer hardware (class 50) and management software packages acquired before January 1, 2030, above a $5,000 exclusion threshold for computer equipment. The rate depends on the region: 15% in Montreal and Quebec City, 20% or 25% depending on the economic vitality of the territory, fully refundable. A $25,000 server installed in a region at 25% yields $5,000 in credit on the portion above the threshold.

The R&D route: since the 2026 reform, SR&ED once again admits capital expenditures, with a 40% refundable credit on the eligible capital portion for a CCPC. If the server is used mainly for experimental development work, that route can beat the C3I. But you have to choose: the same asset cannot be claimed under both, and the C3I cannot be combined with the CRIC on the same asset either. This is a tax adviser’s judgment call, not a matter of principle.

Our practical rule of thumb: if the server is an operating asset that will run your local AI models for five years, the C3I is the natural route. If it is first and foremost an R&D test bench, have the SR&ED route costed before you buy.

All of them lived through, all of them avoidable.

Week 1: frame the project in phases (diagnostic, implementation, any R&D, training) and allocate every expense. Week 2: check ESSOR eligibility (250 employees or fewer, $2.5M in revenue) and call CED about the RAII. Week 3: if the project involves a technical challenge, approach an IRAP advisor and open your timesheets. Week 4: have your tax adviser arbitrate between the C3I, the CRIC and SR&ED capital before you buy any hardware.

At Cogio this cutting plan is part of scoping: every proposal breaks out the gross cost, the plausible support by program and the estimated net, with the assumptions written down. Our Funding page presents each program and its stacking rules, and if your project touches personal information, Law 25 compliance belongs in the same calendar: an application that plans its privacy impact assessment reassures the analyst as much as it does your privacy officer.