Only 7% of French manufacturers use AI: a sector signal worth taking seriously

Only 7% of French manufacturers currently use artificial intelligence, according to data reported by L'Usine Nouvelle in 2026. This figure stands in sharp contrast to the adoption levels seen in services or retail, where generative AI usage now exceeds, by many measures, a third of companies. French industry is not lagging on digital topics overall, but it carries a specific, well-documented gap on artificial intelligence.

This lag has less to do with cultural reluctance than with a structural obstacle widely acknowledged by manufacturers themselves, access to usable data. A sales or administrative department typically has already-digital data, a CRM, invoices, emails. A production site, by contrast, generates data scattered across heterogeneous sensors, maintenance logs sometimes still handwritten, and systems that do not communicate with each other. Before even choosing an AI tool, an industrial SMB often has to do structuring work that its service-sector counterparts do not face to the same extent.

This finding particularly affects industrial SMBs, whose internal resources for this structuring work are, by nature, more limited than those of large groups that can dedicate an entire team to the task. At Global Industrie 2026, one of the sector's major trade events, AI was not the leading investment topic for exhibiting manufacturers, the technology is growing in discussions without yet becoming a clearly stated budget priority.

This context should not lead to the conclusion that AI is out of reach for an industrial SMB. The most accessible entry points do not require a full overhaul of the production line. Predictive maintenance applied to a single line, rather than the whole site, automated analysis of technical documents or quality sheets, and help drafting technical quotes are realistic first uses that do not require having first solved every data structuring problem the company faces.

The lack of structured data is therefore not a permanent obstacle, it sets a clear prerequisite, often underestimated in the timeline of a first project. Structuring already-available data, failure histories, sensor readings, product sheets, before launching an AI pilot avoids the most common pitfall seen on the ground, a technically well-designed project that cannot be fed data reliable enough to produce usable results.

This finding also reminds industrial SMBs that they do not need to copy AI uses built for services. A tool designed for marketing copy or customer service adds little value on a production line, which partly explains why manufacturers have been more cautious, the supply of tools genuinely suited to their constraints is also still young.

This sector lag, real as it is, is not necessarily bad news for industrial SMBs deciding to act now. A sector where only 7% of players have taken the leap offers, to those who structure their data and launch a first targeted use, a competitive edge that already-saturated AI sectors no longer offer. The gap with services is not a fate to endure, it is a window to use before it closes.

This is exactly what the Inventory step of the IMPACT method should account for at an industrial SMB, a specific audit of production data availability and quality, an essential prerequisite before any serious industrial AI project. A five-business-day diagnostic assesses precisely how mature your production data is for a realistic first AI use.