Key Takeaways<\/p>\n
- \n
- The threshold has been crossed.<\/strong> 54.5 percent of companies now use AI, up from 40.9 percent the previous year. A further 16 percent are planning adoption; 21.6 percent are still deliberating.<\/li>\n
- The Mittelstand is stuck in the middle.<\/strong> Large enterprises come in at 67.2 percent, small businesses at 51.2 percent, mid-sized firms at around 47 percent. The traditional Mittelstand is wedged between corporate machinery and the nimbleness of smaller operators.<\/li>\n
- Usage is not yet value.<\/strong> Manufacturing at 58.7 percent, services at 56.2 percent, retail at around 45 percent. The adoption rate tells you something is happening – not that it’s making a difference in core processes.<\/li>\n
- The bottleneck is rarely the model.<\/strong> It lives in governance, in the quality of the data foundation, and in whether anyone is actually empowered to say no to a use case.<\/li>\n<\/ul>\n<\/div>\n
Related:<\/span>AI in the Mittelstand: The bottleneck sits in legacy systems<\/a> \/<\/span> AI agents on the team: Why only one in nine pilots ever goes live<\/a><\/p>\n
What 54.5 percent actually says – and what it doesn’t<\/h2>\n
The figure comes from the ifo survey conducted in May 2026 and published in early June. It’s a positive signal: a year ago the number stood at 40.9 percent; now more than half of all companies are using AI. The upward shift is real, not statistical noise. Those who were still hesitating in 2024 have at least one foot in the door by 2026.<\/p>\n
What does AI usage mean in the ifo survey?<\/strong> It captures whether a company applies AI in its operations at all – from basic text generation to embedded process automation. It is a self-reported measure of breadth, not a gauge of depth or value creation. A team that occasionally uses a chatbot counts the same as one that has integrated AI into production control.<\/p>\n
That’s precisely the trap hiding inside every strategy deck. The 54.5 percent sounds like a breakthrough, but it describes adoption, not impact. In my own projects I’ve seen both: the company proud of its AI adoption rate with not a single model running in a process that moves money – and the one that barely mentions AI at all, yet has been using it to drive its pricing calculations for some time. The headline figure distinguishes neither.<\/p>\n
Why the Mittelstand is caught between two stools<\/h2>\n
The most interesting finding isn’t in the headline – it’s in the breakdown by company size. Large enterprises come in at 67.2 percent, small firms at 51.2 percent. Wedged between them, at around 47 percent, sit mid-sized businesses. The classic Mittelstand, in other words, lags not just behind the big corporates but behind the small fry too.<\/p>\n
\n\n67.2%<\/div>\nLarge enterprises using AI<\/div>\n<\/div>\n\n~47%<\/div>\nMid-sized businesses<\/div>\n<\/div>\n\n51.2%<\/div>\nSmall firms<\/div>\n<\/div>\n<\/div>\n\n ![\"Industriearbeiter](\"https:\/\/mybusinessfuture.com\/wp-content\/uploads\/2026\/06\/mbf-inline-260608-hero.jpg\")
In industry, 58.7 percent of companies use AI. In the Mittelstand, it often emerges on the shop floor rather than in the strategy deck. Photo: Sergey Sergeev \/ Pexels<\/figcaption><\/figure>\n There’s an uncomfortable reason for this. The large corporation has a dedicated AI team, a budget, and a data department that has been tidying up for years. The small firm has none of that apparatus – but it has short lines of communication: the owner tries out a tool on Friday afternoon and rolls it out on Monday. The mid-sized business has too little of either. Too big for the roll-up-your-sleeves fix, too small for its own AI unit. This sandwich position isn’t a failure – it’s structural.<\/p>\n
Anyone who ignores this and approaches the Mittelstand with corporate playbooks will reliably come unstuck. An AI governance framework built for a large enterprise will suffocate a 200-person operation. Conversely, the small firm’s spontaneous tool adoption falls apart the moment compliance requirements, works councils, and legacy systems enter the picture.<\/p>\n
The quiet leap from pilot to process<\/h2>\n
In almost every project, the first AI step looks promising. A pilot in a controlled environment, clean data, a motivated team, a presentable result for the steering committee. The pilot itself isn’t the problem. The problem is the meter beyond it – the transition into live operations, where the model connects to real systems, real master data, and real exceptions.<\/p>\n
That transition is where most of the projects I’ve accompanied stall. The ifo figure counts a pilot as usage already; but the business value only materialises in production. Between the two lies a stretch that never shows up in any showcase demo: who maintains the master data, who takes responsibility when the model makes a wrong call, who funds operations once the pilot budget is gone. These aren’t AI questions. They’re project management questions.<\/p>\n
The bottleneck is rarely the model<\/h2>\n
When an AI initiative stalls in a mid-sized company, the model is almost never to blame. Models are better than most of the processes people try to fit them into. The bottleneck sits in three places – and all three are uncomfortable, because none of them can be solved with technology.<\/p>\n
First: the data foundation. A model is only as good as the master data feeding it. If you’ve been maintaining your product and customer data across three systems with four different versions of the truth for the past decade, AI will mostly give you poorly sorted garbage – just faster. Second: governance. Not the heavyweight kind with committees and policy binders, but the simple question of who is allowed to stop a use case that sounds nice and delivers nothing. Third: accountability. As long as no one is named to own the success or failure of an AI deployment, every initiative remains an experiment that can be abandoned at any point without consequence.<\/p>\n
These are unglamorous problem areas – and that’s precisely where the glamorous initiatives tend to fall apart. The most honest discipline in any AI project isn’t prompting. It’s refusing to tell yourself, your team, and your stakeholders that usage already equals impact.<\/p>\n
What counts next quarter<\/h2>\n
For mid-sized businesses, the path forward means fewer use cases and greater depth. One use case that genuinely enters regular operations and measurably saves money or generates revenue is worth more than five pilot projects stuck in showcase mode. That demands an uncomfortable choice: identify a process where AI actually moves the needle financially, clean up the underlying master data, and name a person who owns the outcome.<\/p>\n
The 54.5 percent figure is good news – as long as you read it as a starting line, not a finish line. Anyone who checks it off as a target will find themselves a year from now with a high adoption rate and thin results to show for it. In the mid-market, the edge goes to the company that solves a single process with AI so thoroughly that the outcome becomes visible. Adoption rates alone won’t get you there.<\/p>\n
Frequently Asked Questions<\/h2>\n
\nHow many companies in Germany are using AI according to ifo?<\/strong><\/summary>\n
In May 2026, 54.5 percent of companies were using AI, up from 40.9 percent the previous year. A further 16 percent are planning to adopt it, while 21.6 percent are still in the discussion phase. The figures come from the ifo survey published in early June 2026.<\/p>\n<\/details>\n
\nWhy do mid-sized companies use less AI than small ones?<\/strong><\/summary>\n
Mid-sized businesses sit at around 47 percent, trailing both large corporations (67.2 percent) and small firms (51.2 percent). They lack the infrastructure of big companies – dedicated AI teams, budgets, and data departments – yet also miss out on the short decision cycles that give smaller firms their agility. This sandwich position is structural, not a sign of backwardness.<\/p>\n<\/details>\n
\nDoes using AI automatically translate into a productivity gain?<\/strong><\/summary>\n
No. The ifo figure measures adoption, not impact. An occasionally used chatbot counts the same as AI embedded in production control. Real business value only materialises when a model runs inside a core process and demonstrably saves money or generates revenue there.<\/p>\n<\/details>\n
\nWhat separates an AI pilot from full production use?<\/strong><\/summary>\n
A pilot runs in a controlled environment with clean data and a motivated team. Production use hooks into real systems, real master data, and real edge cases. Between the two lie the hard questions of data governance, accountability, and operating costs – the exact points where initiatives routinely stall.<\/p>\n<\/details>\n
\nWhere should a mid-sized company without an AI team start?<\/strong><\/summary>\n
With a single process where AI moves money in a measurable way – not with as many pilots as possible. Clean up the master data for that process and assign one person who owns the outcome. Depth beats breadth, especially when resources are tight.<\/p>\n<\/details>\n
Editor’s Reading Picks<\/h2>\n
- The Mittelstand is stuck in the middle.<\/strong> Large enterprises come in at 67.2 percent, small businesses at 51.2 percent, mid-sized firms at around 47 percent. The traditional Mittelstand is wedged between corporate machinery and the nimbleness of smaller operators.<\/li>\n