Flemish companies doubled their AI use in two years

The numbers come from the Flemish AI barometer 2025, published by the Flemish department of economy, science and innovation on 11 March 2026. In 2023, 32.1 percent of companies used at least one AI technology. In 2025 it was 58.8 percent. Generative AI is at 47.6 percent and text-analysis software at 41.2 percent. In almost half of the companies (48.9 percent) the use grows from the bottom up, often because individual employees start on their own.

The press release says use in the Flemish government has also doubled. Civil servants using Microsoft Copilot save 16 to 17 minutes a day on average. There are 72 AI projects running in Flemish government services.

The Flemish AI policy plan started in July 2019 and was renewed in March 2024. Its website now gives about 36 million euro a year. About 9.5 million of that goes to helping companies adopt AI through VLAIO, the Flemish agency for innovation and entrepreneurship. Another 12 million a year or so goes to the digitalisation of small businesses. The barometer press release mentions a larger amount. It speaks of more than 150 million euro a year so that companies and public services use AI responsibly. It does not say what that includes.

The plan, the barometer and the employers' federation Voka all stress competitiveness and speed of adoption. Responsible use comes as a separate track next to it. Minister-president Matthias Diependaele puts it like this: “We should not be afraid of AI as a society. On the contrary. We should embrace it.”

At the federal level, the High Council for Employment (Hoge Raad voor de Werkgelegenheid) published a report on AI and the Belgian labour market in February 2026. It found that 33 percent of Belgian workers used generative AI at work in 2025, against a European average of 21 percent. As far as I know, it is the closest thing to a Belgian institutional assessment of what AI does to jobs. More in the jobs question.

Almost six in ten Flemish companies now run at least one AI system. The models behind those systems come from the same few labs that spent 2026 working through tens of thousands of cases of their models doing things nobody asked for.

Those two facts are discussed in different places. In Flanders the debate is about skills, productivity and not falling behind. The incidents are discussed in English, on American websites, by people who already work in the field. I know of no Flemish or Belgian body whose task is to connect the two. The closest I found was a hearing on a concept note on AI in the WEWIS committee of the Flemish Parliament on 17 September 2026. Valeria Pulignano (KU Leuven) and Vincent Ginis (VUB) were among the speakers. I have only read the announcement, not what was said.

That gap is the reason I started this dossier. Flemish companies take up AI quickly. Almost everything written about the incidents appears in English.

There is also a counterargument, and I find it convincing. Adoption is also how a region builds its own industry. The productivity gains look real. In the barometer, 21.4 percent of companies see higher productivity among low-skilled workers thanks to AI. And a Flanders that waits buys everything from elsewhere and has even less say. The same reasoning is behind the European argument, one level up.

The barometer does not say what Flemish companies and public services run, or on whose models. It measures how many use AI. How vulnerable they are when those models go wrong, it does not measure.

The adoption figures do not match. The Flemish barometer has 58.8 percent of companies for 2025. The High Council for Employment has almost 35 percent of Belgian companies for the same year. These are probably different surveys with different definitions. I have not checked where the difference comes from.

I wrote earlier that the policy plan costs 32 million euro a year, with 15 million through VLAIO. Those may be the 2019 figures. The current website of the plan gives about 36 million, and I could not find the older numbers.

I read the press release on the barometer, not the full report. Of the High Council report I read the summary and the recommendations, not the whole document.