Europe knows how to train top researchers and operates powerful supercomputers. Yet the most powerful artificial intelligence models are being developed elsewhere. This is precisely what a report published by the European Commission’s AI Office warns about. It summarizes the conclusions of more than a hundred experts and sends a message to European leaders. Europe has roughly one to two years to make key decisions. If it fails to do so, it will be negotiating access to the most advanced AI from a weaker position.
The Commission convened an expert forum to assess Europe’s position in cutting-edge artificial intelligence, known as frontier AI. This refers to the most powerful models at the limits of what the technology can achieve at a given time. They write code, accelerate research, and can acquire offensive as well as defensive cyber capabilities.
The forum first met in April. It brought together European developers, researchers, industry representatives, investors, and experts from specialized institutes, as well as officials from Member States and the Commission. The resulting document speaks openly about a closing window of opportunity. That is not something usually found in Brussels documents.
The experts point out that in just three years, models have made enormous progress. Not long ago, they struggled with simple tasks. Today, they are approaching the limits of what current tests can even measure. The market behind this development is advancing at the same pace. Europe, meanwhile, remains on the sidelines. It trains a large share of the world’s successful professionals in the field and produces first-rate research. But the development of the most powerful models is taking place primarily outside the Union.
What Europe lacks and what is holding it back
The starkest warning concerns computing power. According to the report, the European Union hosts only around 5% of the world’s capacity for training and operating models. The experts therefore urge the Union to triple its share.
Computing power is a resource without which cutting-edge AI cannot function. Without enough chips, data centers, and energy, even strong research will remain confined to small-scale experiments. Those without sufficient capacity lose the race for what matters most: the best chips, the ability to pay brilliant minds, and the speed at which models improve.
The shortage of capacity is compounded by the price of energy. According to the report, industrial electricity prices in Europe are more than twice as high as in the United States. Training large models consumes enormous amounts of electricity. The race for AI is therefore shifting from software to land, permits, chip supplies, and electricity supply contracts. The experts identified computing power and the energy that drives it as the most urgent priority for the next two years.
America vs. Europe
The difference between the continents is also evident in revenues. The two largest American laboratories together generated annual revenue of around 60 billion dollars. Europe’s leading companies generate somewhere around one percent of that amount. Revenue is what determines who can buy the best chips, who can pay top developers, and who can improve their models faster than everyone else.
The situation is not hopeless. The report lists the areas in which the Union continues to hold strong cards. These include the size of the single market, the strength of European regulations, its industrial base, and its leading position in certain parts of the AI supply chain. The experts advise first mapping these strengths thoroughly, then protecting and using them wisely. They also mentioned cooperation with like-minded countries facing similar constraints. Together, they can multiply their influence.
Recommended changes and objectives
The experts’ prescription is to speed up permitting for data centers and grid connections. To address more expensive and slower access to electricity. And above all, to provide legal certainty around training data, where developers are currently being held back by ambiguities surrounding copyright and privacy protection. Copyright in particular remains an unresolved issue. Too little leeway stifles European companies. Too little protection, in turn, will pit authors and publishers against Brussels. Finding the right balance will be tricky.
In addition, the experts propose making several bold, high-risk, high-return bets in areas where Europe could take the lead. Whether within the current direction of development or through alternative approaches. Industrial AI, where Europe has an advantage thanks to automation, is mentioned most often.
The experts agreed that the Union’s pace must match the speed of the technology itself. They therefore recommend setting ambitious targets for 2030 and making frontier AI a genuine political priority across European institutions.
And once again, money. European AI is running up against a shallow capital market for companies in the growth stage. Funding can be found for initial ideas, but large-scale expansion is much harder. Making this capital accessible is among the main tasks outlined in the report.
Sources: digital-strategy.ec.europa.eu and euractiv.com
