Transfer Pricing and Artificial Intelligence Value Creation

In recent years, the world has witnessed a tremendous development in artificial intelligence technologies, so that it is no longer just a technical tool used to improve some processes within companies, but has become a key element in creating economic value and generating revenue. Global companies have come to rely on it to develop products, analyze data, understand customer behavior, improve operational efficiency, and even innovate entirely new business models. With this transformation, new legal and tax challenges have emerged, because AI is no longer just a helper, but in many ways It is one of the most important sources of profit and competitive advantage within multinational groups.

The fundamental problem is that traditional transfer pricing rules, chief among them the Organisation for Economic Co-operation and Development (OECD) Neutral Pricing Principle, are designed to deal with more explicit and stable intangible assets, such as patents or trademarks. Artificial intelligence is a different asset in nature, because it does not rely on a single easily identifiable element, but rather on an interconnected set of algorithms, data, technical infrastructure, continuous development, and commercial application in different markets. The value generated by it is not only created in the programming or development phase, but may also arise from the data collected, from the contribution of local markets, and from continuous use that leads to the improvement and efficiency of the system.

Hence the emergence of what can be described as a “valuation vacuum” within international tax rules, because these rules do not fully keep pace with the dynamic and evolving nature of AI. Instead of having a fixed asset with a clear owner and a specific developer, we are facing a renewable system whose value elements are distributed between more than one party and more than one country. It has become necessary to rethink how the value generated by AI is analyzed for transfer pricing purposes, not only from the perspective of legal ownership, but also Also, in terms of actual contribution to development, data provision, risk tolerance, and commercial exploitation. In this sense, it is important to examine the adequacy of the current rules and the need to develop a more flexible approach that is able to accommodate this new type of intangible asset.

Dismantling the Traditional Concept of Value Ownership

In traditional multinational corporate structures, intangible assets were seen through their associated functions, in particular the development, enhancement, maintenance, protection and exploitation  (DEMPE) functions, so that the entity that performs or bears these functions is the most likely to receive the main economic return generated by the asset. This logic has been largely appropriate in the case of traditional software or patents, where the developer or legal owner of the asset can be clearly identified. However, this Visualization becomes less accurate when dealing with artificial intelligence, because value does not arise from the code alone, but from interrelated elements that include data, computing power, continuous training, and actual application in different markets.

This problem is even more pronounced when an AI model is developed in a particular country, while the AI model relies on data collected or provided from a subsidiary in another country. In this case, it is difficult to consider the subsidiary as just a routine service provider that deserves limited compensation based on cost alone, especially if the data it provides is of a unique or strategic nature, such as medical records, customer data, or local consumer behavior patterns. Here it is not. Data is no longer just an auxiliary, but becomes an essential part of the process of creating a high-value intangible asset. Therefore, ignoring this contribution when distributing dividends may result in an outcome that does not reflect the true economic realities within the Group.

Tax authorities in many countries have begun to question the traditional model that the developer or legal owner alone accounts for the entire remaining return. With AI, value creation is no longer limited to the initial development stage, but has become a distributed process in which multiple entities may participate, each adding a substantial element to the asset or its commercial success. Hence, there is a growing trend towards recognizing that an entity contributing material data or an operational environment Allowing the system to learn and improve may have a legitimate right to a portion of the super-profits generated by this success, reflecting a clear shift in the understanding of the concept of “value ownership” in the context of the digital economy.

The Challenge of Continuous Learning and the Impact of Market Interactions

Unlike traditional software, which is often a near-finished product once it is developed and rolled out, modern AI models do not stop at the moment of launch, but continue to evolve through actual use and constant interaction with users. This is evident in large language models (LLMs) and in systems that rely on reinforcement learning through human feedback (RLHF), where the outputs of everyday use themselves become part of the model optimization process. Every interaction within a local market, every observation, response or pattern of use can contribute to increasing the accuracy and efficiency of the system, meaning that economic value is not only generated at the center of development, but may also be created during commercial exploitation in different markets.

Hence the issue of market-related intangible assets, i.e., the value that arises as a result of the product’s presence and interaction within a particular market. If, for example, the use of an AI model in Germany improves its performance due to German user interactions and provides it with feedback data that has a real impact on its development, German tax authorities may consider that part of the economic value associated with this model has been generated within their borders. The market is no longer just a place where the product is sold or exploited. It may even become a contributing element to the formation of the intangible asset itself, and to maximizing its value at the global level.

In light of this changing nature, it becomes difficult to rely on traditional transfer pricing tools, especially those that seek similar transactions between independent parties, because such transactions are often rare or unavailable in the field of AI-owned within multinational groups. The trend towards a transaction profit division (TPSM) approach is therefore more appropriate to this reality, as it is based on aggregating the gross profits generated by the AI product and then distributing it among parties or states different jurisdictions according to their respective contribution to value creation. In this context, the evaluation is not limited to R&D spending, but extends to the volume and quality of the data, the importance of feedback loops from local markets, and the extent to which each element contributes to improving the model and maximizing its profitability.

Regulatory Friction

The biggest risk of an AI-based corporate tax audit lies in the Hard-to-Assess Intangible Asset (HTVI) system. Because the future earnings potential of a new AI model are very uncertain at the time it is transferred internally, tax authorities are given “retrospective consideration” powers. If a parent company transfers an AI model to a subsidiary in a low-tax area for a nominal fee, and that AI subsequently achieves a technological breakthrough that has led to huge unexpected profits, the authorities can adjust the conversion price retroactively.

This creates a complex situation for multinationals: they must price the conversion based on current data, but will be judged by the tax collector based on unpredictable future success. Moreover, the physical location of the GPU clusters and computing infrastructure adds another layer of complexity. With the requirements for “economic core” tightened globally, a company cannot simply place its intellectual property in a low-tax jurisdiction without proving that it owns the technical staff and architecture Physical infrastructure to manage the significant risks associated with the development of artificial intelligence.

Conclusion

The future of AI transformational pricing requires moving away from siloed accounting and toward an integrated techno-tax framework. To defend their global dividends, multinational corporations must be able to measure the added value provided by specific datasets, and map the geographic origin of the human feedback that develops their models. As AI continues to decentralize value creation, the tax world is moving toward a reality in which “intelligence” is treated not as a product owned by a single entity, but as a collaborative asset to be taxed Where they learn and grow.

Frequently Asked Questions

To find out more, please fill out the form or email us at: info@eg.Andersen.com

Contact Us