While the world builds and deploys AI at scale, Europe legislates. The AI Act is a recent example: if regulated without an industrial strategy, we will be vassals of the AI empires.
While the United States is investing hundreds of billions in AI infrastructures and China is training its engineers by the millions, Europe is focusing on producing articles of law by the hundreds. Thus, the AI Act is now a reality, but barely published, its technical obsolescence already seems planned. Behind this protective posture hides a delay and a strategic phase shift.
The first major economic bloc to formalize such regulation, Europe intends to structure the uses of AI at a time when technological and competitive balances are in flux. This text perfectly illustrates Collingridge’s dilemma. Theorized in 1980, this principle postulates that at the start of an innovation, it is difficult to anticipate its impacts in order to manage them; but once these impacts are visible, the technology is so entrenched that it becomes almost impossible to control. The AI Act attempts to break this cycle, but at what cost?
Industrial emergency in the face of regulatory mirage
The AI Act is based on a risk-based approach, a relevant strategy to prohibit abuses such as manipulation or generalized biometric surveillance. However, the true nature of this revolution is missing: artificial intelligence is no longer a simple software tool, it is a heavy transformation industry and not a simple technological evolution of digitalization.
The idea that regulation protects Europe is an illusion. As analyzed by Arthur Mensch, co-founder and CEO of Mistral AI, in his Senate hearing on May 12, heavy European regulations create an entry cost that only large companies and therefore American giants can absorb thanks to their colossal resources and their lobbying capabilities.
Although seen by some as an opportunity for local players to stand out, this regulation stifles the ecosystem and local startups. Unlike the United States, Europe requires juggling multiple labor rights, taxes and languages (key elements of our wealth and identity). Brought down to the scale of a globalized market, this situation considerably slows down the scaling up of our European companies.
In his example and with the trajectory already underway by 2030, if 10% of the productivity gained at the European level is based tomorrow on American models, Europe will in fact pay them a massive technological rent, while having to support the destruction of the associated wage bill (Mensch estimates this transfer of value at several hundred billion annually by 2030). Without reinvestment in AI, it will widen its trade deficit and will not create any leverage in its commercial relationships.
Currently, US giants are deploying hundreds of billions of dollars to monopolize access to components, infrastructure and energy, and are therefore hoovering up future capacity and reserving associated resources to support the infrastructure that will enable these profits.
For Europe, the real wall is no longer regulatory or software, but rather energy and material. Faced with non-European players who fuel demand that exceeds supply, they generate a supply crisis where demand far exceeds global production capacities in chips, servers and electricity.
Open Source: a lever of European sovereignty
In this global race, Europe has a major asset, too often underestimated in regulatory debates: Open Source. Beyond a software distribution method, it is a guarantee of transparency and a decentralized innovation engine to fight against an oligopoly.
Open Source is the only area where a mid-sized European company can today build sovereign AI. But we must be lucid: the open reference models today are primarily Chinese such as Kimi, Qwen, DeepSeek, Mimo ahead of the Americans and Europeans.
If French players like Mistral AI or LightOn manage to exist against the American titans, it is thanks in particular to a strategy of open models and the use of open source in the foundations of AI and agent platforms. They have paved the way in Europe but it has not yet created the ecosystem to support it and bring it to the scale of its competitors.
In reality, digital sovereignty goes well beyond open source and is equally dependent on the cloud, semiconductors, network infrastructure, cybersecurity, skills, public procurement, financing and global industrial capacity.
The specter of a regulatory rather than innovative Europe
Legislating to hope to defend against American monopolies does not work. Without a strong industrial strategy to support the sector, the risk is real: slowing down the rise of local players, while allowing foreign competitors to develop in a more flexible framework before investing in the European market.
European companies are already paying the price: compliance costs that crush SMEs while Big Tech absorbs them without blinking and tirelessly continues their lobbying. Added to this is the flight of our engineers and researchers, who are leaving ecosystems where experimentation is stifled by administration without financial recognition up to international standards.
And in the absence of models trained on our data, we already inherit the cognitive and linguistic standards of those who produce them with the consequences: impoverishment of the language, standardized thinking, inability to escape from a framework that will be imposed on us and disseminated continuously.
For living governance
Rather than a fixed text like the GDPR, artificial intelligence requires building living governance and a flexible framework to serve our society.
First of all, strengthen safeguards where the risks become clear: deep fakes, critical and ideological biases but also censorship. Transparency on training data and model biases must become a selection criterion, not a regulatory constraint.
At the economic level, three measures would be enough to change trajectory. First, index the thresholds of the IA Act on a technical committee with annual review and not on the European legislative cycle. Then, direct a portion of public procurement towards European open source AI solutions, as a lever for indirect financing of R&D. Finally, address AI sovereignty through industrial consortia, not through laws or directives.
Collingridge’s dilemma is not inevitable; in the past we Europeans were able to build and regulate simultaneously, as in nuclear or aerospace.
Europe must choose: endure the innovation of others within a framework that it alone will have respected, or regain its place in the race. Without regulatory alignment and the ability to generate profitable growth, reinvestment of the margin generated in R&D and the creation of strategic interdependencies in our European ecosystem, we will remain vassals of empires.
