SR&ED in 2026: How Canadian SMBs Building AI and Software Can Recover Up to 35 Cents on Every R&D Dollar

Canada's Scientific Research and Experimental Development program distributed $4.5 billion in investment tax credits to more than 19,000 businesses in 2024-2025 — one of the most generous R&D incentive programs available to any OECD country's private sector. (CRA, Annual Program Statistics)

The sectors collecting the largest share are not biotech or manufacturing. Software and technology development represented 40.8% of all SR&ED investment tax credits allowed in 2023-2024 — the single largest category by dollar value, accounting for nearly 4 in 10 claims filed. (CRA, Annual Program Statistics)

For Canadian SMBs that build custom software, develop AI systems, or systematically solve technical problems without ready-made solutions, this is the most directly applicable federal funding program available — and Budget 2025 just made it significantly more accessible.

What Budget 2025 Changed — And Why It Matters Now

The 2025 federal budget introduced the largest SR&ED enhancements in more than a decade, effective for taxation years beginning on or after December 16, 2024:

(KPMG, Canada's SR&ED program enters a new era, February 2026) (MNP, Significant enhancement announced to the SR&ED program)

The doubling of the expenditure limit is the headline number. A Canadian-controlled private corporation (CCPC) spending $6 million on qualifying R&D in 2026 can now receive up to $2.1 million back in cash — not a deduction, not a deferral, an actual refund even if the company has no taxable income. Before Budget 2025, the same company's maximum refundable credit was $1.05 million.

The expanded taxable capital phase-out range also matters for scaling companies. Under the old thresholds, a company with $12 million in taxable paid-up capital was already partially phased out of the enhanced credit. Under the new thresholds, that company sits fully within the eligible range.

Restoration of capital expenditure eligibility — removed in 2014 — means that equipment and cloud computing infrastructure leased specifically for SR&ED work (including GPU instances for AI model training) is now eligible for both the income deduction and the ITC, for expenditures incurred after December 15, 2024. (Welch LLP, 2026 Changes in SR&ED: Largest Expansion in Decades)

What SR&ED Is — and What It Isn't

SR&ED is a federal tax incentive administered by the Canada Revenue Agency (CRA). It is filed with your T2 corporation income tax return — not a separate government grant application with a queue. The deadline is 12 months after your T2 filing due date, which works out to 18 months from your fiscal year end. For a December 31, 2025 fiscal year: the T2 is due June 30, 2026; the SR&ED claim deadline is June 30, 2027.

The program is not restricted to dedicated research labs. Of the 21,524 claims filed in 2023-2024, 91% were accepted as filed — a number that reflects the breadth of companies that legitimately qualify when claims are well-documented. (CRA, Annual Program Statistics) The program reaches pre-revenue startups, established software companies, and professional services firms with proprietary technical platforms.

Qualifying expenditures include:

• Salaries and wages of employees directly performing, supervising, or supporting SR&ED work — plus CPP, EI, and other payroll-related contributions
• Materials consumed or transformed in SR&ED activities
• SR&ED contracts paid to third parties performing eligible work in Canada
• Capital expenditures for machinery and equipment used directly in SR&ED (restored for property acquired after December 15, 2024)

The Three Tests: What Actually Qualifies

CRA applies a three-part eligibility test to every claimed project. All three conditions must be satisfied. (CRA, What work is eligible — SR&ED tax incentives)

1. Technological uncertainty. A genuine technical problem must exist for which no known solution exists — not a business problem, and not a technical task that a competent professional could resolve with standard techniques. The uncertainty must exist at the start of the work, not be asserted after the fact.

2. Scientific or technological content. The work must involve forming a hypothesis, conducting experiments or analysis, and drawing technical conclusions. This describes what the work actually is, not a documentation layer imposed on top of it.

3. Systematic investigation. The investigation must be conducted in a structured, documented way that allows an independent reviewer to reconstruct the technical narrative.

For software companies, eligible work typically includes:

• Developing novel algorithms for problems where existing approaches produce insufficient accuracy, performance, or scalability
• Designing new system architectures to meet performance requirements that standard patterns cannot achieve
• Resolving technical conflicts between components where no established integration pattern exists
• Experimenting with custom ML model architectures or training methodologies to achieve results not achievable with standard methods

Work that does not qualify:

• Applying a known algorithm to a new dataset without modifications to the algorithm
• Building standard applications — CRUD systems, dashboards, reporting layers — using established frameworks
• Writing integrations with external APIs using vendor-provided documentation
• Prompt engineering against foundation models — the CRA treats this as optimization of a known tool, not investigation under technological uncertainty

AI and Machine Learning: The Fastest-Growing Category — and the Most Scrutinized

AI and ML-related SR&ED claims grew by approximately 40% year-over-year since 2023, reflecting the surge in corporate AI development. (Chrono Innovation, AI & ML SR&ED Claims: CRA's Tighter 2026 Criteria) The CRA responded by sharpening its review criteria for this category; internal guidance circulated in early 2026 provides more specific direction to science advisors evaluating machine learning claims.

The distinction CRA draws is between advancing AI technology and applying existing AI technology. Fine-tuning a commercial foundation model with standard vendor methods, training a classifier on domain-specific data using a published architecture, deploying an AI product via a vendor API — none of this is SR&ED-eligible on its own. The tools, methods, and expected outcome classes are all known; the uncertainty is commercial, not technological. (Leyton, SR&ED AI Eligibility: Where CRA Draws the Line)

Eligible AI work exists where the technical approach itself is uncertain. If your team is building a custom model architecture because standard approaches produce unacceptable results in your domain — and you can document your baselines, your experiments, and why known methods failed — that work qualifies. If you are solving a problem where the field has no established solution for your specific constraints, and you are systematically investigating alternatives, that qualifies. (G6 Consulting, The Impact of AI on SR&ED Eligibility in High-Tech Sectors)

The practical implication for Canadian SMBs building AI products: the custom development work your engineers do when commercial AI tools don't fit your use case is precisely the category the program was designed for. Work that simply assembles known tools is not.

Provincial Top-Ups: Stacking Credits Is Legal

Federal SR&ED investment tax credits can be combined with provincial equivalents. The combined recovery rates vary significantly by province:

Ontario's Innovation Tax Credit (OITC) is a 20% refundable credit available to CCPCs with prior-year taxable income below $500,000 and prior-year taxable paid-up capital below $25 million, on up to $3 million in eligible expenditures. (Ontario, Ontario Innovation Tax Credit) An Ontario CCPC with $500,000 in qualifying SR&ED expenditures can recover approximately $275,000 in combined federal and provincial credits — paid out as cash or applied against taxes owing.

Quebec's refundable rates on salary costs are among the highest in any jurisdiction globally, making it one of the world's most attractive locations for R&D-heavy software development when provincial and federal credits are stacked. British Columbia's 10% provincial credit applies broadly to eligible R&D expenditures and combines cleanly with the federal rate.

Documentation: The Difference Between a Claim and a Cash Return

Of the 21,524 SR&ED claims filed in 2023-2024, 91% were accepted as filed; 6% required modifications; 3% were denied. (CRA, Annual Program Statistics) Modifications and denials are overwhelmingly documentation failures — work that genuinely qualified was disallowed because the technical narrative could not be reconstructed from available records.

Contemporaneous documentation is non-negotiable. Reconstructing project history 12 months after the fact from memory and commit logs alone is the most common pattern in claims reduced on audit. What CRA expects:

• Issue tracker entries or engineering notes describing the technical problem, what was tried, and what the outcome was
• Commit messages and code branches that correspond to distinct experimental approaches rather than feature names
• Test results showing performance against baselines — including failed experiments, which are evidence of genuine investigation
• Technical meeting notes or asynchronous discussion threads that show the hypothesis-and-investigation cycle in real time
• Time-tracking records linking engineer hours to specific SR&ED projects

The standard CRA applies is whether a science advisor reviewing your documentation 18 months after the work was completed could reconstruct the technical hypothesis, the experiments conducted, and the conclusions reached. If yes, the claim is supportable. If not, the work may have qualified but the claim is at risk.

CRA processing has also improved under the Budget 2025 changes: 92.6% of accepted-as-filed claims are now processed within 60 days, and 94.1% of refundable claims selected for review are processed within 180 days. (CRA, Annual Program Statistics) The program is materially faster than it was — for claims with complete, organized documentation.

The Filing Process in Three Steps

1. Identify qualifying projects. Review your development work from the fiscal year against the three-part test. For software and AI companies, this is typically a structured technical interview: engineers describe what they built, what technical challenges they encountered, and how they resolved them — and an SR&ED advisor identifies where technological uncertainty existed and documents the technical narrative.

2. Calculate qualifying expenditures. Allowable expenditures include direct SR&ED labour plus payroll levies, materials consumed, SR&ED contracts, and qualified capital expenditures (since December 2024). A proxy cost method allows a flat percentage of SR&ED salary costs to be claimed for overhead without detailed overhead tracking — a significant simplification for SMBs.

3. File Form T661 with the T2 return. T661 captures the technical description of each qualifying project, the expenditure breakdown, and the ITC calculation. The technical narrative component is the most important section — precise language aligned with the CRA's eligibility framework is required. Most companies work with an SR&ED specialist or a tax advisor with deep SR&ED experience for this step.

Sources

• CRA. *Annual Program Statistics — Scientific Research and Experimental Development.* canada.ca
• CRA. *SR&ED Tax Incentive Program — Main Page.* canada.ca
• CRA. *What work is eligible — SR&ED tax incentives.* canada.ca
• KPMG. *Canada's SR&ED program enters a new era: Key impacts for businesses.* February 2026. kpmg.com
• MNP. *Significant enhancement announced to the SR&ED program.* mnp.ca
• Welch LLP. *2026 Changes in SR&ED: Largest Expansion in Decades.* welchllp.com
• Chrono Innovation. *AI & ML SR&ED Claims: CRA's Tighter 2026 Criteria.* chronoinnovation.com
• Ontario. *Ontario Innovation Tax Credit.* ontario.ca
• Leyton. *SR&ED AI Eligibility: Where CRA Draws the Line.* leyton.com
• G6 Consulting. *The Impact of AI on SR&ED Eligibility in High-Tech Sectors.* g6consulting.ca

Cloud Forces works with Canadian SMBs to identify SR&ED-qualifying work in custom application and AI development projects, document technical narratives to CRA standards, and structure development workflows that make future claims defensible. If your team is solving real technical problems that don't have ready-made solutions, a portion of that spending is likely recoverable. Learn about our AI Advisory and custom application development services or book a no-obligation SR&ED eligibility consultation.