Germany risks losing €155bn a year under tighter AI data rules, new study warns

Researchers estimate that restrictions on training data could wipe out more than a third of the potential economic gains from artificial intelligence, as European policymakers review copyright rules

Germany could miss out on €155 billion in additional annual economic value by 2034 if tighter restrictions on artificial intelligence training data are introduced, according to a new study by the ifo Institute.

The Munich-based economic research institute estimates that generative AI could contribute an extra €440 billion a year to the German economy by 2034. But making it harder for developers to access the data needed to train AI models could reduce that potential gain by more than a third.

The largest projected loss, €75 billion annually, would come from less capable AI models, while slower adoption of the technology across businesses could account for another €45 billion. 

Reduced research productivity could cost €25 billion, with a further €10 billion lost through lower levels of AI development in Europe.

The figures, published yesterday, come from new research conducted by Implement Consulting Group in cooperation with the ifo Institute, commissioned by Google and industry association CCIA Europe.

The study identifies widespread business adoption of AI as the largest source of potential economic gains, accounting for €330 billion annually.

Improvements in AI-led research productivity could contribute €65 billion, while AI development within Germany could generate another €45 billion.

But the researchers warn that these benefits depend partly on developers having access to sufficiently large and varied datasets to build competitive AI systems.

Oliver Falck, director of the ifo Center for the Economics of Innovation and Digital Transformation, said that policymakers risked undermining Europe’s technological ambitions by making the development of advanced models unnecessarily difficult.

He said: “Anyone who wants a top-tier European AI model cannot at the same time make access to training data needlessly complicated. You can’t have both.”

The researchers highlight the disparity between the amount of freely available training material in Europe and the datasets used by leading international AI developers.

According to the study, AI Potential at Risk, Europe’s largest copyright-free dataset contains between two trillion and three trillion tokens – the individual units of text and other information processed by AI models. 

Leading international models, by comparison, have already been trained using between 15 trillion and 36 trillion tokens.

The researchers warn that additional restrictions could widen that gap, reducing the performance of European AI models and their commercial usefulness to businesses.

The findings come as the European Commission reviews the EU’s copyright framework, including provisions that allow text and data mining for commercial purposes.

These exemptions permit the automated analysis of digital material, subject to conditions including the ability of rights holders to reserve certain uses of their works.

The European Commission launched a review of the EU’s Digital Single Market Copyright Directive in May 2026, creating a fresh debate over how to protect intellectual property while allowing European technology companies to develop advanced AI systems.

The ifo researchers have urged policymakers to preserve the existing commercial text-and-data-mining exemption, accompanied by proportionate transparency requirements, workable mechanisms for copyright holders to opt out and voluntary licensing agreements.

They also propose considering rules governing what AI systems produce, rather than imposing further restrictions on the material used to train them.

The authors of the new study say that uncertainty over future regulation is already discouraging investment in Europe.

German AI companies have also raised concerns about the consequences of restricted access to training data.

Ilhan Scheer, chief executive of German AI developer Aleph Alpha, told Reuters that companies needed both access to suitable material and regulatory certainty to develop internationally competitive technology.

“Access to diverse, high-quality training data is essential for competitive AI. Unnecessary restrictions and legal uncertainty make its development harder,” he said.

“Europe should make full use of its existing legal framework.”

READ MORE: Bumble shares soar 50% amid $1bn takeover talks. Dating app once worth more than $14bn could be taken private by technology investment firm Ithaca after years of falling users, revenue and share price.

Do you have news to share or expertise to contribute? The European welcomes insights from business leaders and sector specialists. Get in touch with our editorial team to find out more.

_{Main Image: The Reichstag building in Berlin, home of Germany’s Bundestag. Credit: Norbert Nagel/Wikimedia Commons (CC BY-SA 3.0)}