SR&ED Tax Credits and AI: What Changed in 2026 and Whether Your Business Qualifies

The Scientific Research and Experimental Development (SR&ED) program is Canada's largest federal tax incentive for business innovation, and it received its most significant enhancements in years through Budget 2025. For businesses working with AI, those changes matter — but so does understanding the fundamental distinction between what SR&ED is designed to fund and what it is not.

SR&ED is administered by the Canada Revenue Agency. The figures and thresholds cited in this article reflect the enhancements announced in Budget 2025. Tax credit rules can and do change with future federal budgets — verify current rates, expenditure limits, and eligibility criteria at canada.ca/sred before making any decisions based on this article.

What Changed in SR&ED for 2025–2026?

Budget 2025 introduced four meaningful changes that expand the program's reach for smaller businesses:

Doubled refundable credit cap. The maximum annual refundable SR&ED investment tax credit for Canadian-controlled private corporations (CCPCs) increased from approximately $1.05 million to $2.1 million. This is the portion of the credit that is refunded as cash even if a business has no tax owing — the most valuable form for small and growing businesses.

Higher expenditure limit for the 35% rate. The expenditure limit on which CCPCs can earn the enhanced 35% refundable rate increased from $3 million to $6 million in eligible annual expenditures. Above this threshold, the basic 15% non-refundable credit applies. This change lets businesses with higher R&D spending access the premium rate on a larger portion of their work.

Cloud computing costs are now eligible. Prior to this change, cloud computing costs — such as AWS, Azure, and Google Cloud infrastructure — were not considered eligible SR&ED expenditures. As of 2026, cloud computing resources used directly in support of qualifying SR&ED activities are now eligible for the investment tax credit. For businesses developing AI using cloud-based compute infrastructure, this is a significant shift.

Capital expenditures restored. The ability to claim certain capital expenditures — equipment purchased and used primarily for SR&ED work — was restored under the Budget 2025 reform package, reversing an earlier exclusion.

Does Using AI Tools Qualify for SR&ED?

No. Purchasing and implementing commercially available AI software — an AI voice agent, a CRM with AI features, a machine learning platform, or any off-the-shelf tool — does not qualify for SR&ED, regardless of how technically sophisticated those tools are or how much work is involved in configuring them.

This is the most common misunderstanding about SR&ED and AI. The program is designed to fund the advancement of technology itself — the act of pushing what's possible — not the adoption of technology that already exists. The CRA evaluates whether work meets three eligibility criteria, all of which must be satisfied:

• Technological advancement — the work must attempt to achieve something new in science or technology, not simply apply known techniques
• Technological uncertainty — it must not be known in advance whether the technical approach will work; if a competent professional in the field could predict the outcome, there is no uncertainty
• Systematic investigation — the work must be carried out through a systematic process of experiment or analysis, with documented hypotheses and results

Configuration, integration, and implementation of existing AI tools fails on the first two criteria. The technology is already established; the outcome of implementing it (even if it requires significant effort) is predictable by a competent professional.

What AI Activities Do Qualify for SR&ED?

The answer is: development work where genuine technological uncertainty exists. Some examples that may qualify, depending on specifics:

• Building a custom machine learning model for a domain where existing approaches have not been proven to work — where you don't know, before you test it, whether a particular architecture or training approach will produce useful results
• Research into applying a novel AI technique to a specific industry problem where prior published work doesn't establish whether the technique will generalize
• Developing new algorithms or data processing approaches that extend the capabilities of existing AI tools in non-obvious ways
• Work that involves systematic experimentation — testing different approaches, documenting failures and successes, and iterating toward an uncertain outcome

The key signal is uncertainty: if the question "will this work?" has a known answer in the field, SR&ED likely doesn't apply. If competent researchers genuinely don't know whether the technical approach will succeed until they try it, there may be a claim.

The SR&ED vs. IRAP AI Assist Distinction

SR&ED and NRC IRAP serve fundamentally different business profiles, and understanding the difference helps clarify which program — if either — is relevant to your situation:

• SR&ED is for businesses developing technology — writing new code, building new AI capabilities, resolving technological uncertainty
• NRC IRAP AI Assist is for businesses adopting AI — integrating existing AI tools and capabilities into their products, services, or operations

A software company building a custom AI model for a specific industry application should be looking at SR&ED. A service business implementing an AI phone system, booking automation, or AI-assisted scheduling should be looking at IRAP AI Assist. These are different programs for different types of work.

The Documentation Requirement That Most Businesses Underestimate

Businesses that successfully claim SR&ED don't just do the work — they document it contemporaneously. CRA requires that SR&ED claims be supported by records created at the time the work was done: project plans, hypotheses tested, results of experiments, decisions made based on outcomes, and time logs. Retroactive reconstruction of documentation is both difficult and a risk factor in CRA reviews.

If your business is doing work that might qualify — custom AI development, novel algorithm research, systematic experimentation in technical domains — the single most valuable thing you can do before you ever file a claim is start keeping records now. The claim is built from the documentation, not the other way around.

Most businesses that claim SR&ED successfully work with an experienced SR&ED consultant, who typically engages on contingency and will assess whether your work likely qualifies before committing to a claim. SR&ED claims require detailed technical justification for each eligible activity and are reviewed by CRA — inadequate documentation is the most common reason claims are reduced or denied.