Since spring 2026, we spent weeks in the field. Long-standing clients, but also companies simply curious about AI, wanting to understand before deciding. We came back with an impressive volume of notes, and three findings that settled our trajectory.
First finding: everyone is already using AI. Everywhere, in every company we visited, without exception. But often with no regulation whatsoever: well-intentioned employees, from private sessions, on personal accounts, using whatever consumer tool is available. And almost always without realising a simple fact: what leaves for inference servers, in Europe or the United States, never comes back. A contract excerpt pasted into a chat window, a client list submitted for sorting, a strategy summarised to generate a presentation: each of these pieces of information has left the company, permanently, with no receipt and no way back.
Second finding: nobody understands the jargon, and nobody bothers to explain it. Model, inference, RAG, agent, context window: behind these terms lie actions that are often very simple. But confusion reigns, over the words as much as over the very structure of an AI. And this confusion is not always innocent: a client who does not understand what they are buying is a client who cannot compare, challenge, or leave. We saw too many decisions made in a fog to believe that fog is accidental.
AI is still in its infancy, and yet it is already a backbone: everything flows through it, data, strategy, know-how, clients.
Third finding, the most striking: the lucidity is there. Among all the people we spoke with, a real awareness that AI is still in its very early days, its first cries. And yet, even at this stage, it is already becoming, for many, a new backbone of the company. What flows through it is not trivial: data, strategy, know-how, client names and all their specificities. This flow is not mere technical traffic. It is the company's capital, in its most concentrated form.
The decision these visits imposed
From these weeks in the field, we drew one conclusion, and one only: we will do nothing but private AI, on hardware owned by the client. Not as a premium option in a catalogue, not as a variant for demanding clients. As our sole model.
This choice settles two questions at once. Confidentiality first: it is no longer a topic for discussion, a contract paragraph or a vendor promise. It is extreme by design, because nothing leaves. The model runs on a machine that is yours, at your premises or on a dedicated server you own: the question "where does my data go?" simply no longer arises, because it goes nowhere.
The business model next: you step out of pay-per-use billing. No meter running with every question asked, no subscription whose price tracks your dependency. One server, one investment, one piece of infrastructure on your balance sheet that gains in value as your teams adopt it. Sovereignty is total, and it shows even in the accounting.
And if we can offer this model at a price that holds up, it is no magic trick. It is our sister company, FSYS Informatique, whose name is no accident: their business is to supply servers, and nothing else. They are what allows us to build high-performance server configurations without spending a fortune, sized precisely for each company's actual workload. Two houses, two trades, one single chain: the hardware on one side, the intelligence on the other, and your data never leaving either.
We did not invent anything this spring. We simply listened, a great deal, and took seriously what we heard. If AI is destined to become the backbone of your company, then it must belong to you. Everything else follows from that.