After two years dominated by the race for models, artificial intelligence is entering the infrastructure phase. The announcement by Mistral AI of a debt raising of 705 million euros, to finance a data center near Paris provides an illustration of this, with the specific objective of acquiring 13,800 GPUs from NVIDIA and internalizing a computing capacity hitherto largely dependent on American hyperscalers.
The initial project around the Bruyères-le-Châtel site would however have undergone significant adjustments, marked by the withdrawal of Fluidstackinitially announced as a partner, as well as by the abandonment of the data center project in the town of Bosquel, to which Mistral AI was also associated. This withdrawal contrasts with the visibility that Fluidstack enjoyed until recently. At the AI Action Summit, Emmanuel Macron had put forward an investment commitment of 10 billion euros carried by the company.
These adventures illustrate the concrete constraints faced by the deployment of large-scale computing infrastructures, at the crossroads of industrial, energy and territorial issues. AI infrastructure is no longer a simple technological subject and is becoming an object of development, subject to local arbitration as much as to market logic, with Fluidstack now focusing on the American market.
Until now, Mistral had established itself as a model player, positioned as a European alternative to the solutions developed by OpenAI or Google. With this operation, she takes a further step. By internalizing computing, it secures its costs, its deployment capacity and the governance of uses. The movement is consistent with the evolution of the market; as models become more widespread and standardized, the stakes shift towards computing resources.
For the first time, Mistral is using debt, via a banking consortium including BNP Paribas, Crédit Agricole CIB, HSBC and MUFG, in order to support the deployment of tangible, depreciable assets, whose profitability depends on the utilization rate.
The operation is also part of a European ambition to reduce dependence on American clouds, dominated by Microsoft, Amazon and Google, by building local capacities adapted to the requirements of businesses and administrations. Arthur Mensch, Managing Director of Mistral, sums up this logic: “Scaling our infrastructure in Europe is essential to empower our customers and to ensure that AI innovation and autonomy remains at the heart of Europe. »
Another parameter appears as a watermark: energy. Mistral is targeting 200 megawatts of capacity in Europe by 2027, with a second site announced in Sweden. At this scale, AI meets the constraints of heavy industries. The sizing of infrastructures now depends as much on access to electricity as on the availability of chips. The question of the cost, stability and decarbonization of this energy is becoming central, particularly in a European context marked by increasing industrial arbitration.
By committing to this path, Mistral is moving towards an integrated model, combining model development, infrastructure operation and service distribution, particularly to sensitive public stakeholders such as the French armed forces, with whom the startup has just signed a 3-year contract.
