The flagship project NADIKI, funded by the BMUKN, enables the real-time measurement of actual energy consumption and CO₂ emissions of AI applications for the first time. Instead of estimates, it provides transparent metrics from server hardware to building infrastructure. The project establishes the technical basis for fact-based cost-benefit decisions in AI deployment and evaluates an environmental label similar to the Blue Angel for software.
The NADIKI project makes it possible for the first time to measure the actual energy consumption and CO₂ emissions of AI applications in real time. Funded with approximately 880,000 euros by the Federal Ministry for the Environment, Nature Conservation and Nuclear Safety (BMUKN), this flagship project provides transparent metrics instead of estimates—from server hardware to the data center's building infrastructure.
“The specific energy consumption of AI clusters exceeds that of typical computing clusters by multiples, and we anticipate a massive growth of AI applications. Understanding efficiency and aligning energy expenditure with outcomes is crucial.” — Prof. Dr.-Ing. Peter Radgen, Chair of Energy Efficiency, University of Stuttgart
NADIKI is developing an interface that records the real energy and resource consumption of AI models during execution (inference) and provides it as transparent metrics. The infrastructure accurately calculates greenhouse gas emissions and resource usage based on data from server hardware, cooling systems, and the data center building.
What NADIKI enables
Transparent measurement: Captures the actual resource consumption of AI workloads instead of estimates within seconds of AI inference execution—from server hardware through cooling systems to the building infrastructure.
Temporal and locational optimization: AI training processes can be relocated to sites with a higher share of renewable energies in the local power grid.
Fact-based evaluation: Positive environmental impacts are compared to potential negative effects—a foundation for informed cost-benefit decisions in AI deployment.
Environmental labeling: Evaluation of a certification similar to the Blue Angel for software to label AI models based on their environmental impact.
Open source: All results are provided as open-source software with an open data basis.
“Our goal is to make the hidden, full costs of AI services and applications transparent—financially and ecologically. Only in this way can we make informed cost-benefit decisions for the use of AI.” — Max Schulze, SDIA
The project targets data center operators, AI service providers, developers, and the general public. It runs until November 2025 and is carried out in collaboration with the Institute for Architecture of Application Systems (IAAS) and the Institute for Energy Economics and Rational Energy Use (IER) at the University of Stuttgart.