Restricted access to AI data could cost Germany 155 billion euros in lost value added
The German economic institute ifo warns that restricted access to AI training data could cost Germany €155 billion in lost value added every year.
The German economic research institute ifo warns that restricted access to AI training data could cost Germany €155 billion in lost value added annually. The figure is not a measured loss, but the result of a model calculation published on Tuesday, 13 October 2026 — in the middle of ongoing European rulemaking on transparency and opt-out rules for AI training data.
What ifo calculated
The starting point is an estimate that generative AI could add €440 billion a year to the German economy by 2034, according to ifo's model calculation as reported by Reuters (AOL/Reuters). If developers find it harder to obtain training data, this potential would be reduced by €155 billion — roughly $173 billion — in annual value added.
The losses are broken down into four channels:
- €45 billion from slower AI adoption in business
- €75 billion from weaker models
- €25 billion from lower research output
- €10 billion from reduced European AI development
The sum of the four items makes up the €155 billion. The largest single item — weaker models — points to the core of ifo's argument: the quality of the underlying technology depends on data access, not just on how quickly companies adopt it.
How the €440 billion figure is composed
The German-language coverage from dts Nachrichtenagentur provides a more detailed breakdown of the potential (dts via trading-house.net): around €330 billion comes from broad AI use across the economy, €65 billion from more productive research, and €45 billion from AI development in Germany itself.
The analysis was prepared by ifo's Center for Innovation Economics and Digital Transformation, led by Oliver Falck, in collaboration with Martin Thelle of the consultancy Implement Consulting Group.
The data gap: 2–3 trillion tokens versus 15–36
A central figure in the study illustrates why data access carries such weight. According to dts, with the caveat "den Angaben zufolge" (according to information from the study), Europe's largest copyright-free dataset contains only two to three trillion tokens. Internationally leading models have already been trained on 15 to 36 trillion tokens.
The difference is on the order of seven to twelve times. The argument is that European model developers, with a far smaller pool of freely available data, will depend on being able to use protected material under the EU's text and data mining (TDM) exception — and that restrictions here will hit European model competition precisely.
Falck's core message
"Anyone who wants a leading European AI model cannot at the same time make access to training data unnecessarily complicated. You cannot have both," ifo expert Oliver Falck told Reuters.
The study's authors urge policymakers to preserve the EU's TDM exception for commercial use, while introducing proportionate transparency requirements and a practically workable opt-out regime. They claim that regulatory uncertainty is already deterring investment in Europe.
At the same time, the dts coverage points out that the TDM exception does not solve everything: questions about remuneration for rights holders, transparency requirements and protection of creators remain unresolved. There is a real tension in the story — ifo recommends preserving the exception, but the question of compensation for creatives is still open, and the study does not fully address it in the available accounts.
Industry agrees — and has its own interests
Two industry players back ifo's conclusions in the Reuters coverage. Ilhan Scheer, chief executive of the German AI company Aleph Alpha, said: "Access to diverse, high-quality training data is essential for competitive AI. Unnecessary restrictions and legal uncertainty make development harder."
Florian Schewor, DACH director at the data analytics company Celonis, argued that AI can only realize its potential if data access is not unnecessarily restricted, adding that German and European industry holds enormous quantities of largely untapped data.
Both statements should be read as interested-party arguments: Aleph Alpha builds language models that depend on training data, and Celonis's business is about exploiting enterprise data. Their support for ifo's conclusions represents genuine positions, but not independent assessments.
An estimate, not a measured loss
The €155 billion and €440 billion figures are scenario calculations extending to 2034, not observed losses or certain forecasts. None of the available sources — Reuters's news report and dts's study review — document the methodology behind the numbers: the assumptions about adoption rates, model quality, or how restrictions on data access would concretely play out. Readers should therefore treat the €155 billion figure as a politically relevant estimate from a respected institute, not as a fixed quantity.
The final outcome depends on how EU regulation is actually designed. It is worth noting that the study does not argue against regulation as such: the authors want both transparency requirements and an opt-out regime, but with designs they call proportionate and practical. The crux of the dispute is therefore not whether rights holders should have control, but how much friction opt-out and transparency requirements may entail in practice.
Why now
The timing of publication is no accident. Europe is in the process of settling the rules for access to and opt-out from AI training data, in competition with the US and China for advanced models. ifo's calculation is an input into that process: by attaching a concrete figure to data access, the study gives opponents of strict restrictions a quotable argument — while the open questions about compensation and copyright show that the debate is far from settled on any side.