Does AI reduce energy consumption? Evidence from twelve years of German industry data
A new study using ISTARI data reveals the role of firm size and regional conditions, and how the effect has changed over time.
.jpg&w=3840&q=75)
Charlotte Maw
Artificial Intelligence is increasingly viewed as both a promising tool for the optimization of industrial processes, and as a strain on the energy grid and other natural resources. A group of researchers from the University of Mannheim, Friedrich Schiller University Jena, Paris Lodron University Salzburg, IT:U Austria, University of Bremen and Thünen Institute of Rural Economics have now examined what the rising adoption levels of this technology among firms actually mean for their industries’ regional energy usage. Covering 363 NUTS-3 regions in Germany, or roughly 91 percent of the German districts, over the period of 2012 to 2023, the study combines official energy statistics and patent data with ISTARI web indicators on firm-level AI adoption and sustainability engagement. Their article “Thinking machines, shrinking footprints? The impact of Artificial Intelligence on energy consumption in regions” was published in Energy Economics in July 2026.
“AI adoption can reduce energy consumption. But only if adoption becomes broad and not concentrated only on a few frontrunners. As AI continues to spread, we need to actively shape diffusion, especially in industries and regions that have been slower to adopt.”
Dr. Robert Dehghan | Co-Author and COO of ISTARI
Measuring firm-level AI adoption with webAI
Patent filings track technological inventions. They offer no insight into the subsequent dispersion of these technologies. In order to understand the firm-level adoption of Artificial Intelligence the researchers rely on the ISTARI webAI approach. For this study, the website texts from 2.47 million German companies were retrieved for each year between 2012 and 2023 and classified according to ISTARI’s AI Intensity scoring. This score reflects how central Artificial Intelligence is to a firm’s business model. It is based on an award-winning scientific methodology. More information on this, and other webAI indicators, can be found here.
Firm-level AI adoption is linked to lower industrial energy consumption
When mapped over time, the first noticeable AI adoption among German firms appears in 2018 in Berlin, Munich and the Rhine-Neckar region. In the following years, the technology gradually diffuses across the German economic landscape, with these urban hotspots of AI activity still clearly visible.
The study shows regions with a higher AI adoption among firms to have significantly lower industrial energy consumption – a 0.1 percentage point increase in regional AI adoption being associated with a 0.183 percent average decrease in energy usage across the entire study period. This effect, however, is not stable over time: The year 2018 is generally regarded as a crucial inflection point in the diffusion of Artificial Intelligence and its commercial application. And indeed, when researchers compare the pre-2018 to the post-2018 observations, the energy savings drop from 0.974 percent to 0.194 percent.
The authors offer several possible explanations. Early adoption may have targeted the most obvious inefficiencies first, AI applications over time may have shifted to more complex, compute-heavy business objectives and machine learning models. And the technology may have gradually diffused into less energy-intensive sectors with overall lower saving potential.
Policymakers urged to prioritize SME adoption and regional capabilities
The paper goes on to form two clear policy recommendations for reaping the potential efficiency gains on industrial energy usage.
Firstly, the researchers suggest a support program, targeted specifically at SMEs. Adoption of Artificial Intelligence among small firms is associated with a larger percentage reduction in industry-wide energy consumption, yet adoption rates among smaller organizations remain low. Building out regional expertise matters as well. Using patent data, the study measures how close each region’s existing technological strengths are to Artificial Intelligence. In regions with little relevant industry expertise, AI adoption shows no measurable effect on energy consumption. In regions with more, firms appear to be able to generate significant savings out of the same adoption levels.
The researchers are open about the limitations of their work, particularly around a currently much contested topic: data centers. While the study controls for the number of public data centers per region, the authors call for future studies to consider also the energy consumption of these centers themselves, and to account for computational capabilities drawn from data centers outside the region or country in question. Because while energy efficiency gains through the adoption of Artificial Intelligence show up in industries locally, the cost of computation may not.
Reference: Dehghan, R., Grashof, N., Schmidt, S., Kopka, A., & Woywode, M. (2026). Thinking machines, shrinking footprints? The impact of Artificial Intelligence on energy consumption in regions. Energy Economics, 161, 109526. Link to the study
