Back to Blog
AI Adoption9 min read

SR&ED Tax Credits for Canadian SMBs: How the 2026 Reforms Change the Math on AI and Software Development

By Anton Kuznetsov

Most Canadian business owners who develop custom software or AI tools know the SR&ED program exists. Fewer have claimed it — and a significant number who have claimed it left money on the table by misunderstanding which work qualifies. The program just got meaningfully larger. Bill C-15 received Royal Assent on March 26, 2026, delivering the most significant expansion of the Scientific Research and Experimental Development (SR&ED) tax incentive in over a decade. The enhanced expenditure limit for Canadian-controlled private corporations (CCPCs) doubled from $3 million to $6 million, capital expenditures on R&D equipment returned to eligibility after a twelve-year absence, and the taxable capital phase-out thresholds were widened substantially. If your business builds custom applications, develops AI tools, or trains machine learning models, the post-reform SR&ED math is worth understanding before your next T2 filing.

Why SR&ED Matters More Now Than It Did Two Years Ago

SR&ED is Canada's largest single federal program supporting business R&D. It delivers investment tax credits (ITCs) directly against your taxes owing — and for CCPCs, the credits are refundable even when the company has no tax payable. That distinction matters for early-stage and growing SMBs: you do not need to be profitable to receive a cash refund.

The scale of the program gives context for how much opportunity exists at the individual claimant level. For the fiscal year ending March 31, 2026, the CRA processed 24,160 SR&ED claims and reported $4.9 billion in investment tax credits claimed — a 4.3% increase from the prior year and roughly 50% higher than the $3.2 billion total in 2022. Source: The State of SR&ED 2026 Of those processed claims, 90% were accepted as filed, 6% accepted after modifications, and only 4% denied. The program has a high acceptance rate for businesses that document their work properly.

The 2026 Reforms: What Changed and When

The changes enacted by Bill C-15 apply to tax years beginning after December 15, 2024. For most businesses with a December 31 fiscal year-end, the enhanced rules apply to your 2025 fiscal year. Three changes are particularly significant for technology and AI development businesses.

Doubled enhanced expenditure limit. CCPCs can now claim the enhanced 35% refundable ITC on up to $6 million in qualified SR&ED expenditures per year, up from the previous $3 million cap. At the maximum, this translates to $2.1 million in refundable credits annually — more than double the previous ceiling of $1.05 million. (BDO Canada, SR&ED Program Enhancements)

Capital expenditures restored. Capital expenditures on equipment used primarily for SR&ED — at least 90% of the time — were removed from SR&ED eligibility in 2014. Bill C-15 restored this eligibility. For businesses building AI inference infrastructure, GPU servers, or specialized testing hardware, this reopens a category of eligible spending that has been off-limits for over a decade. (Chrono Innovation, SR&ED Changes in 2026)

Widened phase-out thresholds. The taxable capital range over which CCPCs phase out of the enhanced 35% rate increased from $10 million–$50 million to $15 million–$75 million. Businesses that had grown beyond the previous thresholds — and dropped to the standard 15% non-refundable rate — now qualify again for the enhanced rate.

The ITC Rates

Two federal ITC rates apply depending on your corporation structure:

Entity typeFederal ITC rateRefundable?Enhanced expenditure cap
CCPC (Canadian-controlled private corporation)35%Yes$6 million/year
Large corporations, non-CCPCs, partnerships15%NoNo cap

For most Canadian SMBs, the CCPC structure is the default corporate form, which makes the 35% refundable rate the applicable one. The refundable character is critical: a CCPC that qualifies for $300,000 in SR&ED credits but has only $150,000 in tax payable still receives the full $300,000 — the excess is issued as a cash refund.

Most provinces add their own R&D credits on top of the federal credit. Ontario, British Columbia, and Quebec each offer additional provincial credits that can increase the effective combined rate meaningfully for qualifying small CCPCs. Combined federal-provincial credits for an eligible SMB can reach well above the 35% federal rate alone, depending on province and company size. For provincial credit rates and eligibility, confirm with your provincial finance ministry or a CPA familiar with SR&ED in your province. (Government of Canada SR&ED Program)

What Qualifies: The Three-Part Test

The CRA applies a consistent three-part test to determine whether work is eligible SR&ED. All three conditions must be met simultaneously.

1. Technological uncertainty. There must be a genuine technical obstacle that could not be resolved by applying generally available knowledge or established techniques. Technological uncertainty is distinct from technical difficulty: a project can be complex, time-consuming, and require skilled engineers without qualifying as SR&ED if the path to the solution was known. The question is whether your team faced a challenge that could not be solved by applying existing methods from published literature or standard industry practice.

2. Technological advancement. The work must aim to advance the state of knowledge — to produce new understanding, not merely apply existing understanding in a new commercial context. Crucially, advancement does not require success. A systematically conducted experiment that fails is fully eligible SR&ED if it advances your technical understanding of the problem, provided the failure was documented and the investigation was systematic. (WEtech Alliance, SR&ED in 2026)

3. Systematic investigation. The work must be conducted through a structured process: hypothesis, testing, analysis, iteration. Ad hoc problem-solving — even technically sophisticated problem-solving — does not qualify unless it follows a documentable experimental methodology.

When AI and Software Development Qualify

Software development qualifies for SR&ED more frequently than most business owners realize — and AI/ML development qualifies more selectively than the program's general eligibility suggests.

Qualifying software and AI work:

  • Developing a novel algorithm to solve a problem for which no established algorithmic approach exists
  • Building a machine learning model architecture to handle a domain where standard model types — classification, regression, transformer-based generation — cannot adequately address the target performance requirements
  • Training a custom model on proprietary datasets where the combination of training data, domain constraints, and performance threshold creates genuine uncertainty about whether the approach will converge to an acceptable result
  • Reducing inference latency in a production AI application below a threshold that existing optimization techniques cannot achieve without degrading model accuracy
  • Developing custom integration methods where the absence of public APIs or documented interfaces requires systematic reverse-engineering or protocol investigation

Work that typically does not qualify:

  • Fine-tuning a commercially available model (GPT-4o, Claude, Llama) using standard PEFT or LoRA techniques for a business use case, when the technique itself is established
  • Implementing a documented algorithm in a new programming language or framework
  • Connecting existing SaaS tools through their published APIs and documentation
  • Customizing commercial software within its standard configuration options
  • Writing application business logic on top of an established platform where the technical implementation path is known

The CRA tightened its review of AI/ML SR&ED claims in 2026 in recognition of the volume of AI-related claims being filed. Science Advisors specifically examine whether the claimed uncertainty was genuine — whether established approaches were actually tried, documented as insufficient, and the failure informed the subsequent investigation. Re-labelling standard development work as research without that documentation will not survive review. (Chrono Innovation, AI & ML SR&ED Claims: CRA's Tighter 2026 Criteria)

What Expenses Are Eligible

Qualifying SR&ED expenditures fall into four categories, all of which now include the restored capital component:

  • Labour. Salaries and wages of employees engaged directly in SR&ED, including time fractions for employees who split time between SR&ED and other work. Contractor costs are eligible at 80% of the amount paid — CRA's recognition that contract rates may not precisely reflect eligible time.
  • Materials. Raw materials consumed or transformed directly in the SR&ED process.
  • Overhead. You can claim actual overhead costs attributable to SR&ED, or use the proxy method — 55% of eligible SR&ED labour — which eliminates the need to precisely allocate shared costs. Most SMBs use the proxy method.
  • Capital (restored 2026). Equipment used at least 90% of the time for SR&ED qualifies for both SR&ED deductions and the ITC. GPU servers, AI inference hardware, and specialized testing equipment all qualify under this criterion if the usage threshold is met and documented.

Filing: Key Deadlines and Documentation

SR&ED claims are filed with your T2 corporation income tax return using Form T661 (SR&ED expenditures) and Schedule 31 (ITC calculation). The absolute filing deadline is 18 months after the end of the tax year — no extensions are granted by CRA outside narrow statutory exceptions. For a fiscal year ending December 31, 2025, the hard deadline is June 30, 2027.

Documentation is the most common reason claims are reduced on review. Best practice:

  • Maintain a contemporaneous project log — digital is fine — recording the technical uncertainty, the approaches tested, and what each experiment produced or ruled out
  • Track employee time against SR&ED projects separately from other work throughout the year; time reconstructed at claim time carries less evidentiary weight than logs made during the project
  • Document failed experiments explicitly; they are eligible and directly support the systematic-investigation component of the three-part test
  • Retain all technical meeting notes, model training logs, benchmark results, and version control histories related to SR&ED work

The Missed Opportunity

The SR&ED program consistently underclaims relative to its eligible base. Two misconceptions prevent SMB technology builders from claiming what they are entitled to.

The first is the assumption that client-funded development does not qualify. It does. SR&ED eligibility turns on the nature of the work — whether it involves technological uncertainty and systematic investigation — not on who commissioned or funded it. Developing a custom AI tool for a specific client qualifies if the development meets the three-part test. There is no IP ownership requirement in the SR&ED rules.

The second is the belief that documentation overhead outweighs the return. At a 35% refundable credit rate, $200,000 in eligible labour produces $70,000 in cash refund — before provincial credits. For a straightforward claim, documentation typically requires 2–4 days of staff time plus consultant or CPA fees in the range of $3,000–$8,000 CAD. The net cash return on properly documented eligible work almost always exceeds the cost to claim it.


Sources


If your business develops custom applications, builds AI tools, or trains models on proprietary data, SR&ED is the most direct government funding mechanism available to offset that development cost — and the 2026 reforms made it more valuable than it has been in over a decade. Cloud Forces helps Canadian SMBs evaluate their SR&ED eligibility as part of the application development process, ensuring technical work is documented in a way that supports a defensible claim from day one. Explore our custom app development services or contact us to discuss how SR&ED applies to your next project.

Anton Kuznetsov
Founder & Principal Engineer

Anton Kuznetsov is the founder and principal engineer of Cloud Forces, the Toronto firm he started in 2018 to make custom software and AI practical and affordable for Canadian SMEs. He works hands-on across application development, cloud architecture, and the production systems Cloud Forces runs for its clients.

Ready to bring AI to your business?

Book a free AI Readiness Consultation — no commitment required.

Book Free Consultation