TÜV Study 2026:
AI Upskilling
in Germany Doubles

The numbers sound encouraging at first: in 2026, 27 percent of German companies trained their employees in AI - more than doubling their AI upskilling rate. In 2024 it still stood at 12 percent. The TÜV Weiterbildungsstudie 2026, conducted by Forsa between January and March 2026 among 500 training managers, CEOs, and board members from companies with 20 or more employees, documents remarkable momentum.

Look closer, though, and the real problem emerges: while 56 percent of the companies surveyed already use generative AI tools such as ChatGPT, Microsoft Copilot, or Google Gemini in their daily work, only 27 percent have trained their workforce accordingly with targeted AI upskilling. The training rate lags almost 30 percentage points behind the reality of usage. That's not a development gap - it's a structural risk.

It's also worth looking at the quality of the training itself: not every AI training program is equally valuable. The study distinguishes between tool onboarding (which buttons do I press?) and competence-oriented training (how does the technology fundamentally work, what risks am I taking?). The former is quick and cheap - but it doesn't build lasting competence in the sense the EU AI Act intends. The latter requires more investment but has lasting value.

From 12 to 27 percent: what the TÜV Study 2026 really says - and what it leaves out

The doubling of the training rate from 12 to 27 percent is real and significant. It shows that AI upskilling has arrived in the German economy - at least as a topic. But the study also reveals a troubling self-perception among companies: 45 percent of respondents see no need to expand AI competencies in their organization. That's the majority of the companies not yet training - and they're not passive about it, but actively convinced they don't need to act.

56 percent of German companies already use generative AI tools in their daily work. Only 27 percent have trained their employees for it. This gap is no accident - it's a structural failure.

Company size adds another layer: large companies with 250 or more employees reach an AI training rate of 49 percent. Mid-sized companies with 50 to 249 employees reach 32 percent. Small companies with 20 to 49 employees, by contrast, reach only 21 percent. Since the German Mittelstand is structurally made up of exactly these small and mid-sized businesses, the economy-wide risk exposure is substantial.

In parallel, the Bitkom AI Study 2026, surveying 604 companies, found that 41 percent of German companies actively use AI - up from 17 percent in 2024. Another 48 percent plan to adopt it. AI usage is accelerating; competence development isn't keeping pace. The Bitkom AI Barometer 2026 also shows that only 25 percent of German SMEs use at least one AI application daily - and 53 percent cite a lack of internal AI competence as their biggest hurdle to AI adoption.

The study also shows: companies that already ran initial AI training in 2024 report competence problems and misapplications significantly less often in 2026 than companies only now getting started. Competence built early has a compounding effect - it grows with every new use case and every new generation of tools. Late starters not only have to catch up on the basics, they also have to keep pace with ongoing change at the same time.

Also notable is the choice of training format: the TÜV study shows that digital formats (e-learning, videos, webinars) are used significantly more often than in-person training. That's practical news for the Mittelstand: AI upskilling doesn't have to mean full-day absences or expensive travel costs. Asynchronous learning formats make it possible to train employees in small units - 30 to 60 minutes per module - without disrupting operations.

The gap between AI usage and AI competence: 56 percent use it, 27 percent train for it

The 29-percentage-point training gap between AI usage and AI qualification isn't just a statistical curiosity - it has concrete operational consequences. When employees use AI tools without structured knowledge, errors emerge that are hard to undo: misinterpreted AI outputs, uncritically accepted hallucinations, data protection violations from careless handling of company data in cloud services.

The TÜV study also breaks results down by industry. In information and communication technology, 61 percent of companies already train actively for AI - far above the overall average. In manufacturing, by contrast, it's only 19 percent, in retail 18 percent, and in healthcare 16 percent. It's precisely in the sectors where AI-driven processes will become especially relevant in the future - logistics, production, care - that the catch-up need is greatest.

The question of training content is especially revealing. According to the TÜV study, the training conducted so far focuses mostly on operating specific tools. A fundamental understanding of how AI works, critically questioning AI outputs, and the ethical dimensions of AI use are addressed far less often. That produces technically competent users but not users capable of sound judgment - and that's exactly what becomes a problem once EU AI Act Article 4 takes effect on August 2, 2026.

EU AI Act Article 4: from August 2, 2026, AI competence is no longer optional

Article 4 of the EU AI Act obliges providers and operators of AI systems to ensure their staff have sufficient AI competence. The EU AI Act took effect on August 1, 2024; the Article 4 obligation applies 24 months after entry into force, meaning from August 2, 2026. The deadline is no longer a vague future prospect - it's reality in eight weeks.

What does this mean concretely? The term "sufficient AI competence" is deliberately open. The European Commission has made clear that it's a context-dependent requirement: the competence needed depends on the type of AI systems in use, the risk level of the use cases, and the role of the staff involved. A sales employee using ChatGPT to draft emails needs different competencies than a developer operating a high-risk AI system within the meaning of Annex III of the EU AI Act.

In practice, this means for SMEs: they must be able to document that they've actively taken measures to develop competence. A structured training program, verifiable learning objectives, and documented participation aren't bureaucracy - they're the basis for compliance. Anyone who can't produce documentation in the event of a review by national supervisory authorities risks fines and reputational damage.

"From August 2, 2026, EU AI Act Article 4 applies: companies must demonstrably ensure their employees' AI competence. 73 percent of German companies are not prepared for this."

The TÜV Study 2026 makes the scale of this compliance risk clear: if only 27 percent of companies have conducted any AI training at all, and a significant share of that is merely tool onboarding without structured competence-building, then an estimated fewer than 20 percent of companies in Germany today are likely genuinely positioned to comply with Article 4.

Why 45 percent of companies believe they have no training need - and why they're wrong

Perhaps the most troubling finding of the TÜV study isn't the training rate itself, but the reasoning given by those not training. 45 percent of surveyed companies that haven't conducted any AI upskilling say they see no need. In almost all cases, this assessment is a misjudgment - for three structural reasons.

First: silent AI use is underestimated. Many employees use AI tools privately and carry these habits into their working day without management knowing. Shadow AI - the unauthorized, undocumented use of AI tools - is widespread in German companies. Any executive who believes their company doesn't use AI is very likely mistaken.

Second: the competence gap is hard to see from the inside. You can't recognize a knowledge gap you don't know you have. Employees who operate AI tools without a solid foundation produce plausible-sounding results - and often don't notice themselves when those results are flawed. A lack of AI competence rarely shows up as an obvious failure; it shows up as subtly wrong decisions based on AI outputs.

Third: the regulatory reality changes fundamentally from August 2026. Even companies that today use little or no AI will in the future be confronted with AI-generated outputs - through supply chains, customer interactions, or regulated processes. Article 4 of the EU AI Act doesn't distinguish between active AI developers and passive users: whoever operates or uses AI systems bears responsibility for their staff's competence.

A simple practical test helps with a reality check: ask five employees from different departments which AI tools they used in the past week. In most Mittelstand companies, this survey shows that actual AI usage is substantially higher than what's officially known. The gap between visible and invisible AI usage is the first step toward an honest inventory.

What the Mittelstand can learn from large companies

The gap in training rates between large companies (49 percent) and smaller businesses (32 percent) is no accident. Large companies have dedicated learning-and-development teams that systematically survey and document training needs. They have learning management systems that automatically record training records, and they often work with framework agreements with education providers.

This structural advantage is hard for SMEs to replicate - but the method is transferable. Large companies treat AI upskilling as an investment with a measurable return, not as a cost center. Three levers that also work for the Mittelstand: first, appoint an internal owner - even as an add-on to an existing role. Second, use funding: the Qualifizierungschancengesetz and §82 SGB III enable up to 100 percent cost reimbursement through the Federal Employment Agency (Bundesagentur für Arbeit). Third, tie training to concrete pilot projects - anyone who optimizes a process with AI right after training anchors what was learned and generates measurable value.

What SMEs can do right now: a three-stage model for AI upskilling

Given the time remaining until August 2, 2026 and the broad competence gap, SMEs don't need a massive program - they need a pragmatic entry point that secures compliance while creating real value at the same time. The following three-stage model has proven itself in practice.

Stage 1: AI inventory (1–2 weeks)

Before training begins, you need to know what's actually being used. A structured inventory captures which AI tools are used in the company - officially and unofficially - in which departments and for what purposes. This step is also the basis for risk classification under the EU AI Act: high-risk applications require deeper competence measures than standard productivity tools.

Stage 2: define a competence framework (1 week)

Not every role needs the same AI competence. A sensible competence framework differentiates between three levels: foundational understanding for all employees (What is AI? How do language models work? What are the risks?), application-related competencies for AI-using specialists (prompt engineering, output validation, data protection in AI use), and strategic competencies for executives (AI governance, compliance, risk assessment).

Stage 3: implement and document training measures (4–8 weeks)

Implementation doesn't have to be expensive or time-consuming. Internal workshops, e-learning modules, external certificate courses, and structured self-study paths are all valid formats - what matters is documentation: who learned what and when, which learning objectives were defined and demonstrated, which content was taught. This documentation is the central proof for Article 4 compliance.

Important: competence development is not a one-off project but an ongoing process. AI is evolving faster than any other technology in the history of corporate digitalization. An annual refresh cycle is the minimum standard.

Conclusion: the doubling is a start - not a destination

The TÜV Study 2026 paints an ambivalent picture. The doubling of the AI training rate from 12 to 27 percent is a real signal that the German economy is taking AI upskilling seriously. But realistically, it remains a start, not a success: 73 percent of companies still don't train for AI, 45 percent see no need, and usage reality is outpacing competence reality by 29 percentage points.

From August 2, 2026, this gap is no longer just a competitive disadvantage - it's a compliance risk. EU AI Act Article 4 makes AI competence mandatory. The question is no longer whether companies will act, but how quickly.

Anyone who hasn't launched structured measures by then faces a double problem: the operational risk of unqualified AI use and the regulatory risk of demonstrable compliance gaps. Both risks can be significantly reduced with a pragmatic, structured approach - but only if you start now. The point at which waiting was still an option has passed.

AI upskilling isn't a one-off measure but an ongoing process. The technology evolves quickly - new models, new use cases, new regulatory requirements. Companies that build a structured training program today aren't creating a snapshot, but a learning infrastructure that keeps pace with the market. The first step is the hardest - and the most important. Whoever takes it now will have a competitive edge in the race for talent, customers, and efficiency after summer 2026 - one that grows year after year.