What we should expect from the EU’s new Apply AI strategy
The Apply AI strategy is the right strategy at the right time. Yet to bring Europe back into the game, it'll need a strong collective effort.
After a regulatory period marked by the AI Act, the European Commission is now focusing on applying and scaling up AI in key domains such as science, industry and government, as called for by the Draghi report and data highlighting the limited AI adoption in these fields.
In January 2025, the Competitiveness Compass proposed creating a ‘CERN for AI’ which would provide the necessary coordination and support for both fundamental and applied research.
We then saw the launch of both RAISE, devoted to resources for AI in science, and the Apply AI strategy, focusing on applying AI in eleven sectors, all in the name of competitiveness and AI sovereignty.
There are many good things about the Apply AI strategy. It adopts an ecosystem approach based on coordination and support, which means connecting use cases with their infrastructure and fostering data and skill needs.
The strategy doesn’t provide standardised recipes across domains. Rather, it offers tailored solutions and flagship applications. It does so by keeping a link with the broader EU legislative agenda and investment in a common, open-source pan-European Frontier AI model.
So far, so good.
Alas, the Apply AI Strategy comes with a lot to do but no budget. Key questions are looming, which will need a lot of work in the months to come:
There hasn’t been any foresight exercise on what the EU can achieve. So, what’s the vision for the AI-powered transformation of these sectors?
Would the flagship applications meet the specific requirements for each sector? Would the use cases benefit from current GenAI solutions, classical machine-learning or new paradigms grounded in physical world knowledge to launch us into the ‘physical AI’ age?
Simply relying on large-scale data centres and powerful GPUs isn’t going to be enough. Every sector will require a different solution. We’ve already raised concerns on the investments’ direction in a recent paper. So, what are the compute and software architectures that Europe will need to build successful business models? And is infrastructure being optimally placed to achieve this goal?
Will existing cross-cutting initiatives such as the AI (giga)factories serve industrial sectors’ needs? How can these initiatives be leveraged most efficiently?
The EU’s data strategy has struggled to produce results so far. From the European Health Data Space to Catena-X in the automotive sector, industry hasn’t found a good way to share data in a way that supports primary and secondary data uses. Why would it be different this time?
Simply trying to predict the required skills for each sector isn’t going to work. The Commission and stakeholders should proactively work on AI solutions that ‘augment’ humans, rather than replacing them. The Apply AI strategy should be ‘humans first’ rather than ‘AI first’,
The potential links between the new Apply AI Strategy and the European Competitiveness Fund need to be clarified. Without an effort to bring coherence to the flurry of Commission strategies, the risk is a repeat of what we saw in 2021, with the industrial strategy and the ‘transition pathways’ for industry remaining isolated from the AI agenda, the skills agenda, and even the Green Deal.
These are all important questions... and the list isn’t exhaustive.
We believe the Apply AI strategy is the right strategy at the right time. Yet to make a real difference, it requires a strong collective effort with industry and other stakeholders. It can bring Europe back into the game, with a unique mix of competitiveness, ethical alignment, and sovereignty.
CEPS will actively participate in this effort to translate strategies into real practices by launching a dedicated Task Force on the Apply AI strategy. The journey starts with a dedicated session at Ideas Lab on 3 March!
Andrea Renda, CEPS’ Director of Research and Head of the Global Governance, Regulation, Innovation, Digital Economy (GRID) unit
Every brick — regardless of colour, size, or set — shares one universal standard: the connector. That's why a child in Ljubljana can combine a 1987 space set with a 2026 city set and build something neither could produce alone.
The innovative power is not in the bricks. It is in the interface.
Now imagine LEGO produced by 27 manufacturers, each with a slightly different connector, claiming "compatibility." You end up with a drawer full of expensive plastic that doesn't quite connect.
That is the current state of AI data infrastructure in Europe.
The Apply AI Strategy needs the European Interoperability Framework as its spine — not a footnote. Interoperability is not a secondary concern. It is the primary reason AI remains at pilot scale rather than reaching deployment at meaningful scale. That is why any AI system deployed with public funding should meet defined interoperability standards for data inputs, model outputs, and API access.
But interoperability without guardrails can be dangerous — especially in defence. Civilian and military AI pipelines must remain structurally separated. The dual-use risk is real, and last week's OpenAI and Anthropic situation made that concrete for anyone still in doubt.
Europe's genuine competitive advantage in AI is not raw compute or model scale. It is the democratic commitment to building AI that serves human dignity. That is not a constraint on competitiveness. It is the foundation of it.
The "strategy with no budget" gap you identify is the thing that makes the rest of the document symbolic rather than operational. The sector-by-sector approach is fine but the question I keep coming back to is: what happens when European companies who want to build on European AI infrastructure actually try to use it for real work?
I was at the Mistral EU Hackathon last weekend building an EU information hub - using Mistral models on a European use case felt like a natural fit.
The reality was more complicated. Not a knock on the strategy itself, but the capability gap between European and American frontier models is a concrete problem that no amount of coordination solves in the near term: https://thoughts.jock.pl/p/mistral-ai-honest-review-eu-hackathon-2026