German project escate achieves up to 89% energy savings in ai

Professor Wolfgang Maas of the German Research Center for Artificial Intelligence (DFKI) detailed the ESCATE project, which concluded with the goal of developing the world's most efficient AI data center end-to-end. The project utilized neuromorphic computing hardware from TU Dresden, specifically the SpinEca2 chip architecture, which activates only necessary hardware parts upon request, unlike traditional GPUs from NVIDIA and AMD that typically run at 100% load. The research integrated spiking neuron algorithms, inspired by biological brains, to reduce energy consumption. A key application involved partnering with Saarstahl to use drones for sorting steel scrap and measuring quality, replacing manual visual inspection. The algorithms were tested in a large data center in Munich operated by Telekom, where throttling clock rates during inference phases yielded significant energy savings. Maas cited peak values of up to 89% energy reduction under artificial conditions, noting that while the human brain operates at 20 watts, comparable to a 100-megawatt data center, the technology is being transferred to the successor project Halo X for use in small drones and Bundeswehr initiatives. Maas argued that Europe must prioritize energy efficiency and renewable energy sources over the US model of maximizing size and speed, citing water and power constraints in densely populated regions.