BERLIN — German workers face a more uneven threat from artificial intelligence than a simple choice between mass unemployment and higher productivity. Research suggests the technology could leave overall employment broadly unchanged while shifting demand between industries, qualifications and generations, with younger workers facing particular difficulties entering the labour market.
Germany’s Institute for Employment Research, known as IAB, offers a complicated picture. Its long-term modelling finds substantial job creation and destruction that largely cancel each other out. A separate analysis of employment records finds no general collapse in AI-exposed occupations through 2025, but identifies weaker outcomes among younger workers and cautions against treating those patterns as proof of AI-driven displacement.
The distinction matters for training policy and company investment. A broadly stable national employment total would not mean the same people retained their jobs, the same industries kept hiring or new entrants found an equivalent route into skilled work. The research separates the potential economic benefits of AI from the difficult process of moving workers between activities.
The tension is reflected in Siemens‘ chief executive’s argument that the competitive threat comes from people who know how to use AI. Nikkei Asia promoted its November 10 report on his remarks with the headline, “AI won’t take jobs, but those who can use it will.” That proposition places skills at the centre of the debate, but does not establish that displaced workers will be able to move into the roles that emerge.
In its research report 23/2025, IAB examined an accelerated AI-adoption scenario over 15 years against a reference projection. Around 1.6 million jobs would be created or eliminated across sectors because of the additional structural change, while employment at the end of the period would remain close to the reference scenario. This is a conditional model, not a count of announced redundancies.
The study, produced by researchers from IAB, the Federal Institute for Vocational Education and Training and the Institute of Economic Structures Research, assumes businesses make the investments needed to deploy AI. Those include computing infrastructure, suitable software, data preparation and workforce development. Its results depend on those assumptions being fulfilled.
The model also describes a changing sequence of employment effects. Early investment creates demand for workers to build infrastructure, prepare data and develop systems. Efficiency gains subsequently reduce labour requirements in some activities. Later, new business models could generate enough additional demand to offset much of that reduction.
The eventual balance conceals clear sectoral differences. After 15 years, employment in IT and information services is around 110,000 higher than in the reference scenario. Business support activities employ around 120,000 fewer people. Those figures measure differences between modelled outcomes, not changes that have already occurred.
Nor does higher qualification automatically provide protection. The researchers expect AI to affect highly skilled cognitive work more strongly than earlier forms of conventional software. Their model finds different pressures on specialists and experts over time, challenging the assumption that automation is principally a problem for routine manual occupations.
The potential economic gains are substantial. Under the study’s assumptions, AI raises annual economic growth by an average of 0.8 percentage points and generates €4.5 trillion in cumulative additional value added over 15 years. That is an accumulated difference from the reference projection, not an annual windfall or a guaranteed result.
A separate IAB analysis published on October 13, 2026, examines what has happened rather than what might happen. Researchers Melanie Arntz, Michele Dinelli and Oliver Schlenker used monthly administrative records covering an average of about 22.3 million full-time employees aged 20 to 64, from January 2021 through December 2025.
Across the two most AI-exposed occupational groups, employment in 2025 was approximately 1% above its 2022 level. In the three less-exposed groups, it was just over 3% lower. The overall pattern therefore does not support a claim that occupations with greater technical exposure to AI had already experienced a general employment collapse.
For younger employees, the comparison was less reassuring. Among workers aged 20 to 29, full-time employment in highly exposed occupations fell 3.8% from 2022, compared with 2.6% in less-exposed occupations. That 1.2-percentage-point gap describes a difference in employment trends; it is not an estimate of the share of young workers replaced by AI.
Once the researchers also considered occupations’ sensitivity to the weak economy and suitability for homeworking, the negative relationship between AI exposure and younger workers’ employment remained statistically insignificant overall. The relationship with economic weakness was considerably stronger. Statistically significant negative associations did appear in manufacturing and non-academic occupations.
The researchers’ central limitation is important: their exposure measure identifies tasks that large language models could accelerate, not whether employers actually adopted those systems. Establishing causation would require linking observed company AI use with employment records. The analysis also excludes groups including the self-employed, apprentices and people without employment.
Even where new roles offset disappearing ones nationally, the transition can still interrupt careers. The long-term model does not establish that workers leaving shrinking activities will possess the qualifications demanded by expanding sectors. The employment-record study, meanwhile, raises questions about the early-career opportunities through which younger people acquire experience.
Germany’s unresolved problem is therefore not simply whether AI creates as many jobs as it removes. It is whether the workers losing an established role, or trying to secure their first one, can reach the opportunities on the other side of that reshuffle.
