While the President of the Republic has just announced 109 billion euros in private investments in data centers dedicated to artificial intelligence in France, certain reactions have been skeptical. Unfortunately, we hear too often that Europe should avoid copying the madness of American quantities and favoring a logic “Small is beautiful“: a frugal artificial intelligence, exclusively based on small models, without large investments.
The intention is laudable, but the solution seems wrong to us. It is not by abandoning in the United States and China the means to train and use the most powerful AI models that we will manage to impose our own values. If we want to keep our political and economic independence for the decades to come, we need a massive investment effort in AI infrastructure.
Let us recall some figures. The best of large language models (LLM), on the technological border, see each year double their training cost. This cost mainly involves the mobilization of large datasets and ever more numerous and powerful graphic processors. However, while the United States has 70 % of the global calculation power available for artificial intelligence, the share of Europe is only … 4 %. According to our calculations, getting to the level of American capacities, in proportion to our respective GDPs, would imply for France to dedicate data centers specialized in AI on its territory 5 to 6 gigawatts by 2028, or a Investment from 250 to 300 billion euros by 2028.
Guarantee our technological independence
But then, why is such an investment necessary? First of all because AI is preparing to become a central infrastructure of our savings, as are electricity or internet. In the coming years, we will attend a massive automation of cognitive tasks which will concern all sectors, most obvious (sciences, education), most distant from digital (crafts, construction) via the most critical (defense, health). A transformation equivalent to a new industrial revolution, which could allow us to make important productivity gains. However, if we do not want to perpetuate the economic dropout started for twenty years with the United States, quickly climbing this new technological floor appears as essential to the growth of productivity and standard of living in Europe.
This also means that we cannot afford to leave it to foreign powers physical control. If we depend on the American servers and suppliers for training and the use of the best models of artificial intelligence, and they support a large part of our future economy, we will lose all negotiation power and be vulnerable to future Geopolitical blackouts. It is not a question of being self -sufficient – no one is – but of having the essential supports of the power of tomorrow. And although it involves difficult arbitrations, the investment effort necessary in AI infrastructure is achievable. Recent announcements are proof that lines are moving.
To go further, we will have to create special devices in order to guide private investment towards the creation of data centers, but also mobilize European savings available and accelerate the establishment of these infrastructures by an intelligent application of industrial regulations Including current construction deadlines in France exceed five years! This also involves giving priority to the electrical connection of these installations, and to continue our effort in favor of nuclear energy, decisive for their diet.
“” “Europe as France should not make the protection of individual data a totem which sometimes borders on the absurd and often lacks its goal»
Finally, such a objective requires a measured regulation of the sector, which avoids the platform maximalism. The protection of individual data has of course its place in intelligent regulation of the AI sector, and digital in general. Nevertheless, Europe as France should not make it a totem which sometimes borders on the absurd and often lacks its goal, for example when it leads to discourage innovation and the entry of start-ups on the continent, For the benefit of the largest existing players who alone can face an ever more sprawling regulatory charge.
Processors, the key
More than an investment in technological innovation, this project must be apprehended such as securing the crucial resource of the 21st century: computing power. The AI race is already launched and we cannot afford not to participate. The optimization of the model developed by the Chinese company Deepseek certainly impressed, but it must not make believe that a minimalist AI approach would be the way to follow. Especially since this model is based on large -scale infrastructure and investments. Deepseek as Google operate their own data centers, which allows them to innovate at all levels, in particular by going back into the value chain: low level of processors, data centers, specialized processors design, etc.
The major advances in AI often arise from the combination of software progress (algorithms, AI models) and hardware innovations (fleas, data centers). US operators are betting more than ever on continuous growth in investments required to postpone the technological border. Making the opposite bet in Europe would be disastrous, because in 2030, freedom will have a concrete price: that of processors.
* Raphaël Doan, senior official and writer
Antoine Levy, Assistant Professor at UC Berkeley Haas School of Business
Victor Storcom, machine learning engineer and scientific director of the series “Power of AI” published by the Grand Continent.