Every June, Porte de Versailles fills up with more noise than any single human can process. This year it was louder than usual, because VivaTech turned ten. Four days, June 17 to 20, more than 200,000 visitors, over 15,000 startups, 1,155 speakers, and a cumulative social reach the organizers put north of five billion impressions. I went, I walked the halls until my feet protested, and I came back with a notebook full of things that matter for the people who actually have to run the systems once the conference glitter settles.
Let me be honest about what a show like this is. It is part trade fair, part theater, part diplomacy. This year had heads of state in the aisles, with Emmanuel Macron and Indian Prime Minister Narendra Modi walking the floor, India serving as AI Country Partner, and Germany named Country of the Year, the first European nation to get the title. That is the diplomacy. The theater was Jeff Bezos opening the week and Jensen Huang doing his now familiar tour de force on stage. The trade fair is everything in between. My job, as a CIO, is to ignore most of the theater and find the few signals worth acting on.
Here is what I found worth acting on.
The official tagline did half my job for me
VivaTech 2026 ran under a red thread that I wish I had coined myself: « AI: impact, not illusion. » For someone who spends his days separating what an enterprise can deploy from what makes a good demo, that framing is a gift. The whole edition leaned into it, with the program built around AI and productivity, cybersecurity and defense, greentech, space, and deeptech.
The reason this matters for a CIO is simple. The illusion phase of AI is expensive. It fills your roadmap with pilots that impress a steering committee and then quietly die because nobody designed for production, for cost, or for security. An event that puts « impact, not illusion » on the wall is, whether it means to or not, telling every IT leader in the room to go home and audit their own pilot graveyard. I took the hint.
Jensen Huang, and the part everyone misquotes
Huang returned to the VivaTech stage with the GTC Paris keynote, and his message was the same trinity he has been pressing all year: AI factories, agentic AI, and physical AI. The crowd loves the robots. I watched the room light up at the demonstrations, including a genuinely strange and wonderful live experience where a humanoid from Unitree was driven by human brain activity through non-invasive neuro-AI from HABS.
The robot is the headline. The infrastructure is the lesson. When Huang describes AI factories as places that manufacture intelligence the way power plants produce electricity, the part that should keep a CIO awake is not the metaphor, it is the implication. Agentic AI, systems that plan and execute multi-step work rather than answer a single prompt, changes how you think about identity, about access, about audit trails. An agent that acts on your behalf is a new kind of user, and most of our security models were not designed for users that never sleep and never log off. I did not need a robot to tell me that. But standing in front of one made the point land harder.
Sovereignty stopped being a slogan
The most useful shift this year was tonal. Past editions had a consumer flavor. This one was visibly more geopolitical, organized around AI, sovereignty, defense, cybersecurity, and energy. The CNRS built its entire stand around technological sovereignty across ecological transition, cybersecurity and AI, and industrial competitiveness. Yann LeCun showed up wearing a new hat as Executive Chairman of AMI Labs, and the conversation around European foundation models, with Arthur Mensch and Mistral in the picture, was less about national pride and more about where your data physically sits and who can reach it.
For a CIO of a French group operating across many countries, this is not abstract. Sovereignty translates into concrete questions I now have to answer for my board. Where does our AI inference run. Which models can we use in which jurisdictions. What is our exposure if a provider changes terms, or a regulator does. The show did not answer those questions for me. It did something more useful. It confirmed that I am not the only one asking them, and that the market is finally building options.
Cyber and AI, finally in the same sentence
The thing I have been saying internally for two years was, this year, simply the ambient assumption. Cybersecurity and AI were treated as one topic, not two. The threat side and the defense side of generative AI sat on the same panels. That alignment matters because it ends a tired internal argument. AI is not a project that the security team reviews at the end. The two are the same conversation, and the organizations that keep them separate will pay for it.
I left more convinced than before that the next real differentiator is not which model you pick. It is whether your governance, your security, and your data foundations are solid enough to let you change your mind about the model six months from now without rebuilding everything. The companies on stage with real deployments all had that boring layer in place. The ones with only demos did not.
Why this is worth a CIO’s time
People sometimes ask why a CIO with a global organization to run spends hours at a show. Fair question. The answer is that this is one of the few places where I can compress a year of market scanning into a long week. I get to see what 15,000 startups think the future looks like, watch which of those bets the big partners are validating with real money, and hear directly from the people building the infrastructure my teams will be living on. I can read a hundred analyst reports, or I can spend four days watching where the smart money and the smart engineers are actually pointing. The second is faster, and the conversations in the aisles are more honest than any press release.
What I bring home is not a shopping list. It is calibration. A clearer sense of which of my own bets are early, which are late, and which I should quietly stop defending. The startups remind me to stay humble about how fast the ground moves. The founding partners, the BNP Paribas and Orange and LVMH crowd, remind me what scale and constraint actually look like. And the tagline, « impact, not illusion, » is now stuck to my office wall as a filter for every AI proposal that crosses my desk.
VivaTech is ten years old. The organizers were careful to say this edition was not a celebration of the past nine years but the opening of a new decade. For once, the marketing line is also good advice. The job is not to admire the decade behind us. It is to design for the one ahead, deliberately, with the boring layers in place, so that when the next robot walks on stage we are ready to do something useful with it.
Now, back to the roadmap. The illusions can wait. The impact cannot.
