The Economics of Copyright and Data Access in European AI Development

This ECIPE Insight is based on a forthcoming ECIPE publication on Copyright and AI.

Three developments in early 2026 have brought the relationship between copyright law and artificial intelligence to a head. On 28 January, the European Parliament’s Committee on Legal Affairs adopted the Voss Report, calling for tighter restrictions on text and data mining (TDM) for AI training. Separately, the Court of Justice of the European Union is now considering Like Company v. Google Ireland (C-250/25), its first case directly addressing generative AI and copyright. And the European Commission has launched a consultation on technical protocols for the opt-out reservation of rights under Article 4(3) of the Copyright in the Digital Single Market Directive (CDSMD).

The argument in this ECIPE Insight is straightforward: restricting the availability of data for AI training harms EU competitiveness in two distinct but connected ways. It undermines the EU’s capacity to develop AI and it undermines the EU’s capacity to deploy AI across the broader economy.

To understand why data restrictions matter so much, it helps to start with a conceptual framework. Economists have long recognised that a country’s comparative advantage is shaped by economic endowments such as land, capital or labour and by the regulations that govern access to those endowments. In the digital economy, data has become a modern economic endowment. But data does not automatically translate into competitive advantage. Firms must be able to access, process, and learn from data. Regulation acts as the mediating force: it can either facilitate or obstruct that process.

Figure 1 illustrates this logic. Regulation shapes the availability of modern endowments. In turn, the availability of those endowments determines the comparative advantages that firms can develop. And those comparative advantages drive the economic outcomes that policymakers care about: trade flows, investment patterns, and market structures.

Figure 1: Model for understanding the behavioural effects of regulationSource: ECIPE, adapted from Erixon and Guinea (2024).

This framework has a concrete implication. When regulation restricts access to a key economic endowment, it does not simply create compliance costs. It redirects the structure of the economy. The EU has already seen this dynamic play out with the GDPR: two years after its entry into force, EU firms stored 26 per cent less data than their US counterparts and had reduced computational activity by 15 per cent.

Figure 2 applies this framework directly to copyright law and AI development. It is Figure 1 with three additional mechanisms layered on top.

Figure 2: Model for understanding the behavioural effects of regulation – the case of restrictive copyright lawSource: ECIPE, forthcoming.

The first mechanism is access and exclusion. Copyright law defines who may legally reproduce or reuse copyrighted material for purposes such as TDM. Under Article 4 of the CDSMD, rightsholders can expressly reserve their rights, including through machine-readable opt-out signals. If opt-outs are exercised extensively, the pool of legally accessible training data shrinks.

The second mechanism is transaction costs. Where data cannot be freely mined, developers must negotiate licences with individual rightsholders, manage large volumes of opt-out signals, and navigate legal uncertainty about whether their compliance methods meet the requirements demanded by regulation.

The third mechanism is entry barriers. Transaction costs do not fall equally. Large, well-capitalised firms can negotiate licences, build compliance infrastructure, and absorb legal risk. European start-ups and SMEs cannot. The result is a market structure skewed in favour of non-EU incumbents who have already built their models.

Together, these three mechanisms reduce the availability of data which translates into a particular economic specialisation and outcomes.

For example, the EU is a net importer of AI-related products, from foundational models to cloud computing infrastructure. This reflects the underlying forces of the economic model presented in Figure 1 and 2. When the fundamental input for a technology is made artificially scarce, domestic producers are less likely to invest in developing that technology, and more likely to import it from jurisdictions where the input is cheaper and the legal environment is clearer.

As a result, EU producers may end up focusing on downstream applications built on models developed elsewhere. For the EU companies with the resources and ambition to develop foundational models, the rational response may be to relocate to a jurisdiction where the legal environment is more favourable.

The concern about strategic autonomy, which EU policymakers raise in other contexts, applies here. AI models developed outside the EU, trained predominantly on non-European data, will be less useful for European firms. However, Europe will not stop using AI. What Europe risks is becoming just a consumer rather than a consumer and producer of AI.

The damage does not stop at the ICT sector. AI is a general-purpose technology, and its impact on competitiveness will be felt across every sector of the EU economy. The EU may not be positioned to build the next generation of frontier foundational models, but it can – and must – be able to fine-tune and adapt those models for sector-specific applications. This is where European competitiveness is most directly at stake.

The empirical evidence on the relationship between regulatory restrictions and economic performance is unambiguous. ECIPE research shows that a 10 per cent increase in the combined effect of services trade restrictions and digital technology uptake is associated with a 1.3 per cent decline in value added. Regulatory restrictions limit the diffusion of digital technologies, and lower diffusion translates directly into lower business growth.

The TDM exception under Articles 3 and 4 of the CDSMD is one of the few instruments in the EU’s regulatory portfolio that moves in the direction of making data more accessible rather than less. Every other relevant framework – the GDPR, the DMA, the AI Act – imposes restrictions on data access. The TDM exception serves as the counterweight in this framework. Weakening it would not merely reduce a legal permission; it would shift the balance of EU data policy further towards restriction, precisely at a time when Europe’s economic competitiveness increasingly depends on its ability to harness data as a productive resource.