On 30 July 2026, the Frankfurter Allgemeine Zeitung reported on the results of the ESCADE project under the headline "DFKI: Energiehunger der KI um 90 Prozent senken" ("DFKI: Cutting AI's energy demand by 90 percent"). In his article, business editor Bernd Freytag summarizes the approaches that the consortium coordinated by DFKI and led by Prof. Dr.-Ing. Wolfgang Maaß developed over three years to reduce the electricity demand of artificial intelligence.

The article covers the two software approaches at the core of the project: compressing large teacher models into tailored student models that are up to 90 percent smaller and, according to the project's test runs, require up to 89 percent less energy; and neural architecture search for image-processing models, which lowered energy consumption by 40 percent while improving accuracy. It also describes the field test at SHS – Stahl-Holding-Saar, where a compressed visual model classifies scrap steel from camera images, and the work on neuromorphic chips, which ran up to six times more efficiently than conventional processors in the project's tests. "This makes high-performance AI models accessible to small and medium-sized enterprises as well, which was previously impossible simply because of the size of the models," Dr. Sabine Janzen is quoted as saying.

What makes the article valuable beyond the results themselves is the context in which it places them. Freytag points to the scale of the global data center buildout, which he describes as the largest industrial infrastructure investment in a long time: the four large US technology companies alone announced investments of 650 billion dollars at the beginning of the year, with new plans following almost weekly. He illustrates the local conflicts this creates with the example of a gas-fired power plant planned by the municipal utilities in Mainz to supply a single data center, and notes the doubts — including within DFKI — as to whether Europe can win a race decided by sheer scale. Against this backdrop, he frames the ESCADE results as pointing toward a different route: AI that is more targeted, less expensive and considerably less energy-hungry.

The full article is available on FAZ.NET.