AI + the R&D Tax Credits: What Qualifies, What Doesn't, and How to Claim It in 2026
Published: March 11, 2026
Last Updated: March 15, 2026
By Jonathan Cardella
22 min read
Key Takeaways
- Companies that build AI and companies that use AI to transform operations can both qualify for R&D tax credits under Section 41.
- Every AI project must pass the IRS Four-Part Test: permitted purpose, technological uncertainty, process of experimentation, and technological in nature.
- Post-OBBBA, Section 174A restores immediate expensing for domestic R&D, making AI investments more tax-efficient starting in 2025.
- "Vibe coding" and AI-assisted development can still qualify if the developer directs experimentation and resolves technical uncertainty.
- Documentation is critical: maintain contemporaneous records of experiments, failures, and technical decisions.
Does AI Qualify for the R&D Tax Credit? Two Types of Companies, One Answer
The short answer is yes, but the details matter. Two categories of companies claim AI-related R&D tax credits.
First, companies that build AI: these firms train models, develop new algorithms, engineer novel architectures, or create proprietary machine learning pipelines.
Second, companies that use AI to transform their operations: these businesses deploy AI tools to automate processes, improve decision-making, or create new product capabilities.
Both can qualify under Section 41 of the Internal Revenue Code, provided the work meets the IRS Four-Part Test.
The AI Four-Part Test: How the IRS Evaluates Your R&D Claim
The IRS evaluates every R&D credit claim against four criteria.
First, the work must have a permitted purpose: developing a new or improved business component (product, process, software, technique, or formula).
Second, there must be technological uncertainty about the capability, method, or design.
Third, the taxpayer must engage in a process of experimentation, systematically evaluating alternatives.
Fourth, the work must be technological in nature, relying on principles of engineering, computer science, or physical/biological sciences.
AI Activities That Qualify for the R&D Tax Credit
- Training custom machine learning models on proprietary datasets
- Developing novel neural network architectures
- Building computer vision systems for quality control
- Creating natural language processing pipelines for document analysis
- Engineering recommendation engines
- Developing autonomous systems
- Building predictive maintenance algorithms
AI Activities That Do Not Qualify
- Purchasing off-the-shelf AI software
- Using pre-built APIs without modification
- Routine data entry or data cleaning
- Marketing analysis using existing tools
- Standard IT infrastructure maintenance
Real-World Scenarios: Qualifies vs. Doesn't
Qualifies
1.Assuming AI Work Automatically qualifies
every project must independently pass the four part test
Generative AI R&D Tax Credit: Vibe Coding, Agents, and the New Frontier
The rise of generative AI tools like GitHub Copilot, Claude, and ChatGPT has created new questions about what qualifies. "Vibe coding," where developers use AI to generate code iteratively, can qualify if the developer is directing a genuine process of experimentation to resolve technical uncertainty.
The key question is whether the developer is resolving technological uncertainty about the design, method, or capability of the software, not simply using AI as a productivity tool for routine tasks.
R&D Tax Credit for AI Companies: Qualified Research Expenses
QREs for AI companies typically include:
- Wages for ML engineers, data scientists, and software developers
- Cloud computing costs for model training
- Supplies used in experimentation
- Up to 65% of contract research expenses
How OBBBA and Section 174A Changed the Math on AI Spending
The One Big Beautiful Bill Act created Section 174A, restoring immediate expensing for domestic R&D costs starting in 2025. AI development costs are now fully deductible in the year incurred, rather than amortized over 5 years. Small businesses under $31M in gross receipts can also apply this retroactively to 2022-2024.
Documentation That Survives an Audit
Maintain contemporaneous records including:
- Project narratives describing the technological uncertainty and experimentation
- Experiment logs with dates, approaches tested, and outcomes
- Technical decision documents explaining design choices
- Time tracking records for personnel involved in qualifying activities
Five Mistakes Companies Make
1. Assuming AI work automatically qualifies.
Every project must independently pass the Four-Part Test.