European Public AI

Today’s most capable AI systems are increasingly becoming part of Europe’s basic infrastructure and now underpin search, communication, education, healthcare or even security‑relevant systems. Yet this infrastructure is largely developed and governed by a small group of non‑European firms that concentrate investment, data and talent in their hands.

In a new policy brief we take this imbalance as its starting point. We argue that Europe’s dependence on foreign cloud providers and foundation models is not just an economic problem, but a direct risk to strategic autonomy, democratic decision‑making and the European social model.

Public debate about artificial intelligence often swings between uncritical optimism and alarmist scenarios. In this policy brief we deliberately step away from this hype and treat AI as a “normal technology”: powerful and transformative, but ultimately shaped by political and social decisions, rather than an autonomous force beyond democratic control.

In this perspective, the key question is not whether Europe can win an abstract “AI race”, but what kind of AI infrastructure it wants to rely on – and who that infrastructure should serve. The answer we put forward is Public AI: a framework that understands AI as public digital infrastructure organised around the common good.

Public AI rests on three pillars. First, universal, equal and non‑discriminatory access to compute, data, models and applications, supported by open licences, open standards and interoperable, often open‑source ecosystems. Second, mission‑driven public goals that guide investments towards clearly defined public needs and create digital public goods where markets under‑provide. Third, public control through meaningful oversight, public funding or direct provision, with options for citizen participation and public ownership of successful innovations.

The good news is that Europe is not starting from scratch. Last year, the European Commission has launched a range of initiatives – including the AI Continent Action Plan, the Apply AI strategy and the Data Union Strategy – to expand computing capacity, support model development, increase access to high‑quality data and deploy AI in strategic sectors.

Taken together, these steps represent a substantial public investment in Europe’s AI future. Yet, without an explicit Public AI focus, they risk reinforcing existing dependencies instead of overcoming them. If publicly funded computing facilities, models and datasets primarily serve proprietary solutions controlled elsewhere, Europe will have paid for infrastructure that others ultimately steer.

To avoid this outcome, the brief calls for a “full‑stack” perspective. Compute, data, models, applications and software should be treated as parts of a single Public AI stack rather than as separate policy fields. Measures at each layer need to reinforce one another. Equally important are demand‑side tools such as strategic public procurement, targeted support for public‑sector use cases and guidance for responsible deployment, so that new infrastructure is actually used to create public value.

On this foundation, we make policy recommendations with regards to three key layers of the AI stack: compute, models and data.

On compute, we argue that investments must be clearly oriented towards Public AI objectives. In fact, investments in computing power –  and particularly the Gigafactories initiative – need a much clearer orientation towards public AI objectives. While Gigafactories and AI Factories should indeed support a broad range of AI development projects, the strategic importance of public AI should be explicitly recognised and prioritised.

On models, we call for a demand‑driven strategy centred on a family of European public foundation models that are permanently open and democratically governed. These general‑purpose systems should be complemented by smaller, specialised models tailored to concrete domains. Open‑source licensing and transparency about architectures, training methods and data are framed as essential – both for accountability and reproducibility, and to allow European actors to build on publicly funded technologies without creating new structural dependencies.

On data, we argue for a European data commons to counter emerging “data winters”, in which high‑quality datasets are increasingly locked away. A Public AI data strategy should combine legal certainty for using publicly available data in model training with new governance mechanisms for sharing high‑value datasets. Data Labs linked to AI Factories are presented as potential institutional “glue” between data holders, developers and compute providers, provided they are explicitly tasked with serving open‑source development, public‑sector needs and public‑interest applications.

Ultimately, the Public AI approach is not about deploying AI everywhere, but about deploying it where it genuinely serves public needs. We argue against an uncritical “AI first” principle and instead for purposeful deployment guided by evidence, mission‑oriented research and innovation, and strong governance. In some areas, the most responsible decision will be not to use AI at all.

The Public AI policy brief is aimed at European decision‑makers in EU institutions and member states, as well as researchers, public‑interest technologists and civil society organisations that want to shape a different AI future for Europe.

The future of AI in Europe is still open. Whether AI becomes another driver of dependency or a shared public infrastructure will depend on the choices made now. We invite all actors involved in Europe’s digital transformation to engage with its proposals and help build a Public AI ecosystem that truly serves the common good.