Yesterday, the European Commission presented its AI Continent Action Plan, a set of policy measures designed to boost and shape AI development in the European Union. The Action Plan is meant to enable a “distinctive approach to AI” that is both competitive in what is defined as a “race for leadership in AI” and enabling for the “European brand of open innovation”. This will be achieved, according to the plan, by interventions in five key domains: computing infrastructure, high-quality data, AI development and adoption, talent, and reduction of regulatory compliance burdens.
The plan can be seen as a holistic approach to strengthening the EU’s capacity to build and deploy AI systems, also for public interest goals. On the other hand, the Action Plan needs a stronger vision of purposeful AI deployment if it wants to achieve more than just boost commercial AI development.
The Action Plan largely builds on a blueprint for AI development policies proposed in January 2024 with the Communication on boosting startups and innovation in trustworthy artificial intelligence, which established the AI Factories program and was aimed at creating a European “startup and innovation ecosystem for AI”. The fact that the Commission announced an upgraded, and extended Action Plan just over a year later shows the perceived urgency of investing in AI development.
Key measures proposed in the Action Plan include:
- Full deployment of AI factories, with compute power in the range of 25,000 GPUs
- Building five AI Gigafactories, with compute power in the range of 100,000 GPUs – matching largest commercial efforts to date
- Boosting European cloud capacity through the Cloud and AI Development Act
- Updating the European data governance framework through the Data Union Strategy
- Ensuring access to Common Data Spaces for AI uses, through Data Labs connected to the AI Factories
- Accelerate adoption of AI systems in key industrial and public sector with the Apply AI strategy
- Support research and innovation into European generative AI models and establish the Resource for AI Science in Europe, which will pool resources for scientific research
- Enlarge the European talent pool by establishing an AI Skills Academy, facilitating migration of experts and incentivizing return of European experts from abroad.
The Action Plan is meant to be financed through the Invest AI Facility, which aims to match EUR 50 billion of public funding with EUR 150 billion from private investors and industry.
The AI Continent Action Plan is a policy proposal that has the chance of developing a European public AI infrastructure: an AI stack that is democratically controlled, has reduced dependencies on dominant commercial players. While the overall direction is correct, the outcome will depend on the ability of public actors to orchestrate this complex plan, and adhere to public interest principles (enshrined in EU’s Declaration of Digital Rights). The Action Plan needs to support an alternative development path focused on public interest goals, rather than give in to the hype and the race to attain technological power. In order to fulfill this vision, the Action Plan’s approach to data needs to be refined, and a stronger focus on model development is needed.
The AI Continent Action Plan is a public AI strategy
The action plan comes close to being a public AI strategy, with measures aimed at creating a European AI stack – consisting of publicly supported compute, data and model components – that meets the standard of Public Digital Infrastructure, and at supporting an ecosystem of solutions built on this stack that are trustworthy, adhere to European values, and focus on the public interest instead of extracting value for private gain.
The Paris Charter on AI in the public interest, signed during the AI Action Summit, offers the following policy vision:
“The benefits of AI in the public interest rely on building open public goods and infrastructure, providing an alternative to existing market concentration, ensuring democratic participation, enforcing accountability and developing environmentally sustainable solutions. To fulfil this vision, we focus on enabling conditions for infrastructure in areas with demonstrated benefits in the public interest.”
This means, on one hand, increasing public control over key infrastructure used to build and run AI systems, especially computing power. In other words, the Action Plan can help secure Europe’s digital sovereignty by reducing some of the dependencies on commercial actors, often with dominant positions. This is made possible through funding new compute capacity in the form of AI Factories and Gigafactories.
Supply chain dependencies will remain, in particular with regard to GPUs, and export controls on AI chips, introduced by the United States, might also be a limiting factor. It’s worth noting that the Plan signals the need to develop European GPUs. Still, a GPU development program, if any, will only be announced in 2026. Meanwhile the DARE research initiative led by the Barcelona Supercomputing Center that aims to develop open GPU architecture and EU-made chips is already underway – and more resources should be directed towards it. .
Secondly, and maybe even more importantly, it is a strategy that commits to some public interest goals. There are, for example, dedicated measures meant to support European research and levelling the compute divide between commercial and academic research capacity. And European open-source approaches, many of them supported since last year by the AI factories, are highlighted as means for knowledge sharing, collaboration, and increased transparency.
Nevertheless, the public interest goals of the Action Plan could have been made more prominent – instead, the Commission highlights economic growth, sovereignty and competitiveness.
Success will depend not just on the scale of investments
Several factors will be crucial for the success of this complex Action Plan in serving the public interest. Sufficient investments are of course necessary. In the new Action Plan, the European Commission aims to mobilise EUR 20 billion in funding for the AI Gigafactories, and raises the AI Factories budget from EUR 2 billion to EUR 10 billion. The increase in planned investment, year-to-year, shows that the EC is trying to match the pace of investments, public and private, around the world. But sufficient funding will not be enough.
Secondly, it is a matter of orchestrating enabling policies for other layers of the AI stack: data, models and talent. It would be a mistake to think of European AI policy only in terms of the scale of investments – which is the typical focus in current AI debates, often limited to the question: will EU spend enough money on compute? A successful policy needs to look beyond the funding, and look at how these funds are used to shape AI development.
Finally, for the Action Plan to result in meaningful public interest results, care needs to be taken to set proper guardrails and risk mitigation measures. For example, planned pilots in high risk areas like healthcare, education or the justice system can both create meaningful standards for proper deployment of AI, but also – if things don’t go well, generate major social harms.
The Action Plan is a complex policy with multiple initiatives and few mechanisms for ensuring compliance – there is a need for a digital rights framework to be at the core of this initiative. Similarly, conditionalities tied to both procurement and provision of computing power should offer preferential conditions to public interest uses, for example open-source AI development.
Better model and data governance is needed
The Action Plan lacks a key commitment to developing a state-of-the-art, open-source, AI-Act-compliant European frontier model. The document describes AI Factories and Gigafactories as facilities for the development of AI models – but lacks a clearly stated commitment to develop such a “flagship model” for Europe (OpenEuroLLM is not mentioned). For now, European AI labs have built a flock of small, open-source models that prove their capacity to build. An effort to develop, deploy and sustain a frontier model needs to be framed as more than just investments in supercomputing centers.
Another missing element is an institution capable of orchestrating this complex Action Plan. An observatory, to be developed within the AI Office, is the only institution mentioned – and it will clearly not suffice. There is a need for a major investment not just in technical capacity and talent, but also in institutional capacity to orchestrate the deployment of AI across various sectors and contexts.
A stronger vision for data as an enabler for AI development is also needed – and it’s a welcome sign that this will be the topic of the upcoming consultation of the new Data Union Strategy. There is also a need to strengthen the capacity of the European AI actors to work with data. The quick pace of technical development, combined with a sense that we are in a “data winter”, shows that a public intervention is needed to address this imbalance. Data Labs – data sharing frameworks between AI Factories and Common European Data Spaces – are a good starting point. But, once again, there is a sense that more is needed to both make more data available, and to properly govern its use. It is time to take seriously Enrico Letta’s proposal for a European Knowledge Commons, a centralized data sharing platform – that will be as crucial for the success of European AI as the Gigafactories.
AI technologies hold some promise of public value, but also immediate harms – including environmental ones. The expansion of AI infrastructure has significant negative impact, related to, among other, the energy needed to power data centres and water necessary for cooling purposes. It is good to see that the strategy recognises concerns around water and energy consumption and identifies them as important considerations that need to be addressed as part of the strategy. However, there are other concerns that have not been addressed.
An AI agenda that prioritises competitiveness and seeks to position the EU at the forefront of AI innovation may be fundamentally at odds with the Union’s commitments to environmental sustainability and the green transition. Significant resource consumption stands in direct tension with the objectives of the European Green Deal and the European Climate Law, which set targets for carbon neutrality and resource efficiency. Without strong sustainability safeguards, an AI strategy driven by international competitiveness risks exacerbating environmental degradation, contradicting the EU’s overarching commitment to a just and sustainable digital transformation. Without addressing the full range of environmental concerns, the EU’s AI strategy risks creating a landscape where innovation and growth overshadow climate goals.
