gensai.aiAt VivaTech, everyone expected spectacular demos: humanoids, drones, immersive experiences, ever more powerful models. All of that was there for the 10th edition of the Paris show (17-20 June 2026), but another thread ran through the aisles and the keynotes — quieter, and far more structural.
Sovereignty in artificial intelligence imposed itself as the central question of this edition. Not raw model performance, but control: who owns the technology, where the data runs, and which models companies and governments depend on.
Why this shift, and what does it change concretely for French companies investing in AI?
The answer comes down to one idea: sovereignty is no longer an abstract political debate — it has become an architecture criterion for any serious AI project; especially at a time when the US government can decide to cut other countries off from an AI model deemed too powerful.
Why did this 10th edition put sovereignty centre stage?
VivaTech 2026 made European technological sovereignty its backdrop as much as its subject. Germany, country of the year, came with nearly 200 startups and a dedicated pavilion. Europe as a whole occupied the floor — a way of responding, in presence, to a dependence often denounced but rarely addressed.
The context is not neutral. A few days before the show, the US administration ordered a major AI company to suspend access to its most powerful models for "any foreign national", in the name of national security. The signal was received loud and clear on the European side: a tool you can be barred from using overnight is not a tool to build a strategy on.
Digital sovereignty then stops being a slogan and becomes a matter of business continuity.
Yann LeCun's plea for sovereign AI
On the main stage, Yann LeCun delivered a clear plea for sovereign AI for every state. The French researcher, a global figure in AI, asserted that governments are right to want to secure their sovereignty in this field.
His argument goes beyond geopolitics. In his view, before long most of our information consumption will flow through AI assistants. The plurality of these assistants then becomes as important as the plurality of the media: having a single source, controlled by a single player, would mean accepting an unprecedented informational dependence.
Hence his defence of an open foundation model, on which anyone could build an assistant adapted to their language, culture and value system. That is the whole point of the Tapestry project, launched with the AI Alliance coalition to develop open, sovereign AI. LeCun contrasted this vision with the "control" exercised by a handful of players who would reserve the most capable models for a chosen few.
One may agree more or less with his position on open source. But the underlying observation is hard to dispute: autonomy requires the ability to choose, adapt and host your own models.
Is AI sovereignty a form of retreat?
AI sovereignty is neither national retrenchment nor a rejection of foreign technology. It is the ability to keep control over three variables: where data is stored and processed, which models are used, and under which rules.
For a French company, the question is very concrete. An AI project can rely on open models hosted in Europe, on a RAG (Retrieval-Augmented Generation) architecture that keeps business data under control, or on building blocks compliant with GDPR and the AI Act. These are not administrative constraints: they are the conditions of trust.
This is the line Gensai has defended from the start. Frugal, responsible AI, where the question "where does the data run, on which models?" counts as much as headline performance: the right level of intelligence applied to the right problem, in the right place.
What do the booths say about the AI market?
Beyond the conferences, the booths also tell you where the market is heading. VivaTech 2026 confirmed the rise of immersive, interactive experiences, where generative AI no longer just answers but stages, personalises and brings a brand to life.
Large groups have understood this well. Right at the show entrance, the prominently featured L'Oréal booth offered several interactive installations, including two applications Gensai had the opportunity to contribute to.
Coming across these projects between two conferences is another way of measuring a trend: AI applied to customer experience is no longer a demonstrator — it is a product showcase.
Sovereignty: a criterion, not a slogan
VivaTech 2026 confirmed a shift from what AI can do to who owns it and where it runs. Government announcements, Yann LeCun's plea and the place given to Europe all point in the same direction.
For decision-makers, the point is not to tick the "sovereignty" box on principle. The point is to make it an architecture criterion from the design stage of a project: which models, which hosting, which compliance, which reversibility. In enterprise AI, technical sophistication never replaces strategic clarity — and sovereignty is now at the heart of that clarity.