AI must leave the race for more and more: develop more sober, sovereign and useful models, designed to strengthen humans rather than replace them.
Artificial intelligence is advancing at such a pace that each innovation seems immediately overtaken by the next. As if the only possible direction was that of gigantism: more parameters, more energy consumed, more data collected, more computing power. This race cannot be a sustainable strategy. Because if AI transforms our societies, it must also learn to transform itself.
We have reached the limits of “ever bigger”
For five years, the generative AI industry has been founded on a simple principle: what works today will be even better if it multiplies its size tenfold. This logic has enabled spectacular progress. But it leads us to a paradox: to make models more intelligent, we also make them more inaccessible.
Costs are exploding. Electricity needs are becoming dizzying. Systemic risks are increasing. And concrete uses sometimes struggle to keep up with this technological inflation. Businesses don’t want AI that knows everything: they want AI that knows exactly what they need.
The future of AI will be sober or not
The great transformation to come will not be that of the most massive models, but that of the most intelligent architectures. Not those who consume the most, but those who consume better. Not those who remember everything, but those who know how to forget what is irrelevant.
True mastery will arrive when we know how to design models that are as fine as they are powerful, capable of operating on accessible infrastructures, in secure environments, and with a sustainable energy footprint.
Sovereignty cannot be decreed: it is constructed
In a world where data is the raw material of the 21st century, depending entirely on models designed outside our borders is a risk that few sectors can afford. The question is not ideological: it is economic, cultural, strategic and, increasingly, civilizational; Building autonomous AI capacity does not mean refusing international cooperation. This allows us to keep control of our cultural heritage gathered within the major language models.
Europe has a path of its own
Contrary to what some imagine, Europe is not condemned to run behind the United States or China. It is charting its own path — a path that is more like it: diversity of cultures integrated into AI models supported by sober, reliable, auditable, safe technologies.
The AI of tomorrow will be local, specialized and distributed
The idea of a universal supermodel is attractive, but illusory. What will naturally emerge is an ecosystem where different types of models will coexist, each adapted to its purpose. In this landscape, engineering finesse will take precedence over raw size.
The horizon: an AI that augments, not replaces
The goal of AI is not to replace humans but to augment them. Replacing AI creates economic stress. Increasing AI creates value. The essential question is not: how far can AI go? The real question is: what do we want it to do for us?
We have the choice of the future we want to build
We have the possibility of deciding that AI will not be an excessive technology, but a precision technology respectful of societal, economic, political and environmental balances. AI is not destined to escape us: it is up to us to decide what it becomes.
