Grant Thornton tax specialists said during a July 30 webcast that companies deploying artificial intelligence (AI) should assess how the technology changes their operating models, intellectual property, related-party payments, and allocation of profits.
David Sites, national managing partner for international tax in Grant Thornton’s Washington National Tax Office, said AI can change “the value chain for a company.” Existing tax rules still apply, but companies must determine how they apply to a changing operating model.
AI tools can become business assets
Companies often start with third-party foundation models and build tools around proprietary data, internal processes, and institutional knowledge, said Cory Perry, a partner in Grant Thornton’s Washington National Tax Office. Those tools can automate tasks, support employees, or create customer-facing offerings.
Chris Summer, a Grant Thornton international tax partner, identified three broad uses of AI: individual productivity tools, internal productivity systems used at scale, and customer-facing outputs. The first can resemble a conventional software expense, while the latter two can improve operations or alter a company’s products.
Samit Shah, a Grant Thornton transfer pricing partner, said activities producing organization-wide benefits can be “value adding and accretive to the business,” potentially requiring additional compensation for the entities that perform them.
Characterization drives tax results
No Treasury regulations specifically classify AI, Perry said. Companies instead fit an arrangement within existing rules for software and cloud transactions, based on the rights transferred and how a user accesses the technology. Reg. § 1.861-19 generally treats on-demand network access to software and digital resources as a service.
Transactions outside that framework can be analyzed under Reg. § 1.861-18, which distinguishes sales of copyrighted articles from licenses of copyrighted rights. An AI system that replaces services and is accessed remotely can be a cross-border service, while rights to exploit a model can support a royalty.
That distinction can affect withholding tax, sourcing, foreign tax credits, and base erosion and anti-abuse tax treatment. Perry said a royalty paid to a foreign related party can be a base erosion tax payment when it is not subject to withholding tax, while a service payment might qualify for an exception.
DEMPE analysis follows substance
For transfer-pricing purposes, legal ownership alone does not determine which entity receives returns from AI-related intangibles. Companies must examine the development, enhancement, maintenance, protection, and exploitation functions each entity performs — the DEMPE framework.
“Legal ownership really won’t get you there,” Sites confirmed. Companies need evidence of who funded, controlled, and performed development, and who makes deployment and monetization decisions.
The analysis is difficult when several entities contribute data, fine-tuning, prompt engineering, retraining, or monitoring. Companies must also distinguish routine maintenance from value-creating activity. Current IRS and OECD guidance does not directly address AI-to-AI transactions, making documentation of who controls AI decision-making important.
Mat Knudson, a Grant Thornton transfer pricing manager, said the larger disputes often involve a tax authority claiming a share of profits tied to an intangible, rather than a service-provider markup.
Planning can unlock incentives
Perry said early planning can also unlock incentives. Domestic AI development costs may generate research credits under IRC § 41, and revenue from charging foreign affiliates or customers for AI tools may qualify as foreign-derived deduction eligible income. Because research expenses do not reduce the income eligible for that deduction, companies can reach low effective rates when they structure the arrangements correctly.
Whether AI development occurs in the United States or abroad also affects cost recovery, Summer said, because domestic and foreign research expenditures are capitalized and amortized over different periods.
Governance should begin early
The panel recommended building governance into AI projects before broad deployment. An inventory can identify models, training data, algorithms, and infrastructure, along with the relevant entities, jurisdictions, costs, and DEMPE functions.
Companies should also review intercompany agreements to address ownership, data contributions, improvement rights, cost allocations, and payments. “Having a clear delineation and understanding of who owns the output, who owns that value creation is critical,” Shah said.
Agreements must reflect how the business operates, the advisers said. Sites cautioned that contracts and intellectual property registrations cannot substitute for evidence of where development, control, and risk-bearing functions occur.
Knudson recommended tracking each project’s expected and actual return on investment, including cost savings, revenue, or error reductions. “Here’s what we thought we were going to get by developing these tools. Here’s what we tracked, and here’s how it actually went,” he explained.
Summer emphasized that tax teams should begin by understanding what a company spends, where it spends it, and which entities incur the costs.
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.
