The Only Language Left

At VivaTech 2026, something quieter than a revolution happened. Every industry started speaking in the same voice. There is a particular kind of silence that falls when everyone in a room agrees. Not because the debate is over  but because the terms of the debate have quietly shifted, and no one noticed when it happened.

200,000 visitors, 165 nationalities, 15,000 startups, five billion cumulative social media impressions. The numbers are the largest in the event's ten-year history, and they say less than they appear to. What was striking about this edition was not its scale. It was its grammar.

Walk the floors of Paris Expo long enough and a pattern emerges  not in the products, but in the verbs. L'Oréal does not sell cosmetics anymore: it programs the future of beauty, in the words of its own press materials. Netflix does not offer a catalog: it negotiates attention. India's Prime Minister Narendra Modi, on the VivaTech main stage alongside Emmanuel Macron, did not talk about technology per se: he presented the MANAV framework a national AI governance model whose stated goal is a world where "technology serves humanity, not replaces it." Even the reassurance has been optimised.

Three sectors. Three continents. One grammatical structure: predict, personalise, reduce friction. Repeat.

This is not convergence in the usual sense  not industries merging, not products resembling each other. It is something subtler and more consequential: a shared cognitive framework spreading across sectors that have nothing else in common.

The numbers behind the paradox are worth sitting with. VivaTech's own official theme this year was "Artificial Intelligence: Impact, Not Illusion"  a headline that is either a provocation or an admission, depending on how honestly you read it. McKinsey data, cited in the Foxconn-Bull press release announcing a €120 million AI server manufacturing partnership in Angers, places Europe's share of global AI infrastructure capacity at under 5% in several key segments. Meanwhile, Mistral AI and Nvidia confirmed the launch of Mistral Compute a sovereign cloud platform built on 18,000 Grace Blackwell Superchips and Mistral's own 44-megawatt data center near Paris, financed with $830 million in debt.

The investment figures are staggering. The gap between investment and measurable output is staggering in the opposite direction. 80% of large companies now allocate budget to AI. 6% report a profit impact they can actually quantify. A technology being deployed at continental scale, by institutions that structurally, organisationally  do not yet know how to absorb it.

Consider what L'Oréal announced at VivaTech, and what it quietly implies. The beauty group, celebrating its tenth consecutive year at the event, unveiled a landmark partnership with OpenAI structured around what L'Oréal's Chief Digital Officer Asmita Dubey called the "11-minute paradox": data suggesting that as consumers engage with AI chatbots rather than search engines, the journey from discovery to purchase is getting longer, not shorter.

L'Oréal's response to this finding was to embed its products Lancôme, Kérastase, Maybelline  structurally inside ChatGPT's answer layer, feeding verified product data directly into the model's responses. Maybelline's virtual try-on tool, built on L'Oréal's ModiFace technology, will live inside ChatGPT conversations by this summer. OpenAI's life-sciences model GPT-Rosalind will help map skin microbiome data for La Roche-Posay product research.

The ambition is legible and rational from L'Oréal's perspective. The implication for the wider industry is harder to name. When a product is discovered not through editorial, not through culture, not through a friend's recommendation or a window display but through its structural position inside the answer layer of a dominant AI model something has changed about what a brand actually is. It has become, in part, an algorithmic placement.

This is the part VivaTech did not put on a slide.

The sovereignty conversation, which dominated the geopolitical sessions, deserves to be translated directly into creative terms. Thomas Wolf of Hugging Face argued that closed AI models represent a systemic risk citing documented cases of companies abruptly cut off from infrastructure they had come to depend on. Joe Tsai of Alibaba positioned open-weight models like Qwen as a hedge against a fragmented world where trust in foreign infrastructure cannot be assumed. The European Commission's Henna Virkkunen reframed Europe not as a regulator of AI but as an active builder pointing to Mistral, to the Franco-German industrial partnership, to the data center being built in Bruyères-le-Châtel as evidence that sovereignty was no longer rhetoric.

There is an irony embedded in Tsai's argument worth noting: Alibaba, as a Chinese company, operates under China's National Intelligence Law, which compels organisations to cooperate with state intelligence work on demand. The "sovereign" alternative he is selling carries its own dependencies. This is not an exception. It is the condition. Every model, every platform, every infrastructure you deploy is owned by someone with interests that are not identical to yours.

What survives the flood?

Adobe's Shantanu Narayen offered the clearest counter-argument on the main stage: AI will produce more content than any market can absorb. What will be scarce is not production it is meaning. Emotional intelligence, cultural specificity, the capacity to make someone feel something unexpected.

He is right, but the implication is harder than it sounds. Yann LeCun  who left Meta last year to co-found AMI Labs, backed by the largest seed round in European history made a quieter point from the same stage: the race to deployment is running faster than the race to understanding. The architecture, he has long argued, is wrong for what we are asking of it.

That is the actual question for creative industries: not whether AI can produce, but what it means when production is no longer the constraint.

VivaTech's closing image was, eight humanoid robots performed on the main stage movement, music, martial arts as the symbolic send-off of the 10th anniversary edition. A spectacle of optimised motion. Beautiful, precise, and entirely predictable once you understood the choreography.

The question the event left open is editorial, not technological: how much of human experience should be made frictionless? What do we lose when the detour is engineered out?

Those questions have answers. They will be given, implicitly, by every creative decision made in the next decade. The industries best positioned to give good ones are those that understand what they are actually protecting not a process, not a workflow, but a particular relationship to uncertainty that has always been where original work comes from.

The optimisation consensus has no use for uncertainty. That, right now, is the most valuable thing a creative professional can possess.