Europe needs an industrial policy for AI

THE recent debate about reforming the EU AI Act proves that the law does a reasonable job identifying risk, but a poor job solving Europe’s growth problem.

This is a substantial criticism: Unlike previous tech regulations like the Digital Services Act, which primarily affects US-based social media firms, or the Digital Markets Act which regulates large platform businesses, the EU AI Act – just like the GDPR – touches every large European business.

What is needed is an industrial policy for the age of AI. This industrial policy must focus on the most promising sectors rather than be horizontal, it needs to foster competition, not “European champions”, and it needs to leverage the idea of Europe as a “middle power” in AI.

Europe is not home of any of the frontier AI models. But we’re still in the race with Mistral and the recent acquisition of Aleph Alpha by Canadian AI firm Cohere, which provides an opportunity to create an AI lab that has a global perspective on AI with a “middle power” mindset.

Because of this, Europe needs to adopt a very distinct strategy from the US - which bets on being the first nation to achieve artificial general intelligence (AGI) - and China - which focuses on making their models widely available across the globe and selling their whole IT stack along the way.1

Instead of being an AI innovator, Europe should first and foremost become an AI adopter. As Jeffrey Ding argues, the ability to quickly absorb new technologies and drive adoption is a more powerful predictor of political power than actually inventing these technologies.2 In other words: Europe needs to drive value creation with AI.3

This is the essence of what it means to be an “AI middle power”. According to Dean W. Ball and Anton Leicht, middle powers are:

nations with significant institutional and industrial capacity but without frontier AI development capability. Unlike great powers, they do not control whether and how AI gets built, and whether it rips through society or not. Yet unlike the least developed nations, middle powers possess institutional substrate – functioning economies and bureaucracies. They have something to transform. Unburdened by the need to win a race they were never going to win, they are free to – but must – ask different questions on how to deploy the assets they still have to secure the future they want. The race that remains open, the race where the middle powers can compete and even lead, is the race to favorably adapt frontier artificial intelligence into their governments and economies.4

In the past, European industrial policy often coalesced around creating a European search engine, cloud service, or social network. In other words: Picking the winner. In the AI age, Europe needs to choose winning, and this means building on the strength of the European economy.

When comparing the AI action plans of the EU, the UK, and the US, it becomes clear that Europe’s advantage lies in the availability of data and in the strength of Europe’s industrial model (a key factor for an AI middle power).

These advantages can best be leveraged by deeply integrating AI in sectoral legislation and regulatory guidelines – which was one of the most contentious questions in the trilogue negotiations for the AI omnibus that was finalised earlier this month.

CEOs of leading European technology firms, including Mistral‘s Arthur Mensch, have framed Europe’s AI opportunity in similar terms:

For the past two decades, the focus has been on building the digital world. But the next phase of innovation will be defined by how digital capabilities are applied in the real world – across industries, infrastructure and entire economies. To create value, technologies like artificial intelligence must be connected to the physical systems they are meant to improve.

Indeed, the adoption of industrial robots was a leading driver for the emergence of European “superstar firms” during the fourth industrial revolution5 and it seems plausible that this is even more important for AI-powered robots.

On the other end of the spectrum, some of the obligations of the EU AI Act are overly burdensome for small and medium-sized businesses: My wife runs a small social impact cleaning company, Kehrwork. If she wanted to use LLMs for instant, image-based quality controls, this would likely be a “high risk” use case, with all the obligations that come with it.6

In other words: The adoption of technology is key for the success of Europe’s economic model but our laws are too broad to cater to the needs of our leading industries and too strict to encourage broad adoption.

The lack of speed when adopting AI matters because the technology is improving at an exponential speed. Anthropic‘s Jack Clark believes that there is a chance of more than 60 per cent that AI systems will be able to build themselves by 2028.

In the case of AI development, this means that AI systems are able to write code, test it, interpret and integrate the feedback and update the code in iterative circles. This way, the AI system continuously gets better.

In the physical world, this is more difficult (experiments may require real world labs), but the logic still applies: Connected to the right data, AI agents might be able to greatly accelerate research and development (R&D) in fields that today are dominated by European firms such as engineering and chemicals.

This means that Europe has five to ten years to integrate AI tightly into our industrial base – or risking severe disruption.

The objective is clear: Europe needs an industrial policy for the AI age that encourages companies to embed AI into their R&D and production processes.

The EU AI Act explicitly allows for such sectoral regulation: According to Annex I of the act, certain AI applications are regulated under sectoral legislation rather than the EU AI Act. This provision has been strengthened in the recent AI omnibus which clarified that some industrial applications would also fall under sectoral laws.

This approach makes sense: Sectoral law is better suited to find the right balance between safety and innovation, based on the specific context if an industrial sector.

It is also in line with other European laws which for example foresees stricter rules for the use of cloud services for financial services companies because lawmakers consider the use of cloud services in this sector to be more critical.

European law makers should build on this architecture and extend the areas where AI is regulated by sectoral laws rather than a horizontal approach. The EU AI Act will then over time become a frontier AI law, in line with laws that other countries currently consider or have implemented.

In parallel, the EU should convene its major industrial players and orient its AI funding towards supporting breakthrough research and development in these core industries, similar to the Genesis Mission launched by the US administration in November 2025.

This can be achieved by launching European “missions”, where the EU sets goals and provides funding or prices to companies that achieve breakthrough results towards these goals (I have previously written about this idea). This competitive approach to industrial policy means companies can move with greater speed and less committee work.

There is less than a decade to get this right and ensure Europe gets to the forefront of an AI economy. But we need to start walking into the right direction today.