Do AI Investments Qualify for R&D Tax Credits?
Summary
- AI investments may qualify for federal and state R&D Tax Credits when they involve developing or improving technology, not simply using off-the-shelf tools.
- H.R. 1 restored immediate domestic expensing of R&D costs under new section 174A, reversing the mandatory five-year amortization that applied from 2022-2024.
- Small businesses (under $31 million in gross receipts) may be eligible to retroactively amend 2022-2024 returns to claim immediate deductions.
- AI projects must meet the four-part Section 41 R&D test.
- Developing proprietary AI models, algorithms, data pipelines or novel workflows may qualify.
- Routine AI implementation, subscriptions or prompt engineering generally do not qualify.
Artificial intelligence (AI) has rapidly evolved from experimental technology to a core operational driver across many industries. Chief Executive Officers (CEOs), Chief Information Technology Officers (CIOs) and Chief Financial Officers (CFOs) are approving record levels of investment in AI to automate processes, improve decision-making and gain competitive advantage. Yet many organizations are overlooking a critical opportunity to align those AI investments with the tax rules that govern research and development (R&D) deductions and credits.
This disconnect, often described as the “AI Paradox,” occurs when companies aggressively deploy AI while failing to distinguish between using AI and developing AI for tax purposes. This distinction determines whether costs are treated as routine operating expenses or if they qualify for research activities, which are eligible for immediate deductions and value credits.
Recent legislative changes with the One Big Beautiful Bill Act (H.R. 1) ended the period of mandatory amortization and restored domestic expensing under Section 174A. With this relief now in place, the priority is clear; companies must carefully define where AI usage ends and development begins. Getting this right can mean the difference between AI being a pure cost center, versus AI becoming a source of meaningful, dollar-for-dollar tax savings.
The H.R. 1 Revolution: Immediate Expensing is Back (With Lookback Relief)
For tax years 2022 through 2024, companies were required to amortize R&D costs over five years for domestic research and 15 years for foreign research. This requirement created cash-flow strain and reduced the near-term value of R&D incentives.
H.R. 1 reversed course by the introduction of Section 174A, which restores 100% immediate expensing for domestic research activities. For companies working on AI development in the U.S., this change can significantly accelerate tax benefits. However, foreign R&D, such as work performed by offshore development teams, must still be amortized over 15 years.
Equally as important is the retroactive relief available to small businesses. Companies with average annual gross receipts under $31 million may amend their 2022 through 2024 returns to claim immediate deductions for domestic research costs previously amortized. For a significant number of growing companies, this creates a rare opportunity to unlock refunds tied to past AI investments.
It is also critical to understand the interaction between Section 174A and Section 41. The former governs the deduction of search costs, while the latter provides the subsequent R&D Tax Credit based on those same qualified expenses. Together, these provisions can significantly reduce both taxable income and tax liability for qualifying AI investments.
The Four-Part Test for Section 41 R&D Credits
To qualify for the federal R&D Tax Credit under Section 41, AI-related activities must satisfy a well-established four-part test. These requirements apply regardless of whether the project involves traditional software development or more advanced machine learning and data science.
Permitted Purpose
The activity must aim to create a new or improved business component, such as a product, process, technique, formula or software. This could include developing a custom automation process, designing proprietary algorithms, building application programming interfaces (APIs) or creating a new code or custom predictive model for the company’s specific needs. Simply using a chatbot or a third-party AI tool does not meet this standard.
Technological in Nature
Qualifying activities must rely on principles of computer science, data science, engineering or mathematics. AI development that involves model architecture decisions, data engineering, algorithm design or system optimization typically satisfy this requirement.
Elimination of Uncertainty
At the onset of a project, there should be some uncertainty regarding the capability, method or appropriate design of the solution. For AI projects, uncertainty often arises around whether a model can achieve required accuracy, scale efficiently, integrate within existing systems or process complex data sets. Activities undertaken to resolve these unknowns are central to R&D eligibility.
Process of Experimentation
Finally, the company must engage in a process of experimentation to overcome the identified uncertainty. This involves evaluating alternatives through iterative testing, modeling, simulation and trial-and-error. This could include training AI models, testing different data inputs, adjusting architecture or benchmarking performance against defined objectives.
What Qualifies: AI Developments vs. AI Implementation
A common misconception is that R&D must be groundbreaking or new to the world to qualify. In reality, the standard is whether the activity is new to the taxpayer. This distinction is especially important in AI, where a wide range of tools and techniques already exist but are being adapted or extended in novel ways within a specific organization.
Some examples of qualifying activities include:
| Qualifying Activity (Expensed and R&D Credit) | Non-Qualifying Activity (Routine Expense) |
| Developing/enhancing proprietary models on unique datasets. | Integrating ChatGPT or Claude via a standard API. |
| Building custom data pipelines to handle “unstructured” processed data. | Paying for monthly Software as a Service (SaaS) subscriptions (e.g., Midjourney). |
| Developing novel workflows with technical uncertainty. | Routine prompt engineering for content generation. |
| Optimizing model latency/speed for edge computing. | Basic bug fixing and routine software maintenance. |
Capturing the “Hidden” AI Costs
Even when companies correctly identify qualifying AI development activities, they often underestimate the associated costs that may be included as qualified research expenses.
One of the most frequently overlooked areas is cloud computing costs. Computing time incurred on platforms such as AWS, Azure and Google Cloud while developing and testing AI models, excluding production usage, can qualify as a significant component of R&D expenses. Given the resource-intensive nature of AI training and experimentation, these costs can be substantial.
Wages typically represent the largest category of qualified research expenses. While data scientists and machine learning engineers are obvious candidates, eligibility also extends to software engineers, project managers and even supervisory personnel.
Contractor costs can also qualify. Amounts paid to third parties for direct research or technical support related to AI development may be included as qualified research expenses, subject to applicable limitations. This is especially relevant for companies that rely on external specialists or development firms to supplement internal teams.
Next Steps and Recommendations
CFOs should review their AI spending and 2022 through 2024 tax filings for potential refund opportunities, with particular focus on small businesses who have the broadest retroactive options. It’s also important to re-evaluate R&D Tax Credit strategies, including the impact of Section 280C(c)(2) elections, and to prepare current or amended returns promptly.
When approached thoughtfully, AI development can deliver both operational advantages and meaningful tax savings through immediate deductions and R&D Tax Credits.
If your organization is investing in AI and unsure whether those efforts qualify for R&D Tax Credits, contact Brown Plus today!