1 Introduction

Sometimes the answer to a question depends on who you ask. I vividly remember Enrico Letta at the “One Europe, One Market” summit, sharing how he had asked whether it was possible to achieve digital sovereignty for Europe, and someone in the audience asked ChatGPT, which replied “No, for these three reasons”. Shortly after, someone else in the audience said he could not accept that answer and instead asked a European AI model, which responded that digital sovereignty would indeed be possible for three different reasons. 

I find this anecdote particularly fitting, because 1) it illustrates that we need these capabilities in Europe, and 2) it places the topic of frontier AI within the broader context of digital sovereignty. This is, in fact, the first consideration that I would like to share with you now:

In order to succeed in joining the “Premier League” of frontier AI, we must approach this topic within the broader context of digital sovereignty. Let me elaborate: 

2 The broader context of digital sovereignty

We need to consider the connections between different dimensions of technology – such as AI, Chips, Cloud, Robotics, and Quantum Computing – as this allows us to benefit from interdependencies. 

Together with Claudia Plattner, President of the Federal Office for Information Security, and Kristina Sinemus, Minister for Digitalisation and Innovation in Hesse, I published a paper on this topic earlier this year. It contains 37 implementation-oriented measures and connects the fields of technology that I have just mentioned. 

Let me give you an example of why action on chips and action on AI need to happen at the same time: To join the “Premier League” of frontier AI, we need to have access to sufficient computing power and, for that, we need data centres capable of training and deploying AI models as powerful as those in other regions. Using conventional chips, we would face the constraint of being able to supply sufficient energy.

In Europe, however, we have the research and innovation in photonic chips. These chips operate with optical signals and are particularly useful for visual AI applications. Most importantly, they are significantly more energy-efficient than current conventional chips. Photonic chips are ready to go into mass production now. If we finance their production and construct data centres based on this next generation of chips, we would have an opportunity to build momentum when it comes to the training and use of frontier AI models, too. 

3 Concrete measures

In order to get to this point – and this is my second consideration – we need to pool public demand, improve the translation of innovation into marketable products, and invest in application-oriented research in selected cases.

Let us stick with the example of chips. To start production of photonic chips in Europe, we need to identify start-ups that would be able to make this happen. The public sector could guarantee a certain level of demand – where economically useful – and private capital would need to be mobilised, for example through the VC market. In this context, Enrico Letta’s push for the savings and investment union, as well as measures like the 28th Regime and EU Inc., are crucial.

Chips and AI are just one example of these interdependencies. I would like to point out that a lot is already happening, for example the Scaleup Europe Fund to improve VC financing. 

When it comes to bundling public demand, I would like to highlight the example of cloud services. At the Bundesbank, we are currently tendering for EU-native cloud services. For a large part of our multi-cloud strategy, we are aiming for EU-native environments. Within the Eurosystem, too, we are pursuing collaboration with EU-native cloud providers in the context of large integration programs, for example in the area of bank reporting. Bundling public demand at the EU-level would, of course, have an even greater impact. 

4 The creation of an “AI Airbus”

As my third consideration, I would like to mention one of the concrete measures on AI that we are proposing: the creation of an “AI Airbus”.

When I spoke with Jean-Claude Trichet at the beginning of the year, I asked him why we had an “Airbus” in the aviation sector, but not in the automotive, chemicals, or fashion sectors. He pointed out three reasons: 1) In the 1960s, Europe was trailing far behind the United States in terms of aircraft production. 2) There was not a single global champion with European origins. And, 3) With Franz Josef Strauß and François Mitterrand, we had two political leaders who moved roadblocks out of the way. Today, a similar approach for transformative AI is promising. 

We need at least one more European-trained large language model in addition to Mistral, which we also need to support further, for example through public demand as well as private capital. Such an “AI Airbus” would be a European corporation that would build upon an existing open-source model, host it on European infrastructure, and leapfrog its training.

This is critical for several reasons. 

So far, we in Europe are mostly price-takers in the IT sector, and this could affect our price level if we are unable to broaden the oligopoly in this market – aside from the fact that we risk missing out on growth opportunities in this industry. 

In addition, cybersecurity based on high-performance AI models is a matter of survival for Europe – both literally and economically. 

Finally, Europe is home to world-leading firms in manufacturing, engineering, and chemicals, along with decades of highly specialised data. So, in addition to developing another European-trained LLM, we need to develop specialised large language models tailored to Europe’s industrial data. This will unlock new business models made in Europe. 

So, to sum up, we need to act strategically on digital sovereignty to truly reach the “Premier League” of AI. I am optimistic that we can achieve this if we take concrete measures now. I am delighted to be at this summit to find contributors and push forward concrete measures to get there.