Europe wants to develop its uses of artificial intelligence, but the necessary resources remain a central issue. Competition involves processors, data centres, electricity and networks. The quality of the models is not sufficient to ensure their deployment.
At the GTC conference in Berlin, Jensen Huang, head of Nvidia, highlighted modular facilities for data centres. He promises to reduce certain preparation times. However, these modules do not remove energy requirements or connection work.
Artificial intelligence: the gap also relates to calculation
According to estimates published on 22 October, Europe would have about 5% of the world’s computing capacity devoted to artificial intelligence. The United States share would reach about 75 per cent.
This comparison covers the infrastructure available. It does not measure the quality of European research. Nor does it cover all forms of computer computing. This distinction avoids turning a material indicator into a general ranking of scientific capabilities.
European companies need resources to train and exploit models. This requires advanced processors, high-performance memory and fast connections. Facilities must also cool equipment and maintain its operation. If these resources remain scarce or costly, local development becomes more difficult.
The lack of capacity can then increase the use of external platforms. Companies also depend on the commercial conditions set by their suppliers. The problem, therefore, is not only the availability of machines. It concerns the conditions for access to and control of the calculation.
Modules to shorten part of the site
The system presented by Nvidia is based on a modular infrastructure prepared with partners. Units are transported to an already developed site. Their shape is close to that of a container.
Jensen Huang claims that this organization could reduce the preparation of a site from several months to a few days. This leadership promise does not mean that a complete data centre becomes operational within a few days.
The reception area must have sufficient power supply. It also requires cooling, connections, authorizations and connection to the network. The industrial preparation of the modules does not remove these steps. It covers part of the installation, not all the conditions necessary for its operation.
However, standardization can facilitate the assembly of equipment. It also allows for a gradual increase in capacity. An organization can add units according to its needs. In Nvidia’s presentation, this advantage is particularly relevant to actors who wish to develop their infrastructure in stages.
Industrial uses increase demand
Nvidia also presents a passage to what it calls physical artificial intelligence. These are applications related to factories, machines, cars and robots. These uses go beyond the interfaces that answer questions. They place data processing in contact with industrial equipment.
Jensen Huang mentions Siemens, Mercedes-Benz and Audi among its European partners. The group also claims to have invested ten billion dollars in start-ups and European partners in the previous year.
These applications include maintenance, design and planning. Their dissemination can increase the need for data processing and circulation. It also requires reliable connections. Industrial uses thus create their own infrastructure requirements.
Safety and reliability also occupy a special place. An application for a robot or industrial system does not meet the same constraints as a conversation tool. These equipment require tests adapted to their operation and the risks of their use.
Dependence on how software works
Access to processors is part of the problem. Control of data, updates and operation is also important. Palantir’s assurances about his Maven military system illustrate this other dimension of dependency.
In statements reported on 22 October, Palantir sought to reassure Europeans about his Maven military system. The company claims that it does not contain a device to disable it remotely. It also submits that customers retain operational control of the purchased software.
These guarantees are advanced by the supplier. According to the information published on that date, France and Germany have not adopted this system for their armies. Their concerns concern American dependency.
This issue is in line with that of artificial intelligence infrastructures. How do users control the tools they depend on? It also covers data, updates and the availability of alternatives when supply conditions change.
This reading of digital sovereignty does not necessarily require the local manufacture of each component. She asked to know the dependencies and their practical consequences. It concerns continuity of operation as well as ownership of equipment. The actual control is therefore not limited to the physical location of the servers.
Electricity and skills remain essential
The development of European capabilities depends on several material and human conditions. Public and private investment must be accompanied by appropriate electricity networks. It is also necessary to train the engineers needed for installation and operation. The purchase of equipment does not address these needs.
Clear data rules are also needed. Competition is important to ensure that resources are not concentrated between a few players. Access to calculation concerns enterprises, but also universities. Their capacity to use the facilities depends on the usefulness of the investment.
The increase in infrastructure finally raises the question of their energy costs and their environmental impact. These constraints must be integrated into deployment. Faster units to install do not dispense with this examination.
The success of an investment is therefore measured by the uses it makes possible, rather than by the number of servers alone. Organizations must be able to develop their solutions, exploit them and examine them. They must also retain options to their suppliers. For the modular facilities presented, the costs, connections and results of actual deployments remain to be documented.
Sources
Al Joumhouria, November 2, 2026, page 10.
Comments