New tax law permanently restoring immediate expensing for domestic research costs is reshaping the deductions and credits available to companies investing in artificial intelligence, as well as the compliance work behind them, Grant Thornton specialists said during a September 29 webcast.
Immediate expensing for domestic R&E restored
The One Big Beautiful Bill Act (OBBB), P.L. 119-21, created IRC § 174A, which permanently restores immediate expensing of domestic research and experimental (R&E) expenditures for tax years beginning after December 31, 2024. The provision reverses the capitalization and five-year amortization rule that applied from 2022 to 2024.
For tax purposes, software development is an R&E expense regardless of how it is treated for financial reporting, a divergence at the center of artificial intelligence (AI) spending. David Sites, national managing partner at Grant Thornton’s Washington National Tax Office, who moderated the session, said each dollar of AI spend raises the question “what do we do with those for tax and accounting purposes?”
Monica Bambury, a partner in accounting methods at Grant Thornton, said the tax result often does not follow the books. If a company is “doing the hands-on development, there could be some opportunities, depending on where the activity is being performed, to actually immediately deduct or expense” the costs, she said.
Location and downstream effects
Costs for research performed outside the U.S. remain subject to capitalization over 15 years under IRC § 174. “I think location is everything,” said Caleb Cordonnier, a partner in Grant Thornton’s Washington National Tax Office. The first question, he said, is “where are the people sitting, who’s doing the work, and where are they, internal or external.”
For domestic costs capitalized between 2022 and 2024, the panelists pointed to Rev Proc 2025-28, modified by Rev Proc 2026-32, which lets companies continue amortizing or accelerate recovery over one or two years.
The decision to expense immediately carries downstream consequences that require modeling, Cordonnier said. A larger current deduction can reduce adjusted taxable income and shrink the interest deduction under IRC § 163(j), add to net operating losses constrained by the 80% cap, or interact with the corporate alternative minimum tax. “You don’t want to necessarily trade a temporary item for a permanent benefit or against a permanent benefit,” he said, adding that it “takes a little more than just saying, ‘Yeah, deduct it, and then I’ll get to use it in the future.'”
The research credit and a coordination trap
The expensing changes also affect the research credit under IRC § 41. “AI is not inherently qualified research and it’s not inherently disqualified for the credit,” said Kevin Benton, a partner in credits and incentives at Grant Thornton. “We still have to identify the business component and apply the four-part test to the development activities.”
To qualify, the work must serve a permitted purpose, be technological in nature, involve technological uncertainty, and use a process of experimentation. “Some of the stronger fact patterns are going to be developing proprietary models, developing new algorithms,” Benton said, while some things “are more on the bubble,” such as prompt engineering.
Internal-use software faces a higher bar, Benton said. Beyond the four-part test, it must clear three additional tests: it must be innovative, involve significant economic risk, and not be commercially available. Software available on the market to resolve the uncertainty does not qualify.
The new law also creates a coordination trap, according to the firm’s presentation. Because the OBBB ties the credit to the expensing regime, a cost is a qualified research expenditure only if it is also treated as domestic R&E under § 174A, so the credit study, the § 174A method, and the IRC § 280C(c) election must be reconciled before the return is filed.
Coordination and documentation
The widening split between tax and accounting rules heightens the need for internal coordination. “In the organizations I talk to, the tax folks aren’t always in the know right away on what the companies are doing on AI spend,” Sites said, urging tax and accounting teams to align early.
Panelists closed with practical steps. Christine Janis, a partner in Grant Thornton’s accounting principles group, advised companies to “inventory your significant AI arrangements, classify what you’re buying, what you’re building, what the use is, who the users are, and how those users are going to access it,” then track costs to specific projects.
Bambury urged companies to “identify where your development is happening, whether it’s domestic or foreign, so you can properly treat it under” § 174 or § 174A. For credit purposes, Benton recommended “collecting documentation during the development, so you can understand how it relates to the business component and that nexus to the qualified cost and activities.”
For more on immediate expensing for domestic research or experimental expenditures, see Checkpoint’s Federal Tax Coordinator 2d ¶ L-3075.
Take your tax and accounting research to the next level with Checkpoint Edge and CoCounsel. Get instant access to AI-assisted research, expert-approved answers, and cutting-edge tools like Advisory Maps and State Charts. Try it today and transform the way you work! Subscribe now and discover a smarter way to find answers.
