In Paris, Jensen Huang has delivered an industrial and distributed vision of AI, in which Europe now plays a central role. A change of scale supported by massive investments in infrastructure, an opening to the local ecosystem, and a rereading of the data center which has become “factory in Tokens”.
We had to listen carefully to inflections, even more than the announcements. During the long -awaited and very successful Keynote of Jensen Huang, CEO of Nvidia, at the living room Vivatech, the manager has continued to reposition Europe as a strategic production territory for the era of distributed artificial intelligence. Starting with this introductory declaration: “Europe has now awakened to the importance of these ai factories.”
These “AI Factories”, which Nvidia designs as installations capable of generating billions of tokens via specialized architectures, have become the new industrial standard of AI. Far from the traditional data centers, they are no longer defined by storage but by intensive calculation capacity dedicated to inference.
“We will multiply by ten the capacity of calculation AI in Europe over the next two years, with more than 20 factories in tokens already in preparation”said Jensen Huang. Some of them will exceed the threshold of Gigawatt, which would place them among the most powerful calculation infrastructure in the world.
An architecture thought to reason, not just execute
At the heart of this rise in power, architecture Grace Blackwellpresented as the first machine specifically designed for the era of reasoning. The figures make it vertigo: 130 terabits per second of bandwidth, more than the traffic peak in the entire world Internet, concentrated in a copper column weighing 60 kilos. “This is How You Shrink the Internet Into 60 Pounds”illustrated Huang, pointing to the Nvlink spine connecting 144 Blackwell chips in an entirely non -blocking topology.
But beyond the figures, it was the very purpose of the architecture that has changed. Nvidia no longer sells only cards or clusters, but a “cognitive factory”. Each bay is thought of as a tokens production unit with high value, where the marginal cost is now assessed on the scale of the generated token, either from the request.
A sovereignty infrastructure … built by Nvidia
The transition to these factories is not a purely technological logic. It is part of an infrastructure policy, on the states level. “These data centers are no longer data centers. These are tokens generation factories. And they will become an integral part of the infrastructure of each country”said Huang. It is in this context that Nvidia multiplies partnerships with European industrialists.
In France, the partnership with Schneider Electric is highlighted as a co-construction model. The objective is to build complete digital twins of the AI factories, allowing to optimize their design and exploitation before a part is physically assembled. The CEO insisted: “These are factories are so exisseed – 50 Billion, sometimes 100 Billion in the future – that ifn’t optimal, the cost to the factory owner will be incredible.”
Emmanuel Macron grabs the ball to the leap
In the early evening, the president Emmanuel Macron rose to the main scene of the show, alongside Jensen Huang and Arthur Mensch, founder of Mistral AI. Together, they formalized the creation of Mistral Computea sovereign ia cloud platform, hosted in Europe, equipped with 18,000 blackwell superpudes.
“The partnership announced today between Nvidia and Mistral is in my eyes historic”said the French president, stressing his role in European digital sovereignty. Several French companies, Orange, BNP Paribas, SNCF, have already joined the project. “Not only will this partnership increase your sovereignty, but it will allow you to go even further”he launched to the address of entrepreneurs and public present.
The Head of State also formulated the wish to install a Electronic flea production capacity on French territory.
Mistral, Siemens, BMW … an assumed European dynamic
Nvidia is now anchored in a European strategy based on deep and differentiated industrial partnerships according to the country. Each alliance illustrates a facet of the current switch, AI is no longer a peripheral tool, but a structural transformation brick for large groups on the continent.
In France, the agreement with Mistral ai crystallizes this evolution. Launching Mistral Computesovereign cloud platform with 18,000 Blackwell chips, should allow the startup to offer secure access to its open source models, while offering a hosting base for the European IA ecosystem. The project, hosted in Europe, is explicitly intended for startups, companies and institutions wishing to escape excessive dependence on the American hyperscalrs.
Besides Mistral, several major French accounts like BNP Paribas,, Orange,, SNCF, announced their membership in the project. Emmanuel Macron, when he speaks to Vivatech, encouraged other actors to join this “alliance”, qualifying the initiative of “Game Change” for European digital sovereignty.
In Germany, collaboration with Siemens Give body to a new generation industrial AI. The stake is not limited to modeling or simulation and it is a question of integrating AI agents into the design chains, predictive maintenance processes, and operational planning.
Bmwfor its part, works with NVIDIA on the complete digitization of its industrial sites via Omniverse3D simulation platform and real -time collaboration. Its Debrecen factory, in Hungary, was fully designed in a digital twin even before laid the first stone. Objective: optimize logistics flows, anticipate bottlenecks, and allow fluid interaction with robotic agents.
To these examples are added Kion Group (logistics), SNCF (digital infrastructure twins), Bouygues (construction), Stmicroelectronics (semiconductors), or Prologis (Data Centers). So many collaborations that testify to a shared will: to integrate AI from the design of the systems, and not as a secondary optimization layer.
This European network is based on the conviction that the industrialization of AI is not a homogeneous process. It requires tailor -made architectures, adapted infrastructure and an ability to dialogue with the sectoral specificities of each actor. NVIDIA does not impose a unique model, but offers an open, interoperable, and extensible technological stack, on which European companies can build.
IA campus, megawatts and state strategy
The rise of Nvidia in Europe is part of a broader state dynamic, embodied by Emmanuel Macron’s strategy in terms of artificial intelligence. The French president for several months defends a sovereign AI approach for several months, based on both massive investments and a logic of critical infrastructure.
It is in this spirit that a IA campus project in Île-de-Francepresented as the largest in Europe. The initiative combines The sovereign fund Émirati MGX,, Bpifrance,, Mistral ai And Nvidiawhich will ensure operational governance. At the top Choose FranceEmmanuel Macron had detailed the contours with its construction from 2026, funding between 30 and 50 billion eurosand the ambition to cover the entire life cycle of artificial intelligence.
The campus will include several additional bricks:
-
- A giant data center of an announced capacity of 1.4 Gigawattor the energy equivalent of an EPR type nuclear reactor;
- Training centers intended to train engineers, developers, technicians and operators of the IA economy;
- Applied research laboratoriesin connection with startups, industrialists and public laboratories.
This emblematic project is added to a series of more dispersed announcements. Since the beginning of 2024, more than 126 billion euros in private investments have been argued towards AI -related projects in France, according to figures from the Élysée. These amounts include:
-
- The project Mistral Compute and its 18,000 Blackwell fleas;
- The 20 billion euros mobilized by Brookfield for a data center in Cambrai (north);
- The 6.4 billion euros that PrologisAmerican actor in logistical real estate, plans to devote to calculation centers.
To structure this wave of investments, the state has identified 65 “conducive” sites for the construction of data centersdistributed in all French regions: Hauts-de-France, Auvergne-Rhône-Alpes, Provence-Alpes-Côte d’Azur, etc. A network designed to meet energy, land and connectivity constraints, in a context where energy density becomes a strategic factor.
A diplomacy of silicon on the move, Jensen Huang at his head.
The initiative is part of a change of nature of Nvidia. More than a GPU supplier, the firm is now positioned as an infrastructure power. “These ai factories are now part of a country infrastructure, which is which you see me running around the World Talking to Heads of States”said Huang.
With an adaptable architecture, compatible with public clouds as well as with on-premise solutions, Nvidia weaves its canvas. In France, the co-construction posture adopted by Emmanuel Macron reinforces this logic of industrial anchoring.
An intervention praised by an audience from around the world
Nvidia seems to have understood that the era of the AI will not be played only on the computing power, but on its distribution, its vertical integration and its location. By betting on Europe as a field of production and experimentation, and by leaving industrial and state partners, the company becomes the catalyst for a new form of digital sovereignty. For Emmanuel Macron, this strategy marks a step in the redefinition of French technological power to which it will now be necessary to adapt.
Benchmarks
- Mistral Compute: sovereign cloud planned for 2026, 18,000 Blackwell fleas
- Investments IA France: € 126 billion announced in 2024-2025
- IA Campus Ile -de -France: Piloted by MGX, NVIDIA, Mistral, Bpifrance; 1.4 GW for power, training centers and R&D
- Flagship architecture: Grace Blackwell GB200, 130 to/s, 1.2 million components, entirely liquid-coolded
- Partners: Siemens, BMW, Schneider Electric, Stmicroelectronics, SNCF
- NVIDIA strategy: Cuda-Q (Quantum), Nemotron (Open-Source Optimized LLM), Omniverse (digital twins), DGX Lepton (Unified Cloud)
- NVIDIA Foundation: Founded in 1993, based in Santa Clara, capitalization> $ 3,000 billion (June 2025)
