Artificial Intelligence as an Opportunity for Growth

Page 1


RESEARCH PAPER | ECONOMIC AND SOCIAL POLICY

Artificial Intelligence as an Opportunity for Growth

The overall economic potential of AI for Germany

2 October 2025

Remarkable technological progress has been achieved in the field of artificial intelligence (AI) over the last few years, which is fundamentally changing the world of work and production. AI is regarded in political, scientific and economic circles around the world as a key technology with huge economic potential, also for Germany. In this context, this research paper investigates to what extent AI can bolster overall economic growth in Germany.

Economic opportunities: In the medium term, AI can make a tangible contribution to overall economic growth by lifting productivity, creating new business models and triggering momentum in research and development. AI could increase the potential growth of Germany by around 0.3 percentage points per year until the end of the decade. This does not amount to a surge in growth, by any means, but is nonetheless a relevant boost particularly given the low rate of growth currently anticipated.

Current position: Germany is currently in the middle of the pack among G7 countries regarding expected effects of AI on productivity. The structure of Germany’s economy with a strong industrial sector and limited service intensity curbs the direct impact of AI. Further impeding a swift and widespread introduction of AI throughout the German economy are structural challenges including a corporate landscape dominated by mid-sized enterprises, deficits in digital infrastructure, a shortage of skilled staff and limited financing options, particularly for start-ups.

Strategic recommendations: Germany should make targeted use of its strengths in industry, its listed corporations and midsized family businesses, to advance the use of AI, both in development and in practical applications. To make this happen, the political framework must enable targeted investment in sovereign infrastructure, promote the development of AI skills in education and in enterprises, and create the legal and economic prerequisites to facilitate the innovation and application of AI. Enterprises are called on not just to use AI in isolated applications but apply it strategically for the restructuring of business models and value added. Germany will only be able to secure its sovereignty and adopt a leading role in industrial AI if policymakers and industry act in unison.

Executive Summary

▪ AI as a potential driver of growth: Artificial intelligence (AI) is a general-purpose technology with a wide spectrum of application. It can automate processes, support complex tasks, create new fields of work and accelerate research. First studies show considerable efficiency gains in some areas, particularly in knowledge-intensive services. The economic potential of AI largely depends on how well it is integrated into existing work and production processes.

▪ Industrial application: The application of AI is also increasing in the industrial sector. According to the German research company, the ifo Institute, 46.7 percent of companies in the manufacturing sector already use AI. The use of AI is particularly widespread in the automotive sector, mechanical engineering and in the chemicals industry. However, use is so far focused largely on administrative processes. Only individual pioneer companies have already integrated AI broadly into their production Empirical evidence of measurable productivity effects is also limited to date. Case studies and estimates of the potential of AI clearly show that considerable efficiency gains are possible in individual industries. Overall, the emerging trend is clear that AI will also become increasing economically relevant in the industrial sector, even if its broad use in production is only just beginning.

▪ Overall economic potential of AI: Studies forecast an annual productivity growth through AI of 0.1 to 0.8 percentage points for Germany (based on total factor productivity). When factoring in structural effects, such as, for example, that additional demand often takes place in less productive areas, and AI-related investment, the expected overall economic growth impetus is around 0.3 percentage points per year. AI could thus tangibly increase Germany’s production potential, which is currently only estimated at between 0.2 and 0.4 percent per year. On an international comparison, Germany ranks in the middle among G7 countries in terms of expected productivity effects, with limited sectoral impact but moderate AI adoption momentum.

▪ Locational factors of Germany: Current studies only take limited consideration of the key national parameters. The dominance of midsized enterprises in the corporate landscape, deficits in digital infrastructure, shortages of skilled staff and limited financing options for start-ups and smaller enterprises are all factors that encumber the swift and broad application of AI in the economy. Access to high-quality data and the transfer of scientific findings into commercial application remain key challenges. These factors relativise the modelled productivity effects and show that technological potential alone is not enough

▪ Recommended action: Germany should apply its industrial strength to the targeted development and application of AI technologies. For policymakers, this means creating the necessary conditions for the broad application of AI throughout the economy, through investment in sovereign infrastructure, the development of AI competencies and innovation-friendly framework conditions. In addition, a priority should be enabling technology transfer and scalability. Enterprises of all sizes are called on to use AI not just for isolated tasks but strategically for the restructuring of business models, value added and organisational methods Germany will only be able to secure its sovereignty and adopt a leading role in industrial AI if policymakers and industry act in unison

1. Introduction

Remarkable technological progress has been achieved in the field of artificial intelligence (AI) over the last few years, which is fundamentally changing the world of work and production AI is regarded in political, scientific and economic circles around the world as a key technology with huge economic potential, also for Germany.

In this context, we investigate to what extent AI can bolster overall economic growth in Germany. We analyse the theoretical impact mechanisms of AI, use micro and macroeconomic findings and compare the locational factors of Germany on an international level. The objective of this research paper is to provide a factually solid assessment of the potential impact of AI on growth and outline political approaches to enable this potential to best unfold.

We focus particularly on setting out the conditions under which AI can be used for broad effect in the German economy and not just for isolated applications, and what political and structural framework conditions are necessary for this to happen.

2. The Growth Potential of Artificial Intelligence

2.1 Theoretical foundations and impact mechanisms

Artificial intelligence is increasingly regarded in scientific literature as a general-purpose technology (GPT), a technology with the power to affect the whole economy and whose economic significance is comparable to historical innovations such as electricity and the internet (OECD 2025). As a generalpurpose technology, AI has the potential to not only make individual processes more efficient, but fundamentally transform entire production and work structures, with far-reaching consequences for productivity, innovation and overall economic growth.

From a theoretical perspective, several impact mechanisms can be identified through which AI can influence growth and productivity. A core channel is automation AI can perform tasks that have so far been performed by humans, particularly cognitively easy, rule-based or heavily standardised tasks which can lower costs and free up manpower To what extent this impacts overall economic productivity depends on how many tasks are actually performed by AI and whether the freed up manpower can take on other more productive tasks (Acemoglu, 2024). At the same time, AI is not labour-replacing but rather labour-augmenting in many areas, meaning it can support employees in certain tasks, such as automating partial processes or the structured processing of information, which, in turn, increases their productivity.

First applications show that AI has so far been used mainly by the service sector to improve access to information and to create digital content (Acemoglu 2024, ECB 2025). AI technologies are also being used increasingly in the manufacturing sector, although mainly in the form of individual pilot projects so far. Some enterprises use AI-based image recognition systems to automate the identification of production errors and thus reduce rejects (see Fraunhofer ITWM, not dated). Industrial enterprises are also working on initiatives to control production processes over cloud platforms in future, including AIsupported maintenance, process optimisation and quality control In addition, some companies are

developing so-called AI agents that can make decisions independently, such as automatically analysing the causes and suggesting suitable solutions in the event of machine failures

Furthermore, AI could also trigger the creation of entirely new tasks and the reconfiguration of existing work processes. These kinds of creation effects are regarded as potentially increasing productivity with positive long-term effects on employment and wages, particularly if they go beyond individual tasks and enable new fields of activities or business models (Acemoglu 2024, OECD 2025). Such effects have been a key driver of long-term growth in previous technological breakthroughs (OECD 2025). As they are difficult to measure, these effects have not yet been systematically considered in many studies

An additional channel for productivity increases that has also not yet been analysed at great length is the application of AI to accelerate research and innovation. With the help of AI, companies can analyse large amounts of data, simulate experiments and identify new correlations, for example, in biomedicine and materials research. These effects are difficult to quantify but could nonetheless contribute considerably to productivity growth in the long run. The potential of AI as an invention technology is highlighted both in an OECD paper (2024), which shows how AI can accelerate research and open up new paths to innovation, and in a report by the German Federal Ministry of Economic Affairs and Energy (2024), which underlines its indirect effects on productivity through research and development as well as its direct effects on productivity through automation.

It should nonetheless be remembered that not every task enabled by AI leads to an increase in welfare Acemoglu (2024), for example, sounds the alarm about the emergence of so-called bad tasks, such as deepfakes, manipulative advertising systems and automated disinformation. Although these kinds of applications can generate economic activity, they potentially undermine social benefits and welfare.

Furthermore, on an aggregated level, structural growth may be limited by what has been termed the Baumol effect (OECD 2024, 2025). The Baumol effect explains why aggregate productivity growth can remain limited, particularly if sectors expand in which higher productivity is difficult to attain, such as healthcare, nursing or personal services The effect emerges because technological progress in highly productive sectors (such as in IT and financial services) lead to lower prices in these sectors and incomes increase. However, the freed up purchasing power is not spent exclusively in these sectors but increasingly on services which cannot be automated to any great degree and whose productivity only grows gradually. These areas, such as personal services, experience greater demand because they are typically consumed more when incomes increase. At the same time, the proportion of highly productive sectors drops in relation to the overall economy because their prices fall and demand does not grow to the same extent. This means that less productive sectors account for a larger proportion of overall economic output which curbs aggregate productivity growth. In the OECD scenarios, the Baumol effect reduces the aggregate productivity growth through AI by between around one sixth and, in extreme cases, one third depending on the assumptions.

It should be noted that almost all available analyses primarily assume a gradual further development of existing AI technologies. The possibility of so-called artificial general intelligence (AGI), a crosssector human-level AI, is usually explicitly excluded as this could lead to a significant acceleration of the rate of innovation, the impact of which lies beyond the methodological framework of current models and cannot be reliably quantified.

The following table provides a compact overview of the possible effects of AI on productivity

Table 1: Potential impact mechanisms of AI and their impact on productivity

Impact mechanism Impact on productivity

Automation

→ Labour-replacing

Automation

→ Labour-augmenting

Task creation

Research & innovation

Socially dubious applications

Baumol effect

AGI (artificial general intelligence)

Positive: Efficiency gains through the automation of individual tasks

Further comments

Impact dependent on reallocation of manpower

Positive: Supporting human labour through partial automation and better processing of information –

Positive: New tasks and business models could potentially increase longterm productivity

Positive: Faster analysis, simulation and findings

Difficult to measure, therefore often not considered in studies

Difficult to quantify, indirect effects often not investigated much

Negative: Can generate economic activity, but also undermine social benefits and welfare –

Negative: Curbs progress in productivity as it shifts demand to less automatable sectors –

Not evaluable: Potential to greatly increase productivity, but not included in current analyses –

2.2 Microeconomic evidence

Company level

While theoretical considerations indicate that AI can generally lead to an increase in productivity, the question arises whether there is also empirical evidence to prove that this is actually the case A number of microeconomic studies have delivered the first solid indications on this point, showing that AI has already led to efficiency gains in specific operational contexts – above all in knowledge-intensive services (see Box 1).

The empirical evidence presented until now focuses largely on the service sector, some surveys, including the ifo Institute (2025) survey, show that AI use is also increasing in the industrial sector. According to the ifo Institute, 46.7 percent of companies in the manufacturing sector already use AI in their business processes, which is even more than in the service sector, where AI use is at 41 percent. Industries with particularly high rates of AI use are the automotive sector (70 4 percent), mechanical engineering and chemicals (around 50 percent in both cases). The industrial sector therefore has a considerably higher proportionate use of AI than sectors such as the food service industry (31 3 percent) and textile production (18 percent). Across the board, 40.9 percent of German companies already use AI and a further 18.9 percent are planning to do so soon

While there are hardly any solid microeconomic studies with measured productivity effects in industrial applications, some case studies and estimates of potential effects show that AI is certainly capable of enabling significant efficiency gains in individual industries. AI is already deeply integrated in industrial processes in some areas, ranging from optimised production lines and highly automated driver assist systems up to generative applications on end devices (BDI/BCG 2025). The increasing operational

use and expected efficiency potential indicates that AI is gradually moving beyond experimental application and becoming increasingly relevant to the whole economy.

At the same time, a report by the MIT of Project NANDA of July 2025, The GenAI Divide: State of AI in Business 2025, paints a sobering picture. According to the report, around 95 percent of enterprise GenAI pilots have so far not had a measurable impact on either increasing revenue or reducing costs. The report is based on a systematic analysis of more than 300 publicly disclosed AI initiatives, structured interviews with representatives from 52 organisations and survey responses from 153 senior leaders. A central obstacle identified by the report is a learning gap. Many systems are unable to process feedback, further develop based on context retention or be seamlessly integrated into existing work processes. While the use of generative AI tools such as ChatGPT and Copilot is growing rapidly and already increasing individual productivity, most companies are not yet able to translate this potential into concrete productivity gains or effects with a tangible impact on profit and loss

The major obstacles identified by the report are less technical limitations such as model quality or regulatory issues than organisational shortcomings, such as the integration in operational processes is not deep enough, or investments directed more at visible but less profitable fields of application such as sales and marketing. Successful initiatives are particularly in areas where AI is closely connected to operational processes, particularly in back-office tasks such as customer service, finance and sourcing The report clearly shows that the realisation of productivity gains is less a question of technology than the capability of AI systems to be integrated and continuously further develop within existing processes.

Box 1: Empirical evidence of the impact of AI on corporate productivity

The following studies show how the application of AI can increase productivity at the company level, based on specific application contexts. They do not just document the positive effects of AI on output and efficiency but also clearly demonstrate under what conditions and in what areas AI can be particularly effective. The selection of studies presented below include both experimental and field-based analyses of various industries.

1. Czarnitzki et al. (2023): AI in German companies – revenue up between 5.5 and 13.9 percent with same input

- The study is based on data from 5,851 German industrial and service enterprises. AI users, defined as companies that had adopted at least one AI method (for example speech processing, image recognition, machine learning or knowledge-based systems) by 2018 in one application area (for example, products, processes, customer interaction, data analysis), increased revenue by 13.9 percent (simple OLS estimate). In models with control variables, the increase was between 5.5 and 5.7 percent. The authors interpret the higher revenue using the same input as an indication of higher productivity, while productivity was not measured directly

2. Ju and Aral (2025): Human-AI collaboration – productivity up 60 percent

- In a controlled experiment with 2,310 participants, AI-supported teams were 60 percent more productive than all-human teams, measured by the number of successfully created advertisements. At the same time, the human-human communication was 23 percent lower with a higher proportion of emotional messages, while human-AI communication was more focused on efficiency. The text quality increased, the image quality dropped slightly.

3. Dell’Acqua et al. (2023): AI in business consulting – productivity up 25 percent

- In a field experiment with 758 people in a leading business consultancy, the application of GPT-4 speeded up processing by 25 percent with the quality of solutions up by more than 40 percent for typical consulting tasks. Particularly lower-performing participants benefited greatly. In the case of tasks beyond AI capabilities, the performance of those using AI deteriorated.

4. Brynjolfsson et al. (2023): AI in customer service – productivity of support employees up 14 percent

- The study with 5,179 customer support agents shows that an AI-supported chat assistant increases productivity by an average of 14 percent, measured by number of resolved customer queries per hour. Less experienced staff benefit particularly, with their productivity increasing by 34 percent, while the productivity of experienced staff barely changed. The authors see indications that AI transmits tried and tested procedures thereby supporting the learning process of employees.

5. Peng et al. (2023): AI in software development – efficiency rises 56 percent

- The experiment in software development demonstrates that the use of GitHub Copilot considerably shortens the time needed to complete programming tasks. Developers with AI support completed their tasks 55.8 percent faster than the control group.

6. vfa & BCG (2024); Dermawan & Alotaiq (2025): AI in research & development –development 50 percent faster, revenue up 8.9 percent

- These studies show that AI can increase the productivity of pharmaceutical research and development. The industry report by vfa (German Association of Research-Based Pharmaceutical Companies) & BCG (2024) based on case studies, documents that AI reduces the number of laboratory tests needed and considerably accelerates development time. According to BioNTech, AI can automate the evaluation of up to 97 percent of all experiments. The share of revenue of producers with biopharmaceuticals increased by 8.9 percent. Dermawan & Alotaiq (2025) confirm these effects in a systematic analysis. AI shortens the time from preclinical research up to clinical application by up to 50 percent and improves the success ratio of active agent candidates. Both studies clearly show that AI can significantly increase efficiency, particularly in the early phase of drug development

7. Kodumuru et al. (2025): AI in pharmaceutical production – development up to 50 percent faster, errors down by 40 percent and costs down by 25 percent

- In their review, the authors show that AI can cut the development time of new drugs by up to 50 percent. In production, AI reduces the error rate by up to 40 percent and cuts production costs by up to 25 percent through automating processes. The review underlines that AI enables significant efficiency gains, particularly in process monitoring, quality control and predictive maintenance.

Industry level

Based on the evidence from the company level, a look at the industry level as an analytical interim step will serve to better understand in which parts of the economy AI can be particularly effective and where the overall economic impact is likely to be the highest.

The OECD report (2025) shows that the potential exposure of AI varies greatly between the individual industries, potential exposure being to what extent the typical activities of a sector can generally be supported or replaced by AI. The OECD report identifies knowledge-intensive services such as financial services, information and communication technology (ICT), telecommunications, media and professional services as having the highest AI exposure rates. In these sectors, between 50 and 80 percent (depending on the scenario) are potentially exposed to AI. This estimate is based on task-level AI exposure mapped into specific occupations with AI exposure of sectors then derived based on their occupational composition

Sectors with a high proportion of manual and physical activities, such as agriculture, mining and construction, are considerably less affected. The estimated exposure of these sectors is only between ten and 30 percent. These sectoral differences reflect the different structure of tasks. While knowledgeintensive services are heavily dependent on cognitive tasks, other industries have a much higher proportion of manual and physical tasks and tend to have a lower exposure to AI.

Alongside the degree of exposure, the economic significance of industries is also a relevant factor. According to the OECD (2025) study, the five sectors with the highest AI exposure, ICT, telecommunications, media, finance, and professional services, account for around 15 percent of gross value added in Germany, a much lower proportion than in other G7 countries such as the United States and the United Kingdom

The OECD uses the sectoral exposure to AI outlined above and combines it with microeconomic productivity estimates and forecast adoption rates to calculate the average sectoral growth of total factor productivity (TFP)1 for the whole G7 over the next ten years (Figure 1). The results show considerable differences between the individual industries. In highly manual sectors, such as agriculture and mining, productivity is only expected to grow by around one percent, while in knowledge-intensive services, such as IT, financial and technical services, predicted growth is more than ten percent depending on the scenario. In manufacturing, including industrial production, machinery, motor vehicles, electrical equipment and related industries, the projected TFP growth is between around one and seven percent depending on the scenario. This puts manufacturing in the middle of the range, more affected than highly manual industries such as agriculture and mining, but much lower than knowledge-intensive services. The highest growth is in the scenario “high adoption and expanded capabilities”, that assumes a rapid diffusion and the development of complementary software.

1 TFP measures the impact on economic growth of technological advancement, better organisational methods and improved efficiency in the use of resources.

*Growth rates under different scenarios. Source: OECD (2025)

2.3 Macroeconomic studies

The overall economic potential of AI is increasingly the subject of macroeconomic studies. The objective of these studies is to quantify expected productivity effects across countries and time periods. The results of the studies vary widely due to their different methodological approaches and divergent assumptions regarding technology development and adoption.

The studies differ particularly in the following aspects:

▪ Time horizon: Some studies model effects over five years, some over ten years of more, depending on the definition of the medium-term horizon

▪ Indicator: While many studies analyse total factor productivity (TFP), others look at labour productivity2 or, particularly in studies closely linked to companies, direct impact on GDP

2 Labour productivity defines how efficiently labour is applied in a production process. It measures how much output (goods or services) is generated by labour unit (for example, per hour worked or per worker).

Figure 1: Predicted sectoral productivity growth over ten years*

▪ Model approach: The methodological approach varies from bottom-up calibrations up to macroeconomic simulation models with sectoral differentiation. In many cases, microeconomic data on the exposure of tasks is combined with macroeconomic structural parameters (such as wage costs or industry structure).

▪ Assumptions on adoption: Assumptions on AI adoption, in part, go beyond technical adoption and model to what extent adoption is economically feasible, using, for example, empirically estimated rates of adoption, cost assumptions and country-specific scenarios.

▪ Country focus: The countries and regions investigated in the studies vary widely, in part (for example, United States, Europe, G7), which influences the respective context of analysis and the comparability of results.

International perspectives

Given the methodological differences, the studies arrive at partly very divergent results. Acemoglu (2024), the International Monetary Fund (IMF 2025) and the OECD (2024b, 2025) calculate comparatively low to moderate contributions of AI to productivity growth:

▪ Acemoglu (2024) develops a macroeconomic model based on the concept that AI either completely takes over individual tasks or supports humans carrying out these tasks. The overall economic productivity gain is a result of two factors here: the proportion of affected tasks and the estimated cost savings per task. Based on empirical studies of AI exposure and on time and cost savings, Acemoglu estimates an accumulated TFP growth of around 0.7 percent over a ten-year period for the United States (around 0.07 percent per year). With more conservative estimates, including the limited capabilities of AI in complex, context-dependent tasks, this figure drops to a maximum of 0.55 percent.

▪ The IMF (2025) forecasts for Europe a growth in TFP of around 1.1 percent over a five-year period in its preferred scenario, which is almost 60 percent more than Acemoglu. The IMF regards the higher results as being caused primarily by more optimistic assumptions on the capabilities of AI. The study simulates productivity gains for 31 European countries based on Acemoglu’s model and, in addition, factors in diverse scenarios and country-specific differences in AI adoption and in the impact of regulation.

▪ The OECD (2024b) estimates for the United States an annual growth in total factor productivity through AI of 0.25 to 0.6 percentage points over ten years. Labour productivity could thus increase annually by between 0.4 and 0.9 percentage points. On a historicalcomparison, these figures are certainly significant but lower than the surges in productivity triggered by former technological booms. It is estimated that information and communication technologies (ICT) triggered a TFP growth surge of up to 1.5 percentage points in the United States between 1995 and 2004. This corresponds to a contribution of around 1.5 to 2.25 percentage points to labour productivity. The figures are significant, given that growth in labour productivity in the United States in this period was between 1.5 and three percent and 2.5 percent on average (OECD 2025b)

▪ The OECD (2025) study follows on directly from its previous study (2024b) to expand the analysis to all G7 countries. Methodologically, the study builds on Acemoglu (2024) and combines three central elements: (1) microeconomic productivity gain on task level, (2) sectoral exposure to AI and (3) forecast adoption paths under different scenarios. In the central

scenario, which is medium adoption speed and expanded AI capabilities, the estimated annual growth in labour productivity is between 0.5 and one percentage point over ten years, with particularly high growth in knowledge-intensive economies such as the United States and the United Kingdom

Studies closely linked to the market such as McKinsey (2023) and Goldman Sachs (2023) are based on more optimistic assumptions and forecast considerably higher effects. McKinsey expects the combined application of generative AI and other automation technologies to result in an annual growth of labour productivity of up to 3.4 percent. Goldman Sachs forecasts GDP growth of up to seven percent over ten years. These estimates are based on a broader definition of technology, use different indicators and assume a particularly rapid diffusion. Generative AI alone could, according to McKinsey, increase the annual growth of labour productivity by between 0.1 and 0.6 percent until 2040.

Germany in focus

The international studies outlined above provide valuable insights into the overall economic potential of AI. How these developments will impact the German economy specifically depends to a considerable extent on the national framework conditions. In view of the specific factors involved, such as industry structure, adoption momentum and regulatory parameters, the following section looks at selected studies that analyse Germany specifically:

▪ IW Consult (2025) analyses the overall economic impact of AI based on a growth accounting model. For Germany, it calculates an annual growth of labour productivity (real GDP per hour worked) of 0.9 percent on average (2025-2030) and 1.2 percent (2030-2040), compared to only 0.4 percent in the 2020s so far. This puts the anticipated growth at the level of the 2000s, which does not correspond to a ‘productivity miracle’ but represents a clear improvement compared to recent growth. The productivity gains are mainly a result of technical and organisational progress and higher capital intensity, although the latter is not primarily due to new investment but due to the demographically induced reduction of labour. The contribution of AI to total factor productivity is estimated at between 0.1 and 0.3 percentage points per year, based on scenario assumptions

▪ The OECD (2025) forecasts for Germany in its central scenario, which is estimated medium adoption speed and expanded AI capabilities, an annual growth in labour productivity of 0.86 percentage points over the next ten years. In the more optimistic scenario (fast adoption) this figure increases to 1.16 percentage points. In the conservative scenario (slow adoption) this figure drops to 0.34 percentage points. This puts Germany in the middle of the range among G7 countries, on a par with Canada, behind the United States and the United Kingdom, but ahead of France, Italy and Japan. The OECD identifies several factors that puts Germany in this position. These are, firstly, Germany demonstrates a moderate starting level in AI adoption, the central OECD estimate of high-intensity AI use among companies is 4.8 percent (US: 6.1 percent, Japan: 1.9 percent). Secondly, the proportion of knowledge-intensive, strongly AI-exposed sectors at around 15 percent of value added is comparably low, much lower than in the United States and the United Kingdom, and on a similar level to Japan and Italy, with industrial production accounting for a larger share of the economy than knowledgeintensive services. Despite these structural limitations, Germany is propelled by a comparatively dynamic trend in AI adoption until 2034 in the model, which, in combination with the other factors, results in a forecast labour productivity increase that is middle of the range compared to the other G7 economies.

▪ The IMF (2025) forecasts for Germany a cumulative TFP gain of around 1.3 percent over five years in the preferred scenario (estimate based on Figure 5), which puts Germany seventh among the 31 European countries. In the model used, this is due particularly to two factors. First, Germany’s wage level is well above the EU average according to Eurostat, based on the regression-based models (2025) of the IMF, and higher wage levels tend to display higher AI adoption rates. Second, the sectoral structure of Germany plays a role. In the IMF model, the level of productivity gains is largely dependent on the degree to which the national value added is produced by AI-exposed sectors. Countries with a higher proportion of knowledge-intensive services, such as financial services, tend to achieve higher productivity gains according to the model. Even if the IMF study does not include any detailed sectoral data for Germany, the sectoral structure of Germany is likely to amount to a medium proportion of AI-exposed value added. While the higher wage level will tend to contribute to an above-average AI adoption, the sectoral structure of Germany, in contrast, will lead to a medium AI exposition in the model which, when combined, results in a productivity growth that is in the upper medium range among European countries. Alongside the direct effects of AI on productivity, the IMF also analysed the role of regulation in Europe. The IMF roughly estimates that national and EU regulation around occupation-level requirements, AI safety, and data privacy combined could lower AI exposure by 50 percent lower in tasks, occupations and sectors affected by regulation This could also reduce Europe’s productivity gains by over 30 percent.

▪ An ECB (2024) study ranks Germany third in a comparison of 18 Euro area countries regarding the estimated TFP gain through AI over ten years (based on Figure 15). The study does not specify an exact figure for Germany, but the visual classification indicates growth of more than three percent, which is slightly above the euro area average of 2.9 percent (corresponding to 0.29 percentage points per year, with a range of 1.3 to 4.5 percent). The estimate is based on an adapted model framework according to Acemoglu (2024), in which country-specific differences in task structure and share of GDP is factored in. Other parameters such as automatability and efficiency gains are assumed to be constant in the model. Detailed countryspecific parameters for Germany are not provided in the study.

Classification of results

Table 2 shows an overview of the findings of the studies. For improved comparability, the TFP effects were converted to labour productivity and vice versa 3 For Germany, the studies demonstrate an estimated annual growth of labour productivity through the application of AI of between 0.15 and 1.16 percentage points (corresponds to between 0.1 and 0.77 percentage points of TFP gain), depending on the scenario and the methodology On average, the four Germany-specific studies estimate an annual increase of 0.51 percentage points in labour productivity and 0.34 percentage points in TFP. Particularly noteworthy are the comparatively more optimistic forecasts of the OECD (2025), which forecasts in its medium scenario an increase in labour productivity of 0.86 percentage points and a TFP gain of 0.57 percentage points.

On an international comparison, the estimated growth rates in labour productivity through AI range from between 0.1 and 1.3 percentage points (0.07 to 0.87 related to TFP). Germany is therefore in the

3 Based on a capital multiplier of 1.5: That means that labour productivity grows about one and a half times as fast as TFP. Put differently, around 70 percent of growth in labour productivity is accounted for by TFP if capital growth remains moderate. This simplified rule of thumb is also used in the literature (see OECD 2024b).

middle of the range overall and partly also in the upper end of the range. The exact positioning of Germany greatly depends on the assumptions used and the speed of implementation.

The expected productivity gains through AI are therefore well below the growth surge triggered by the information and communication technologies (ICT) in the United States between the mid-1990s and the 2000s. As mentioned above, ICT are estimated to have contributed around one to 1.5 percentage points to TFP growth per year or, put differently, this would correspond to a contribution of around 1.5 to 2.25 percentage points to the annual increase in labour productivity.

Table 2: Study overview on the overall economic potential of AI

Source Country/region

McKinsey (2023) Global

points*)

0 07-0 4 (2 27 with combined application) 0 1-0 6 (3 4 with combined application)

OECD (2025) G7 0 13-0 87 (medium scenario: 0 33-0 67) 0 2-1 3 (medium scenario: 0 51 0) IMF (2025) Europe

22

OECD (2024b) United States 0 25-0 6 04 -0 9

Consult (2025) Germany 01 -0 3 0 15-0 45

OECD (2025) Germany 0.23-0.77 (medium scenario: 0 57) 0.34-1.16 (medium scenario: 0 86)

(2025)

* The estimates in productivity growth in the different studies do not always differentiate clearly between percent and percentage points. In line with the OECD (2024b), the following table assumes that the figures are percentage points as these more accurately map the additional effects of AI.

The estimates for Germany are interesting in that labour productivity in Germany only increased on average by around 0.9 percent per year between 1986 and 2022. It stands out that growth of capital per worker has barely contributed to productivity growth since 2009. In contrast, the contribution of total factor productivity in the last 15 years was 0.4 percentage points per year on average, but 0.7 percentage points over the whole period between 1986 and 2022.

Figure 2: Development of labour productivity (in percent)

Source: OECD

Contribution of trend capital per worker growth

Contribution of trend MFP growth

Overall, the present studies contain some significant uncertainties. Alongside model-related assumptions on technology adoption and efficiency gains, other factors that are often not considered include the possible progress in AI research, structural changes and the adaptability of employees, including potential impact on cognitive capabilities, and the additional effects of AI-related investment (see ECB 2024).

Furthermore, most of the studies referred to in this research paper do not explicitly factor in the Baumol effect, which reduces overall economic productivity. The estimated productivity effects could therefore be overestimated by an average of about one sixth (see OECD 2024b and 2025). While the OECD study (2024b), for example, explicitly models the Baumol effect in a central scenario and quantifies its curbing effect (reduction of annual growth of labour productivity from 0.93 to 0.8 percentage points), its follow-on study (2025) discusses this effect extensively but does not factor it into its central estimates. The Baumol effect is not explicitly considered in the other studies.

On the other hand, major capital investment in AI expansion is also left out of the equation which could lead to an underestimation of productivity effects. In the United States, in particular, considerable funds are being spent on the expansion of AI infrastructure. Leading technology corporations are building huge computing centres, modernising hardware and expanding their cloud capacities, often connected with a higher energy demand and investments in electricity grids and energy supply. Goldman Sachs (2025) estimates that the global energy demand of computing centres could increase by around 50 percent until 2027 and up to 165 percent by 2030 (compared to 2023; forecast with considerable uncertainty). McKinsey (2025) estimates the potential global investment volume in computing centres until 2030 at around 6.7 trillion US dollars, 5.2 trillion US dollars of which on AI-specific infrastructure (including GPUs, storage, networks). Such investments increase the capital stock and can trigger longterm growth momentum However, these investments are often not explicitly factored in yet, which means that the calculated effects are possibly underestimated. At the same time, it is unclear to what

extent the high investments will actually lead to overall economic gains in productivity, particularly as they result in considerable costs and structural adjustments, especially regarding energy consumption.

Breaking down the global McKinsey estimate, that AI-specific infrastructure including computing centres, specialised hardware such as GPUs, storage and network technology, and cloud and edge capacities will require around 5.2 trillion US dollars of investment worldwide, to Germany amounts to a required annual investment volume of around 30 billion euros. This figure is a rough estimate based on the aggregate global investment required throughout the remaining years until 2030 (about 867 billion US dollars per year) and the share of Germany in the global economy (about four percent in nominal terms) With the current exchange rate of one US dollar to 0.85 euros, this results in the above estimate of annual investment needed for Germany.

On the one hand, this estimate can be regarded as being on the conservative side. Germany could take on an overproportionate share in the expansion of European AI infrastructure if it wants to become a leading digital and industrial location in Europe. Additional investments in energy supply and security are probably also not included in full in the calculations. On the other hand, public funding is already being spent and is scheduled to be scaled up in the next few years but not easy to factor in. In addition, private investment that has already been earmarked for other areas could be reallocated to investment in AI. Furthermore, if the expansion of infrastructure towards AI is not productively used in the long term, then these investments would ultimately only have a limited effect on potential growth.

There is naturally always a risk of overinvestment in every large technology surge and the discourse in the United States about the valuation of leading technology firms shows that uncertainty about the anticipated growth in revenue and earnings factored in by the capital markets here is considerable and that any small change in the news can play out in huge changes to valuation levels.

Given all these factors, the contribution to long-term potential growth of the required AI-related private and public infrastructure investment can roughly be estimated as follows: The annual required investment volume of 30 billion euros corresponds to around 0.14 percent of the total capital stock of Germany (about 20.8 trillion euros; Destatis 20194). Assuming that additional investment typically contributes with a capital elasticity of around 0.3 to potential growth (see OECD 2025c), this results in a marginal annual growth impetus of around 0.05 percentage points. This estimate relates to potential growth and does not factor in any short-term effects on demand.5

Conclusion

In sum, the potential overall economic growth impetus through AI can be estimated as follows: The increase in total factor productivity for Germany is estimated at between 0.1 and 0.77 percentage points. Taking into account the Baumol effect, this range drops to between 0.08 and 0.64 percentage points. Adding the additional growth impetus of around 0.05 percentage points though capital accumulation caused by AI-related investment, brings the overall effect to between 0.13 and 0.69 percentage points.

4 According to Destatis this is the replacement costs of fixed assets available for production, which means the value as new of all production goods at year end, without discounting depreciation

5 This rough estimate is very simplified: It assumes that the total investment amount is used immediately and completely, does not factor in any time delays or short-term multiplier effects and is based on a rule of thumb for the long-term contribution of capital to growth. As the stock of fixed assets is used as a reference value, the effect can tend to be slightly underestimated.

Based on the average figure of the four studies specifically on Germany of 0.34 percentage points, then discounting the Baumol effect and adding the investment impetus, this results in a growth impetus of 0.33 percentage points. For reasons of clarity, this paper will use a slightly conservative assumption of 0.3 percentage points in its further analysis.

This figure does not include curbing effects caused by, for example, socially problematic applications (see Acemoglu) or regulatory barriers in Germany and Europe (see IMF). At the same time, potentially positive effects, such as the creation of new tasks and business models and a possible acceleration of research and development through AI, are also mostly left out of the equation.

To put these figures into perspective, it is helpful to look at the forecasts for production potential in Germany. In its Spring Report 2025, the Council of Economic Experts estimated a growth of production potential in 2025 and 2026 of only 0.3 to 0.4 percent per year, and an annual growth as low as 0.2 percent from 2027 to 2030. It estimates an annual contribution of capital to growth of 0.3 percentage points, and a contribution of total factor productivity to growth of between 0.1 and 0.3 percentage points. The factor labour, in contrast, is expected to increasingly curb growth, with a negative contribution of minus 0.2 percentage points expected for 2025 and 2026, with a larger negative contribution up to 2030, with as much as minus 0.4 percentage points per year from 2028 onwards. Additional investment from the fiscal package adopted in March and possible AI effects are not explicitly included in these figures.

The German Federal Ministry of Economic Affairs and Energy (2025), on the other hand, does factor in the possible effects of the fiscal package in its German Medium-Term Fiscal Structural Plan and estimates the average growth in the production potential of Germany at 0.9 percent per year through to 2041 (technical assumption, starting with the year 2025). Artificial intelligence is specified as a key technology but is not explicitly included in the explanation of this estimate

With this backdrop, a moderate growth impetus through AI of around 0.3 percentage points could considerably increase potential production compared to the low rates expected by the German Council of Economic Experts. Purely arithmetically, this would almost double the growth in potential output in the next few years, from around 0.3 to 0.4 percent up to 0.6 to 0.7 percent per year, thus substantially supporting productivity growth, though not enabling extraordinarily high growth. In view of the historical trend in productivity and growth, the forecasts of the federal ministry seem to be on the optimistic side. At the same time, sources of major impetus, such as the fiscal package, have so far not been considered in the calculations of the German Council of Economic Experts. Including the realistic effects of AI, the growth of production potential in Germany until the end of the decade could be conservatively estimated at somewhat under one percent. AI would make a tangible contribution to growth but cannot be expected, as already indicated in other studies, to produce a productivity or growth miracle in Germany (see IW Consult 2025, Stiftung Marktwirtschaft 2025).

3. Locational Factors and International Ranking

The studies presented above put Germany firmly in the (upper) middle range on an international comparison in terms of the anticipated productivity effects of AI, but well behind countries such as the United States and the United Kingdom. This is due, among other things, to its sectoral composition: The German economy has a relatively low share of knowledge-intensive, highly AI-exposed service sectors (such as IT and financial services). AI solutions tend to be easier to scale up and can be more

readily integrated in productivity-relevant processes in these sectors which is why countries with a higher intensity of services will initially be able to profit more from AI.

To allow a better overview of Germany’s position in the international AI competition beyond the study results, it is helpful to look at Germany as a business location and the central factors that affect the dissemination and economic impact of AI. Without claiming to be exhaustive, these can roughly be divided into the following relevant categories: First, basic conditions such as infrastructure, skilled staff and capital; second, adoption conditions at company level; and third, systematic barriers to implementation, particularly in the transfer of research findings into commercial application and in the access to and use of high-quality data.

Infrastructure

The first category is the infrastructural conditions, including the digital infrastructure which is not accorded much consideration in the studies presented above. A key aspect of digital infrastructure is the access to high-performance chips, the availability of which is a decisive factor in the scalability and competitiveness of AI applications. The demand for AI chips in the EU is estimated at around 40 billion euros in the next few years, according to industry experts such as Schwarz Digits. Imports from the United States are likely to play a substantial role in covering this demand. Of the investment needed, around 25 to 30 billion euros could be used for the construction of new AI gigafactories (with estimated costs of three to five billion euros per site) and another ten to 15 billion euros for the upgrading of existing computing centres. Some industry representatives, including the digital industry association, Bitkom, regard this estimate as rather conservative for Germany, considering additional needs such as investment in national flagship projects, university HPC clusters and cloud infrastructure (Table Media, 2025). Although Germany has a strong computing infrastructure in the field of research and ranks high up in terms of number and power of scientific supercomputers, according to OECD (2024c), it does not have the industrial and public sector capacities of countries such as the United States and China that would take it right to the front. Germany also has structural shortcomings in digital connectivity. On an international comparison, for example, Germany has a low proportion of highspeed fibre-optic connections which is only gradually increasing (OECD 2024c).

Another infrastructural bottleneck is energy supply. The energy consumption of German computing centres has increased continually since 2010 and was already at around 17 billion kWh in 2021, which corresponds to 3.3 percent of the electricity supply in Germany, a higher share than in the Netherlands (2 7 percent) and the United Kingdom (2.5 percent) (OECD 2024c). Forecasts assume a further increase to over 30 billion kWh by 2030. The electricity prices, which are comparatively high internationally, represent a disadvantage for Germany as a business location (see BDI 2024) and could hamper the expansion of energy-intensive AI infrastructure. Decisive factors for a forward-looking development of AI infrastructure are therefore not only the expansion of technical capacities but also the availability of affordable and sustainable energy.

Skilled staff

Another central factor for Germany as a business location is the availability of skilled staff (not least in rural areas) and skills in the use and development of AI technologies. Although Germany has made some improvements here with targeted measures, including the establishment of 150 additional AI professorships and the location of international talents in technology centres such as Berlin and Munich, the proportion of AI-related job advertisements is still lower than in countries such as the United States and the United Kingdom and many positions remain vacant for an above-average length

of time compared to other sectors (OECD 2024). The OECD identifies lacking (digital) skills and a limited understanding of potential applications as key obstacles for the implementation of AI, particularly in SMEs. The BDI/BCG (2025) study also regards talent as one of the most important currencies in the race to artificial intelligence. Especially given that the United States already has the biggest pool of talent in the AI field with almost 60 percent of the globally leading AI researchers. The study emphasises that the availability of skilled AI staff is increasingly becoming a bottleneck for innovation and competitiveness. Without targeted investment in education, immigration and social acceptance, Germany is in danger of dropping further behind in the global race for AI talent.

Financing

Weaknesses in financing represent another key barrier. Although the number of AI start-ups in Germany has significantly increased in the last ten years, the start-up momentum could still be better on an international comparison. Almost half of the AI start-ups between 2013 and 2024 took place in the United States, more than ten percent in China. With just under 400 AI start-ups in Germany in the same period, Germany was behind France and the United Kingdom (Stiftung Marktwirtschaft 2025). A key obstacle in Germany is the considerably lower availability of venture capital on an international comparison. In 2022, the venture capital invested in German AI start-ups was around 14 times lower than in the United States and five times lower than in China. Countries including the United Kingdom, India and Israel, all recorded higher volumes of venture capital investment in AI (OECD 2024). Given the rapidly rising investment in AI, particularly in the United States, with leading technology corporations such as Microsoft, Alphabet, Amazon and Meta planning to invest more than 300 billion US dollars in 2025 alone, with a considerable portion of that going into AI infrastructure, the divide to Germany will have widened since then. For AI to have a stronger economic impact, Germany needs not only a powerful digital infrastructure and an adequate number of skilled staff but also better financing options for technology-based start-ups.

While the financing of innovative start-ups remains a central challenge in Germany, current developments show that at least large German industrial enterprises are increasingly stepping up their investments in AI. Robert Bosch (2025), for example, plans to invest more than 2.5 billion euros in artificial intelligence until the end of 2027 with the objective of accelerating processes and making products more innovative. Siemens (2025), together with partners including Microsoft and NVIDIA, is pursuing the development of industrial AI solutions including digital twins and software-defined automation. Schaeffler (2025) is also investing in AI and has entered a technology cooperation with NVIDIA to digitalise its production processes. According to the BCG AI Radar 2025, 65 percent of leaders surveyed in Germany said that they were planning to increase their AI investments in the next few months, which is still lower than the international average of 73 percent The examples and findings set out above show that particularly larger companies in Germany are making substantial contributions to the development and application of AI, either through direct investment or through strategic collaborations and technology initiatives. There is not yet any systematically collected data available for corresponding activities of smaller enterprises

Adoption conditions – corporate structure, culture, regulatory aspects

Alongside the infrastructural, knowledge-based and financial conditions, company-level parameters also influence the extent and speed that AI is integrated into the economy, particularly also the corporate landscape in terms of size, which does not feature much in many of the studies analysed above. The OECD (2023) shows that larger companies not only use AI more frequently but also benefit more from the advantages of AI in terms of productivity, particularly if they have the necessary digital

capabilities and infrastructure. With this in mind, the dominance of industrial midsized enterprises in the German corporate landscape could prove to be an especially large challenge. In Germany, 99.3 percent of companies classify as small and medium-sized enterprises They employ 55 percent of employees and account for 42 percent of value added (Destatis 2022). Only industrial enterprises over a certain size will be able to feasibly develop their own AI expertise and applications, a size that is larger than SMEs and usually also larger than small mid-cap enterprises. This highlights the relevance of promoting the diffusion of AI technologies throughout the economy and reducing structural barriers for midsized enterprises in their broadest definition. Providing these companies with easy and affordable access to AI services would be of critical importance.

In addition to the size of a company, cultural and organisational aspects of companies can also influence the adoption momentum of AI. There are individual examples in the literature on the subject which propose that cultural factors in Germany, in particular, tend to lead to a more cautious introduction of new technologies, above all in cases where the short-term benefits are difficult to estimate (see Hofstede Insights 2023, OECD 2024). However, there are hardly any empirical studies available that quantify this correlation. Whether and to what extent these kinds of cultural characteristics really influence the speed and depth of AI diffusion is therefore an open question

The regulatory environment is another factor that influences adoption momentum, as addressed above. Unclear or contradictory requirements, for example in the AI Act or in conjunction with European data protection requirements, create legal uncertainty and curb investments, particularly among SMEs. Without practicable, European-wide uniform regulations, Germany is in danger of losing appeal as a business location in international competition.

Systemic barriers to implementation – research transfer and data access

While many locational factors refer to the prerequisites for the development and introduction of AI, their actual economic impact is often determined by the last mile, which means the successful translation of research into operational application and the access to high-quality data. This is exactly where Germany has key weaknesses, but also untapped potential.

A major barrier to implementation is the limited transfer of scientific findings into commercial application. Despite Germany’s internationally strong research base, the transfer to economically relevant products and processes has so far only been limited, particularly among midsized industrial enterprises. Although there are numerous programmes to support the corporate use of AI, they are often not very well known, difficult to access or insufficiently tailored to the needs of smaller enterprises (OECD 2024).

Structural obstacles further exacerbate the corporate application of AI, such as sluggish digitalisation, limited access to high-quality data and uncertainties in the compliance with data protection regulations and other regulatory requirements. Small and medium-sized enterprises (SMEs) are particularly affected by this. According to Eurostat (2023), Germany only ranks seventh within the EU regarding the degree of digitalisation among SMEs. Many companies do not have either structured, interoperable data formats or the necessary competence to integrate different data sources (OECD 2024c). Open public data is also often not used, even though it would be highly relevant in the training and validation of AI models.

This also represents an opportunity for Germany. The country has a large number of industrial databases that could form an important basis for applied AI solutions given the right legal, technical and organisational parameters.

Conclusion

The studies analysed in the previous sections provide valuable insights into the potential impact on productivity of AI in Germany but tend to ignore key influencing factors such as the corporate landscape in terms of size, the regulatory environment and the limited commercial application of scientific findings Although Germany occupies a good international position in terms of AI research, it has significant weaknesses in other areas, particularly in digital infrastructure, the availability of skilled staff and commercial implementation. The findings of the studies and our estimate of potential growth should therefore be treated with considerable caution. It cannot be ruled out that, in some aspects, the position of Germany has been too optimistically assessed. The identified shortcomings need to be tackled resolutely for the opportunities presented by AI to be realised and to prevent Germany from falling further behind in international competition.

4. Action Recommendations for Policymakers

AI opens up considerable potential for Germany to increase its growth, productivity and competitiveness. To tap into this potential, decisive factors will be the implementation in practice and wide availability of technological developments. Given the dominance of midsized enterprises in the country’s corporate landscape, the importance of the value networks of small, medium and large companies, and the sectoral and regional complexity of industrial value added, policymakers have a key role to play. They must establish suitable framework conditions for innovations to be able to unfold their economic potential.

The BDI/BCG (2025) study on deep tech, which investigates not only AI but also robotics, quantum technologies, mRNA drugs and gene and cell therapies, sets out six overarching priorities that serve as the basis for the following recommendations.

1. Define targets, plan backwards, implement resolutely: Define targets for each technology in the framework of the High-tech Agenda and plan backwards with binding steps and deadlines

2. Focus innovative power in selected technology hubs: Merge fragmented innovation landscape with over 120 innovation clusters into a smaller number of powerful technology hubs

3. Pool rather than scatter financial support for more punch: Target support instruments to increase the volume of funding available to scale up European deep tech champions

4. Establish industry as the engine of growth for deep tech start-ups: Systematically network deep-tech start-ups with industry to combine the power to innovate with resources and scalability

5. Commercialise scientific findings: Strengthen the technological transfer and commercialisation of Germany’s large research institutes to increase economic value added.

6. Secure key positions in value added using industrial strength: Identify and take on key roles in value chains at an early stage, in addition to selected full stack approaches.

While the recommendations above refer to deep tech in general, they can be used to derive specific strategic priorities for the field of artificial intelligence. The BDI/BCG study (2025) points out that the current momentum in AI development emanates from two parallel races in international competition, and Germany is competing in this competition. The first race is the development, scaling up and provision of basic AI technologies and models Germany is far behind the hyper-scalers and AI pioneers from the United States and China in this race. The second race is the development and implementation of products, processes and business models based on existing advanced AI technologies In this application-oriented competition, Germany is playing up front on the international arena and is even pioneer in individual segments.

The necessary strategic course is therefore clearly as follows: In the first case, Germany should focus its efforts on strategically relevant areas such as industrial AI, through the development of large industry models, for example, with specific industrial applications. The larger economic potential, however, lies in the second race, the race to application. Here, Germany has the opportunity to build on its current strengths and develop new skills to secure its leading position for the long term.

The High-Tech Agenda Germany adopted in July 2025 represents the first important step in this direction. It pursues the objective of generating ten percent of Germany’s economic output through AI technologies by 2030 and establish Germany as a leading AI location on the international stage. Further targeted measures are needed if these targets are to be met. The following three recommendations for action from the BDI/BCG study are particularly relevant to anchor AI across broad swathes of the German economy and particularly within industry.

1. Restructure the economy resolutely with a focus on AI: Companies of all sizes should optimise their products, services and processes not just through the isolated use of AI but restructure them with an AI focus across the board, with the objective of creating new business models, organisational methods and value chains as well as significantly increasing efficiency. Midsized industrial companies also need to develop their own future scenarios (see BDI/Z_punkt, 2024)

2. Strengthen AI competencies and acceptance through national educational initiatives: AI skills need to become more widespread in the economy and in society with practical programmes at schools and in further education institutions. In addition, simple and persuasive targets (for example, up to 2030) should be communicated in order to provide orientation and reduce social anxieties.

3. Development of sovereign AI infrastructure: Germany and the EU should reduce strategic dependencies and strengthen their technological independence for the long term by investing in computing centres and supercomputers, AI gigafactories, energy infrastructure, opensource solutions and the European production of critical components.

Bibliography

Acemoglu, D. (2024). The Simple Macroeconomics of AI. April. Massachusetts Institute of Technology. Cambridge, Massachusetts.

BDI (2024). Transformation Paths for Germany as an Industrial Nation. September. Berlin.

BDI and Z_punkt, 2024 The Future of Industrial Mittelstand in Germany in 2030 November. Berlin.

Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative AI at Work. April. National Bureau of Economic Research. Cambridge, Massachusetts.

BCG & BDI (2025). Deep Tech in Deutschland. Strategische Weichenstellungen für ein leistungsstarkes Ökosystem. September. Berlin.

BCG (2025). Zukunftstechnologie KI: 2025 trifft weltweite Dynamik auf deutsche Zurückhaltung. January Cologne

Czarnitzki, D., Fernández, G. and Rammer, C. (2023). Artificial intelligence and firm-level productivity. Journal of Economic Behavior & Organization, Vol. 211, pp. 188-205.

Dell’Acqua, F., McFowland III, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. September. Harvard Business School Technology & Operations Management Unit. Boston, Massachusetts.

Destatis (2019). 20.8 trillion euro capital stock on a gross basis at the end of 2019 Wiesbaden.

Deutscher Bundestag (2025). Finanzplan des Bundes 2025 bis 2029. September. Berlin.

Destatis (2022). Small and medium-sized enterprises Wiesbaden

Eurostat (2025). Wages and labour costs. Data extracted on 28 March 2025. Luxemburg.

---(2023). Digitalisation in Europe – 2023 edition. Luxemburg.

EZB (2025). AI can boost productivity – if firms use it. ECB Blog. March. Frankfurt am Main.

---(2024). The Past, Present and Future of European Productivity. July. Frankfurt am Main.

Financial Times (2025). Inside the relentless race for AI capacity. July. London.

Fraunhofer Institute for Industrial Mathematics (Fraunhofer ITWM). (Not dated). Automatisierung von Prüfaufgaben mit KI und Maschinellem Lernen. Accessed on 18 August 2025. Kaiserslautern.

German Council of Economic Experts (2025). Spring Report 2025. May. Wiesbaden.

German Federal Ministry of Finance & German Federal Ministry of Economic Affairs and Energy (2025). German Medium-Term Fiscal-Structural Plan from 2025 to 2029. July. Berlin.

German Federal Ministry of Economic Affairs and Energy (2024). Künstliche Intelligenz: Für mehr Produktivität braucht es die richtigen Rahmenbedingungen. May. Berlin.

Goldman Sachs (2025). AI to drive 165% increase in data center power demand by 2030. February. New York City.

(2023). Generative AI could raise global GDP by 7%. April. New York City.

Hofstede Insights (2023). Country comparison: Germany. Hofstede Insights.

IW Consult (2025). Wie wird KI die Produktivität in Deutschland verändern? Gutachten. Commission by: Gemeinschaftsausschuss der Deutschen Gewerblichen Wirtschaft (joint committee of German business associations) February Cologne

IWF (2025). Artificial Intelligence and Productivity in Europe. IMF Working Paper WP/25/67. April. Washington DC.

Ju, H., & Aral, S. (2025). Collaborating with AI Agents: Field Experiments on Teamwork, Productivity, and Performance. Massachusetts Institute of Technology. March. Cambridge, Massachusetts.

Kodumuru R., Sarkar S., Parepally V., Chandarana J. (2025). Artificial Intelligence and Internet of Things Integration in Pharmaceutical Manufacturing: A Smart Synergy. Pharmaceutics. January Basel.

McKinsey & Company (2025). The cost of compute: A $7 trillion race to scale data centers. April. New York City.

---(2023). The economic potential of generative AI. The next productivity frontier. June. New York City.

MIT NANDA (2025). The GenAI Divide: State of AI in Business 2025. July. Cambridge, Massachusetts.

OECD (2025). Macroeconomic productivity gains from Artificial Intelligence in G7 economies. May Paris.

---(2025b). Compendium of Productivity Indicators 2025. Paris.

---(2025c). Global long-run economic scenarios 2025 update. Paris

(2024). The Impact of Artificial Intelligence on Productivity, Distribution and Growth. Key Mechanisms, Initial Evidence and Policy Challenges. OECD Artificial Intelligence Papers. April. Paris. (2024b). Miracle or Myth? Assessing the macroeconomic productivity gains from Artificial Intelligence. November. Paris.

(2024c). OECD Artificial Intelligence Review of Germany. June. Paris.

---(2023). A portrait of AI adopters across countries: Firm characteristics, assets’ complementarities and productivity. February. Paris.

Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023). The Impact of AI on Developer Productivity: Evidence from GitHub Copilot. February. Cambridge, Massachusetts.

Robert Bosch GmbH (2025). KI, die bewegt. Bosch erleichtert mit Algorithmen den Alltag. CEO Blog. June. Gerlingen.

Schaeffler AG (2025). Digitale Fertigung: Schaeffler und NVIDIA schließen Technologiekooperation. June Herzogenaurach.

Siemens AG (2025). Siemens beschleunigt mit Innovationen und Partnerschaften den Weg zur KIgesteuerten Industrie. March Munich

Stiftung Marktwirtschaft (2025). Künstliche Intelligenz – Wie lassen sich Wachstumspotentiale freisetzen? August. Berlin.

Table Media (2025). AI chips from the USA: How high demand in Europe is Europe Table #1003. 12 August. Berlin.

vfa & BCG (2024). KI in der Arzneimittelentwicklung. June Berlin.

Publishing Information

Federation of German Industries (BDI)

Breite Strasse 29, 10178 Berlin www.bdi.eu

T: +49 30 2028-0

Lobby register number: R000534

EU Transparency Register: 1771817758-4

Author

Frederik Lange

T: +49 30 2028 1734 f.lange@bdi.eu

Editors / Graphics

Dr Klaus Günter Deutsch T: +49 30 2028 1591 k.Deutsch@bdi.eu

Dr Thomas Koenen

T: +49 30 2028 1415 t.koenen@bdi.eu

Polina Khubbeeva T: +49 30 2028 1586 p.khubbeeva@bdi.eu

BDI document number: D 2162

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.
Artificial Intelligence as an Opportunity for Growth by Bundesverband der Deutschen Industrie e.V. - Issuu