In late March 2026, researchers from three European universities organized a "Green AI Hackathon" focused on building artificial intelligence systems that actively reduce their carbon footprint. Nine students from Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) traveled to Budapest University of Technology and Economics (BME) to collaborate with peers from Italy and Hungary on developing prototypes for more sustainable AI models.

Why Should We Care About AI's Energy Consumption?

The energy demands of artificial intelligence are staggering. According to the International Energy Agency (IEA), AI is estimated to consume between 60 to 70 terawatt-hours (TWh) of electricity annually, roughly equivalent to the total energy consumption of Switzerland, Austria, or Finland combined. This calculation accounts only for data center operations, not the energy required to manufacture the chips and materials that power AI systems. When that energy comes from fossil fuels, the carbon footprint becomes a significant environmental concern.

"According to the IEA, Artificial Intelligence is estimated to consume as much energy as 60-70 TWh per year, roughly as much as Switzerland, Austria or Finland. And this is only estimated from data centers, not taking into account the energy used to produce the precursor materials and chips themselves," explained Prof. Dr. Andreas Kist.

Prof. Dr. Andreas Kist, Department of Artificial Intelligence in Biomedical Engineering at FAU

The hackathon, funded through the EELISA (European Engineering Learning Innovation and Science Alliance) Joint Call for Communities, brought together students from diverse technical backgrounds to brainstorm and prototype solutions to this growing problem.

What Practical Solutions Did Teams Develop?

Over three intensive days, three interdisciplinary teams worked on specific use cases aimed at reducing AI's carbon footprint. The results ranged from early-stage concepts to nearly functional prototypes that demonstrate real-world applications:

• Wildfire Detection on Edge Devices: A team called "tinyfire" developed a computer vision algorithm designed to run locally on battery-powered edge devices, potentially powered by solar panels. The system would be attached to camera drones to detect wildfires at their earliest stages, enabling faster intervention while avoiding the need to send data to energy-intensive data centers for processing.
• Carbon-Aware Batch Processing: The "SustainaBatch" team created code that schedules AI workloads to run only when sufficient green energy is available on the electrical grid, automatically pausing computations during periods when fossil fuels dominate the energy mix, such as nighttime or cloudy days.
• Waste Heat Recovery: Researchers discussed mechanisms for capturing and repurposing the waste heat generated by servers running AI systems to heat nearby buildings, turning an environmental liability into a practical resource.

The SustainaBatch team advanced their work significantly during the hackathon and released their code as open-source software, making it available for other developers to test and build upon.

How to Approach Green AI Development in Your Organization

• Optimize at Multiple Levels: Green AI improvements can happen through hardware optimization, such as using more efficient processors, or through software optimization, such as writing more efficient algorithms and scheduling computations strategically.
• Monitor Energy Sources in Real Time: Implement systems that track the carbon intensity of your electrical grid and adjust computational workloads accordingly, prioritizing energy-intensive tasks when renewable energy is abundant.
• Leverage Edge Computing: Move AI processing closer to data sources by running models on local devices rather than sending all data to centralized data centers, reducing both energy consumption and latency.
• Capture and Reuse Waste Heat: Design data center infrastructure to capture thermal energy from computing equipment and redirect it for practical uses like building heating or industrial processes.

"Making an AI green means optimizing the hardware or the software. For example, using the waste heat of the servers that are running the AI to heat buildings nearby," noted Prof. Dr. Ágnes Urbin.

Prof. Dr. Ágnes Urbin, Department of Mechatronics, Optics, and Mechanical Engineering Informatics at BME

What Made This Hackathon Unique?

The event brought together researchers and students from FAU in Germany, BME in Budapest, and Scuola Superiore Sant'Anna (SSSA) in Pisa, Italy, creating an international collaboration that would have been difficult to organize outside the European university framework. The organizers emphasized that the ability to collaborate across borders without visa requirements and bureaucratic obstacles was itself a significant advantage.

Prof. Dr. Calogero Maria Oddo from The BioRobotics Institute at SSSA in Pisa noted that while three days is insufficient to develop fully functional, complex prototypes, the teams generated genuinely innovative ideas and demonstrated strong motivation to continue the work. The hackathon succeeded in raising awareness about Green AI among technical students from diverse backgrounds and exposing them to practical mechanisms for reducing carbon footprints in artificial intelligence systems.

The EELISA alliance continues to support collaborative research and education initiatives. The call for proposals for "Joint Academic Programs" remains open until June 20, 2026, while the "Research Retreats" program accepts applications year-round. The next call for "Joint Call for Communities" will open in September 2026.