The EU AI Continent Action Plan

Our breakdown on the EU's proposed AI Continent Action Plan

Hey all! 

We have been taking a short hiatus, as Frederike is on a leave, and Leevi has been busy teaching EU policymaking to the bright young minds at the University of Amsterdam. However, today, in a break from our regular programming, we wanted to share our dive into the EU AI Continent Action Plan released last Wednesday. 

The Plan, along with fresh calls for public consultation on key initiatives like the “Apply AI Strategy” and the “EU Cloud and AI Development” Act provides a clearer map of the current state of play at the European Commission, and a roadmap for key milestones in the development of AI industrial policy in Europe. 

The EU AI Continent Action Plan elaborates the European response to the geopoliticization of AI

As we have documented extensively, the tide is turning in AI policymaking. Governments are now turning toward boosting AI, after a period of rule-making and regulation driven by the European Union and at least tacitly accepted by the United States and other partners. Increasing geostrategic salience of AI, shifting domestic political balances, and a hostile geoeconomic situation have created a structural openings for a range of nationalistic policy agendas driven by ideals of sovereignty and national competitiveness.

While the United Kingdom and the United States have a clearer articulation of their visions, the European Union’s response to this changing geopolitical moment has not crystallized to the same extent. This is now changing, with the EU AI Continent Action Plan.

The plan, which aims to make Europe the 'leading AI continent'  echoes similar proposals in the UK and  elsewhere: to increase the supply of AI inputs for national champions (compute, data, talent, and frequently, energy) and stimulate demand (public and private adoption programs), while also removing or simplifying regulatory constraints.

Below, we provide a breakdown of the suggestions and highlight key questions to watch.

Compute, data, talent, AI adoption and regulatory simplification

Compute

• Compute is the core of the proposal. Noting the need compute for training, finetuning, testing and deployment, the European Commission has been leveraging existing the European High Performance Computing capacity towards AI development. As a landmark initiative, the European Commission has lately selected 13 new AI Factories in member states, ecosystems which centralize and coordinate AI-optimised supercomputing, data resources and talent for AI development. 

• In a sign of noteworthy intensification of the large scale scaling paradigm, the Plan extends the AI Factory concept to Gigafactories. These are seen as necessary conditions for the pursuit of AGI, consisting of models with “hundreds of trillions” of parameters (by comparison, DeepSeek R1 model has 671 billion parameters, 0,5 % of that number).

• To achieve such models, the Commission suggests the procurement of 100,000s of advanced AI processors, with a hope that these initiatives will motivate the  development, and even eventually manufacturing, of the leading AI chips in the European Union. The funding for this is a mixture of the 20 billion InvestAI program by the European Commission, European Investment Bank and public-private partnerships. The call for preliminary interest will be launched shortly, with an official call to follow afterwards. 

• The plan also calls for the expansion of cloud and data center capacity. The strategy identifies the lack of cloud and edge computing capacity as key obstacles for European AI adoption, with development of EU-based cloud capacity seen as a strategic imperative. As a solution, the plan suggests will be the rumoured EU Cloud and AI Development Act that that will be released in the Q4/2025, with the objective to “triple the data center capacity in the EU in the next 5-7 years”. The telecommunications lobbyists also succeeded to include a namecheck to “telco edge” as a sign of increased positioning of telcos as the European sovereign compute alternatives.

• N.B The call for evidence for the Cloud and AI Development Act was launched today and runs until 4th of June. You can contribute here. 

Things to watch: 

The increased importance of “sovereign AI” (a term in vogue for public investment in the AI market) is  welcome news for the chips manufacturers, who have struggled to diversify their customer base away from the big tech giants. The strategic panic of the European Union and other nation states will guarantee demand even if the commercial AI development will falter. The default beneficiary of the increased compute demand is Nvidia, which has been strategically positioning as the provider of “sovereign AI” solutions, with deepening partnerships with some of the local European AI champions.

However, some competition is emerging. AMD has been recently making inroads to Europe, by acquiring a leading nordic AI, partnering with the largest AI adoption program in Germany, funded by the key power brokers in the German AI ecosystem and providing computing resources for goodwill on the Open European LLM project. From the sidelines, Cerebras is also making inroads in the European AI, with partnership with Mistral solidifying their role as an alternative in the European market..

For the EU Cloud and AI Development Act, the core question is whether the objective for “tripling the data center capacity in Europe” will mean enticing existing digital giants to localize operations in Europe, or whether this initiative will provide space for European alternatives.  

The overarching problem in the compute section is the overindexing on training compute as the key bottleneck. Notably, the market has been recently moving decidedly away from training to inference computing, as more efficient models drive down the costs of training.

Who are the actors, and what are their incentives in developing and training new models on the huge gigafactory infrastructures? As the market for models gets commodified and the efficiency of training goes down, the business case for training such models seems elusive. The latest set of huge models, such as OpenAI’s GPT 4.5 and Llama 4, did not deliver capability increases that would justify the astronomical costs. If the marginal benefit of increasing scale keeps decreasing, the trend towards more economical models will intensify. The pipe-dream of models with “hundreds of trillions of parameters” might lead to deployment of capital to infrastructure that is looking for use and purpose. 

Data

• The Action Plan announces a new Data Union strategy, with a focus on making more data available for AI development and cutting red tape. The public consultation for the strategy will follow in Q2/2025, with publication in Q3. Proposed initiatives also include Data Labs, which will be located in the AI factories and tasked with managing and curating the dataflows. Shared cloud software is proposed to create interoperability between various European data spaces. Finally, the EU sweetheart initiative to create multilingual language data,  ALT-EDIC, also gets a routine name-check.

Things to watch : 

The proposed Data Union Strategy makes it easy to forget that the European Union had a previous data strategy focused on AI during the last legislative term. This strategy included key legislative files like the Data Act and the Data Governance Act, which aimed to create data markets in Europe. These initiatives faced significant headwinds because various stakeholders did not share the Commission's enthusiasm for data sharing due to trade secrets, conflicting rights and privacy concerns. It remains to be seen how the new strategy will overcome these challenges.

The politically contentious topics are the copyright and other privileged data discussions. While not explicitly discussed in the action plan, the availability of data for AI training has internationally often pitted copyright holders and privacy advocates against the AI developers. We expect these questions to intensify in the upcoming months in the EU as well, signs of which are starting to emerge in the ongoing follow-up negotiations on the AI Act. 

AI adoption

• The third pillar is increasing AI adoption, with a focus on key verticals and the public sector. The key initiative here is the Apply AI Strategy, expected in Q3. In practice, this will mean actions such as increasing data and talent, training, and exploring ways to use the AI (giga)factories and other policy tools to drive adoption. 

• The European Digital Innovation Hubs will be key drivers of increased adoption, by helping SMEs adopt AI technologies. There will also be increased R&I investment in AI, as well as initiatives to establish the European AI Research Council.

• The call for evidence for the Apply AI Strategy was launched today and runs until 4th of June. You can contribute here  

Things to watch : 

It is unclear how the ongoing initiatives will link to the private sector driven AI deployment initiatives, such as the announced European AI Champions Initiative backed by the largest industrial giants in Europe and apparently providing the bulk of the 200 billion euros of promised AI investment by the Commission - a fact conspicuously absent from all public communications. 

Talent 

• The fourth pillar is increasing AI talent by developing new Bachelor's, Master's, and PhD programs in AI, establishing a dedicated AI Skills Academy to provide training and education for the development and deployment of (generative) AI, and attracting “top” international AI experts and PhD students through visa programs and by opening legal gateway offices in partnering countries.

Things to watch : 

The push towards increased strategic importance of artificial intelligence will also intensify the increased polarisation of international movement of peoples. Paralleling similar initiatives in the UK and US, also the action plan testifies to the role of the ‘AI elite’ of founders, funders and researchers as a key group over which countries compete over in a pursuit of staying involved in the AI race.  The increased competition on this limited pool of talent can be expected to further drive up compensation.

The term “AI talent” hides crucial variation. Even though Europe has a strong academic history and 30 % more AI scientists than the US, the European AI champions are disproportionately founded by a small pool of ex-Big Tech engineers with core practical experience in developing leading AI models. 

Regulatory simplification

• Lastly, the Action Plan suggests adjustments to the current regulatory framework. The Plan is careful to avoid the explicit language of deregulation, instead aiming to ensure the ‘successful implementation’ of the existing rulebook.  In practice, the Commission proposes internalize some of the costs of compliance, by providing free compliance help with the formation of dedicated AI Act Service Desk, simplifying compliance through further templates, guidance, webinars and training courses as well as collection of stakeholder input as part of the Apply AI Strategy public consultation .

Things to watch: 

As Leevi noted recently, the Commission is in the midst of a delicate balancing act of between the domestic push for a more innovation-friendly framework being balanced out by the need to appear in the face of pressure from the United States. Against this background, it is telling that the regulatory simplification happens under the heading of technical “clarification”. To what extent such clarification of the rules will be purely informational, and to what extent they are used to shape the implementation of the law to a more flexible direction against the legislative intent, remains to be seen.

Conclusion

The debate on the concrete implementation of the Plan is likely to continue during the spring. However, the key actionable milestones for policy wonks are the two calls for evidence on the Apply AI Strategy and Cloud and AI Development Act, both due 4th of June. Mark your calendars and see you soon!

Cloud and AI Development Act - here.

Apply AI Strategy - here