European Public AI Policy Brief
Europe’s AI landscape is dominated by capital-intensive, proprietary systems controlled by a small number of non-European actors. This creates an unprecedented concentration of power and limits European sovereignty in a technology that is rapidly reshaping society and the economy. The European Union has responded with ambitious strategies, but risks reinforcing dependencies without articulating a clear sovereign logic for AI development.
Public AI offers an alternative path: AI systems developed under transparent governance, with public accountability, equitable access to core components, and a clear focus on public-purpose functions. This policy brief—commissioned by Bertelsmann Stiftung through their reframe[Tech] – Algorithms for the Common Good project and authored by Alek Tarkowski and Felix Sieker—provides concrete policy recommendations for European policymakers to advance Public AI.
Recommendations across three dimensions of Public AI infrastructure
Compute: infrastructure for open source AI
Europe must leverage its Frontier AI Initiative to build sovereign compute capacity that supports open-source AI development. This includes establishing accessible computing resources for researchers, public institutions, and SMEs, ensuring that European AI development is not dependent on non-European cloud providers.
The brief identifies a gap: while the AI Continent Action Plan invests in AI factories and gigafactories (19 established by late 2025, up to five gigafactories planned with €20 billion funding), it lacks a strategic orientation toward public AI goals. A multi-client model aggregating demand from companies, startups, academia, and the public sector is better suited to Europe’s needs. This approach aligns more closely with public AI requirements, as many public-interest applications don’t require extreme compute levels. Key policy recommendations include:
- Strengthen the focus on public AI goals: Computing investments must explicitly prioritize public AI objectives and recognize the value of open-source AI, addressing what the brief identifies as the most significant shortcoming of the AI Continent Action Plan.
- Adopt a clear operating model for AI gigafactories: Gigafactories should function as one-stop shops offering not just raw compute, but permanent storage for open datasets and models, public inference services, commercial workload capabilities, and technical support—serving a diverse user base rather than single anchor customers.
Models: A family of European public foundation models
Europe must develop a family of European public foundation models that are permanently open-source, democratically governed, and aspire to remain at or near the frontier of AI capabilities. The brief provides clear analysis of how the EU’s AI Continent Action Plan, including the Frontier AI Initiative, can be oriented toward public AI development. It examines existing European efforts—from OpenEuroLLM to national initiatives like Spain’s Alia and Germany’s Teuken—and identifies how these can be coordinated into a coherent strategy.
The brief emphasizes that Europe need not choose between large foundation models and smaller alternatives. Both paths are viable and mutually reinforcing. A two-track approach combines capable general-purpose models with specialized, efficient solutions tailored to specific needs, fostering an open-source AI ecosystem that serves diverse public purposes. Key policy recommendations for model development include:
- Ensure open-sourcing and model transparency, with models available under terms that grant users freedom to use, study, modify and share
- Prioritize development of a family of European public foundation models that are state of the art in capability
- Invest in software commons and open standards, supporting key tools as digital public goods and developing open benchmarks that evaluate public value
Data: Breaking the Data Winter Through a European Data Commons
The brief identifies a fundamental challenge: European AI developers face a widening data gap as proprietary actors secure exclusive access to high-quality datasets while the social contract of the open web weakens. Major AI developers rarely share reusable datasets, creating a “data winter”—a declining willingness to treat data as a common good. This asymmetry particularly damages open-source AI development in Europe.
The Data Union Strategy must address this gap through a two-track approach: establishing legal certainty for the use of publicly available data, while building a data commons of high-quality, specialized datasets. Key policy recommendations include:
- Provide legal clarity on AI model training: Explicitly confirm that existing legal bases for text and data mining cover AI model training on lawfully accessible data, supporting harmonization and cross-border research collaboration
- Build a public data commons: Establish data labs as stewards of a European data commons, enabling access through clear frameworks, ensuring collective governance through trusted institutions, and generating public value through public-interest-oriented licensing
Purposeful deployment over “AI first”
Building on our previous White Paper on Public AI, this brief argues for purposeful, demand-driven deployment grounded in clearly identified needs, rather than an uncritical “AI first” approach that accelerates technology rollout without a clear purpose. European investment should prioritize problem-oriented approaches that respond to genuine societal needs and deliver tangible value. This requires careful balance between supporting industry and advancing public value objectives, with strong governance mechanisms capable of developing precise deployment roadmaps—including the willingness to identify cases where AI is not an appropriate solution.
