Transfer pricing and the AI value chain
November 2026 | TALKINGPOINT | CORPORATE TAX
Financier Worldwide Magazine
November 2026 Issue
FW discusses transfer pricing and the AI value chain with Paul McSavage, Samit Shah, Keith To, Pascal Luquet and Charles Marais at Grant Thornton.
FW: What do you see as the main trends surrounding transfer pricing and the AI value chain over the past 12-18 months?
McSavage: We are still in the very early stages of understanding artificial intelligence’s (AI’s) ultimate impact on global business models and profitability as it becomes embedded into the day to day operations of multinational enterprises. The last few years have demonstrated that the scale of investment in AI capabilities and digital infrastructure, combined with the pace of innovation, is already dramatically reshaping how multinational groups create value. As AI becomes fully embedded in business as usual operations, transfer pricing (TP) frameworks must evolve at a similar pace, requiring near real-time assessment of where value is being created and how profits should be allocated across jurisdictions.
To: The most significant trend over the past 12-18 months has been the growing recognition that AI is fundamentally reshaping what constitutes value creation within multinational groups. Activities that were historically regarded as key value drivers are increasingly being augmented or automated by AI. As a result, businesses should be challenging whether existing TP outcomes remain aligned with the economic reality of how value is created. In my view, the biggest TP debate over the next few years will not be who uses AI, but who controls, develops and bears responsibility for it. While AI may change the relative contribution of traditional value-added functions, it does not change the fundamental allocation of risk. Multinational groups remain accountable for the decisions made, the risks assumed and the consequences arising from AI-driven outcomes. This means the risk-bearing entities within a group are unlikely to disappear from the value chain, even if AI materially changes how functions are performed.
“One of the most significant TP developments for AI-enabled businesses may be increased controversy over the allocation of residual profits.”
FW: As AI becomes a more significant creator of enterprise value, how are organisations reassessing where value is generated across global value chains?
Shah: As AI becomes a value driver within an enterprise, organisations are reassessing traditional value chain models that were largely built around people, process, capital and physical assets. Increasingly, value may be created through data, algorithms, AI models and digital platforms that can be developed in one jurisdiction, trained on data from many others and deployed globally at scale. This is prompting companies to reevaluate where economic value occurs, what contributes to value creation and how risks and returns should be allocated across jurisdictions. From a business and tax perspective, the focus is shifting toward identifying the locations of strategic decision making, AI development, data governance and ongoing model enhancement, rather than relying solely on headcount or physical presence.
Marais: AI is also prompting organisations to reassess which activities genuinely drive value across the group. As certain functions become increasingly automated, their contribution to value creation may diminish, while activities associated with the deployment, oversight and governance of AI capabilities become more strategically important. In parallel, the increasing centralisation of AI platforms may result in value creation becoming concentrated within a smaller number of entities responsible for managing and controlling these capabilities. This is leading multinational enterprise groups to re-examine whether existing TP models and entity characterisations remain appropriate, particularly where local entities are primarily users, rather than developers or controllers, of AI tools. The challenge is ensuring that profit allocations continue to reflect the evolving functions, risks and economic contributions that underpin value creation across the value chain.
“AI may produce analyses, predictions or recommendations, but it does not sit in management meetings, approve commercial strategies or assume entrepreneurial risk.”
FW: To what extent does AI require a rethink of traditional DEMPE analyses, particularly regarding data, algorithms, model training and ongoing optimisation?
Luquet: The emergence of AI may ultimately reinforce a fundamental TP principle: value is not created by technology alone, but by the people who make the key decisions regarding its development, deployment and use. As such, at the current stage of development, AI does not necessarily require a rethink of the development, enhancement, maintenance, protection and exploitation (DEMPE) approach. In most situations, AI remains a powerful tool rather than an autonomous creator of value. Human decision makers continue to define objectives, select relevant data, oversee model development, validate outputs and assume the key business and economic risks. Consequently, the core DEMPE principles remain highly relevant: the focus should continue to be on identifying which entities perform and control the economically significant functions and have the capability to manage the associated risks. AI does require taxpayers to look more closely at the contribution of data, model training and ongoing optimisation activities. The challenge is not to replace DEMPE, but to apply it carefully in environments where technology may enhance productivity while strategic direction, judgment and control remain predominantly human-driven.
To: AI does not require a wholesale rethink of the traditional DEMPE framework, but requires significant consideration of how value creation is identified and measured. While control over proprietary data, source code, confidential know-how and access rights have always been relevant, AI elevates their importance as key drivers of competitive advantage and profit generation. This is particularly important because AI models are continuously trained, refined and improved, meaning value creation is increasingly dynamic rather than fixed at the point the intellectual property is developed. As a result, TP policies that were appropriate when an AI model was first deployed may become misaligned over time unless they evolve alongside the underlying AI ecosystem.
“While the technology is new, the underlying governance principles are not. AI does not transfer accountability from management to software.”
FW: How are businesses approaching the valuation and profit attribution of proprietary datasets and other AI training assets that may underpin commercial success?
To: Where AI assets are generating measurable value through revenue growth or cost reduction, businesses are increasingly applying TP methodologies that recognise the contribution of proprietary datasets, training data, algorithms and AI models. Valuation approaches may include discounted cash flow analyses based on expected future benefits, comparable uncontrolled prices derived from third-party data acquisitions or licences, and profit split methodologies. These approaches are supported by DEMPE analyses that identify which entities are responsible for the creation, maintenance and enhancement of AI-related assets, as well as which entities bear the associated risks and investment costs. However, many businesses remain in an investment phase, committing significant resources to developing AI tools and datasets while commercial returns remain uncertain. In these circumstances, determining an arm’s length return can be challenging given the difficulty of valuing assets with limited market comparables and uncertain future benefits. Businesses should clearly document the nature of contributions, ownership and control of datasets and models, the allocation of development costs and risks, and the basis for any charges or future profit-sharing arrangements. This helps establish a clear and defensible link between investment, economic ownership and future economic benefit.
McSavage: Businesses where AI is now an integral part of their value chain are increasingly treating proprietary datasets, training data, user information and model-improvement feedback loops as economically significant intangible assets rather than routine operational inputs. As data underpins the quality, results and commercial success of AI-driven business models, there is growing focus on the uniqueness, quality, scalability and regulatory constraints of datasets that feed the AI solutions. As such, it is anticipated that tax authorities globally will increasingly question whether entities contributing proprietary data should share in residual profits generated by the successful deployment of AI solutions.
FW: Where AI systems are increasingly involved in decision making, how should companies think about risk control, governance and economic substance?
Shah: As AI systems play a larger role in operational and strategic decisions, companies need to determine who is accountable. While AI can inform decisions, organisations retain responsibility for the risks, outcomes and compliance obligations associated with them. Effective governance requires clear oversight structures, documented decision rights, robust controls around data quality, model performance and ongoing monitoring of AI outputs. From a TP perspective, economic substance becomes increasingly important. Organisations should assess where key AI-related functions are performed, including model development, training, governance, risk management and strategic decision making. The entities exercising control over AI risks and directing the deployment of AI capabilities should generally be aligned with the economic returns generated. Regulators are likely to focus not only on who owns the technology, but also on who controls and manages the underlying risks and value-creating activities.
To: AI is rapidly becoming embedded in tax risk management, from processing large volumes of data and detecting anomalies to identifying compliance risks and prioritising matters for management review. While the technology is new, the underlying governance principles are not. AI does not transfer accountability from management to software. Taxpayers remain responsible for the accuracy of information provided to tax authorities and for the judgements made based on AI-generated outputs. For multinational enterprises (MNEs), risk control and governance frameworks should focus on ensuring that AI-generated conclusions are transparent, reproducible and subject to appropriate human oversight. Just as importantly, AI outputs must reflect the economic reality of how the business operates. Economic substance is determined by what people do, the risks they control and the functions they perform, not by what an algorithm concludes. As AI adoption accelerates, the challenge for businesses will be ensuring that technology enhances decision making without weakening accountability, evidentiary support or alignment with commercial substance.
“As AI systems play a larger role in operational and strategic decisions, companies need to determine who is accountable.”
FW: What practical steps should businesses take today to strengthen their transfer pricing governance and documentation around AI-related activities, particularly in anticipation of increased tax authority scrutiny?
To: TP risk management has often been treated as an annual compliance exercise. The rapid adoption of AI and the complexity of identifying, valuing and monitoring AI-related activities within MNEs require a more real-time approach. Robust governance frameworks can help businesses identify and manage TP implications as they arise, while generating the evidence needed to support positions and respond to increasing tax authority scrutiny. Practical measures include maintaining an AI activity register to track significant AI initiatives, incorporating tax and TP review into AI investment approval processes, and performing periodic TP impact assessments of significant AI initiatives. Businesses can benefit from establishing clear DEMPE and data ownership protocols, implement processes to track AI-related benefits and document key decision-making events. Clear escalation triggers for new AI initiatives can further ensure that TP implications are identified early and addressed before material risks arise.
Luquet: Businesses should resist the temptation to create an ‘AI file’ in parallel to their existing TP documentation. The real challenge is not to describe the technology in abstract terms, but to evidence who gives it direction, who challenges its outputs and who bears the consequences of its use. AI may produce analyses, predictions or recommendations, but it does not sit in management meetings, approve commercial strategies or assume entrepreneurial risk. A practical governance framework should therefore trace the human chain of command behind AI-enabled processes: who defines the use case, selects and curates the data, validates the model’s relevance, overrides or accepts its outputs, and decides how those outputs are used in the business. Intercompany agreements and TP documentation should then be tested against that reality. In future audits, the strongest taxpayers may not be those with the most sophisticated AI tools, but those able to demonstrate that human control, risk assumption and profit allocation remain consistently aligned.
“As AI becomes fully embedded in business as usual operations, transfer pricing frameworks must evolve at a similar pace.”
FW: What transfer pricing developments do you expect to have the greatest impact on AI-enabled businesses over the next few years?
To: The biggest impact for AI-enabled businesses is not necessarily where AI is developed, but whether taxpayers can defend the assumptions that underpinned their pricing decisions before AI created unexpected value. The Organisation for Economic Co-operation and Development’s proposed changes to the services guidance move TP further toward an environment where tax authorities can challenge arrangements using ex-post outcomes. For AI-enabled businesses, where commercial value can scale exponentially and unpredictably, this creates a significant risk that today’s TP decisions will be judged against tomorrow’s success. My view is that this will become a major source of TP controversy over the next few years. Tax authorities are likely to challenge pricing arrangements using the benefit of hindsight, particularly where AI-related services or capabilities subsequently generate value far in excess of what was anticipated when the arrangements were put in place.
Marais: One of the most significant TP developments for AI-enabled businesses may be increased controversy over the allocation of residual profits. As AI enables businesses to scale rapidly, tax authorities are likely to pay closer attention to which entities are entitled to the resulting returns. At the same time, AI may weaken the historic relationship between people, costs and profits, with relatively small teams able to generate value that previously required much larger workforces. AI value creation is also increasingly continuous rather than linked solely to initial development, with ongoing model training, refinement, governance and data management contributing to value over time. This may lead to greater focus on the maintenance, enhancement and protection elements of DEMPE, particularly where key decisions regarding model evolution, data strategy and risk management are made. As tax authorities become more familiar with AI-enabled operating models, disputes are likely to centre on whether profit allocations appropriately reflect how value is created, controlled and sustained across the group.
Paul McSavage is an international tax partner leading Grant Thornton’s Irish transfer pricing (TP) practice. Based on 20-plus years of ‘big four’ and in-house TP and international tax experience across Ireland, UK, Europe, Australia and globally, his broad experience and skillset goes beyond tax technical. He is adept at providing clear, pragmatic advice in relation to the full spectrum of TP matters, with particular expertise in financial transactions, intangible property and controversy. He can be contacted on +353 87 031 6103 or by email: paul.mcsavage@ie.gt.com.
Samit Shah leads multinational tax pricing solutions for Grant Thornton Advisors. In this role, he leads the firm’s international tax, transfer pricing, global mobility, employer solutions and global indirect tax teams around go to market strategies and technology enablement. He also leads the firm’s TP team and develops initiatives around TP technology, global integration, as well as leading back office development. He can be contacted on +1 (404) 475 0111 or by email: samit.shah@us.gt.com.
Keith To has significant experience advising digital and technology-enabled businesses. His work has included analysing business value chains, intellectual property arrangements, and alignment of economically significant and value creation activities to transfer pricing (TP) outcomes. Central to his approach is understanding how businesses use TP for risk management and aligning operational and tax strategies. He can be contacted on +61 3 8663 6103 or by email: keith.to@au.gt.com.
Pascal Luquet is a transfer pricing and international tax specialist, advising multinational groups on compliance, controversy and tax risk management matters. He also leads the firm’s artificial intelligence and innovation initiatives, promoting the adoption of emerging technologies to enhance client service and operational excellence. He can be contacted on +33 6 1012 1217 or by email: pascal.luquet@grantthornton.fr.
Charles Marais is partner and head of the transfer pricing (TP) practice at Grant Thornton Netherlands since 2016 and is based in Amsterdam. He also serves as a joint global head of TP and head of EMEA TP for Grant Thornton International. Mr Marais has over 20 years of experience in TP and advising a broad spectrum of clients operating in multiple jurisdictions. He can be contacted on +31 (0) 88 676 9259 or by email: charles.marais@nl.gt.com.
© Financier Worldwide